Showing 1–50 of 242 results for author: Cheng, D

FinGAIA: An End-to-End Benchmark for Evaluating AI Agents in Finance

Authors: Lingfeng Zeng, Fangqi Lou, Zixuan Wang, Jiajie Xu, Jinyi Niu, Mengping Li, Yifan Dong, Qi Qi, Wei Zhang, Ziwei Yang, Jun Han, Ruilun Feng, Ruiqi Hu, Lejie Zhang, Zhengbo Feng, Yicheng Ren, Xin Guo, Zhaowei Liu, Dongpo Cheng, Weige Cai, Liwen Zhang

Abstract: The booming development of AI agents presents unprecedented opportunities for automating complex tasks across various domains. However, their multi-step, multi-tool collaboration capabilities in the financial sector remain underexplored. This paper introduces FinGAIA, an end-to-end benchmark designed to evaluate the practical abilities of AI agents in the financial domain. FinGAIA comprises 407 me… ▽ More The booming development of AI agents presents unprecedented opportunities for automating complex tasks across various domains. However, their multi-step, multi-tool collaboration capabilities in the financial sector remain underexplored. This paper introduces FinGAIA, an end-to-end benchmark designed to evaluate the practical abilities of AI agents in the financial domain. FinGAIA comprises 407 meticulously crafted tasks, spanning seven major financial sub-domains: securities, funds, banking, insurance, futures, trusts, and asset management. These tasks are organized into three hierarchical levels of scenario depth: basic business analysis, asset decision support, and strategic risk management. We evaluated 10 mainstream AI agents in a zero-shot setting. The best-performing agent, ChatGPT, achieved an overall accuracy of 48.9\%, which, while superior to non-professionals, still lags financial experts by over 35 percentage points. Error analysis has revealed five recurring failure patterns: Cross-modal Alignment Deficiency, Financial Terminological Bias, Operational Process Awareness Barrier, among others. These patterns point to crucial directions for future research. Our work provides the first agent benchmark closely related to the financial domain, aiming to objectively assess and promote the development of agents in this crucial field. Partial data is available at https://github.com/SUFE-AIFLM-Lab/FinGAIA. △ Less

Submitted 23 July, 2025; originally announced July 2025.

CKAA: Cross-subspace Knowledge Alignment and Aggregation for Robust Continual Learning

Authors: Lingfeng He, De Cheng, Zhiheng Ma, Huaijie Wang, Dingwen Zhang, Nannan Wang, Xinbo Gao

Abstract: Continual Learning (CL) empowers AI models to continuously learn from sequential task streams. Recently, parameter-efficient fine-tuning (PEFT)-based CL methods have garnered increasing attention due to their superior performance. They typically allocate a unique sub-module for learning each task, with a task recognizer to select the appropriate sub-modules for testing images. However, due to the… ▽ More Continual Learning (CL) empowers AI models to continuously learn from sequential task streams. Recently, parameter-efficient fine-tuning (PEFT)-based CL methods have garnered increasing attention due to their superior performance. They typically allocate a unique sub-module for learning each task, with a task recognizer to select the appropriate sub-modules for testing images. However, due to the feature subspace misalignment from independently trained sub-modules, these methods tend to produce ambiguous decisions under misleading task-ids. To address this, we propose Cross-subspace Knowledge Alignment and Aggregation (CKAA), a novel framework that enhances model robustness against misleading task-ids through two key innovations: (1) Dual-level Knowledge Alignment (DKA): By aligning intra-class feature distributions across different subspaces and learning a robust global classifier through a feature simulation process, DKA enables the model to distinguish features from both correct and incorrect subspaces during training. (2) Task-Confidence-guided Mixture of Adapters (TC-MoA): A robust inference scheme that adaptively aggregates task-specific knowledge from relevant sub-modules based on task-confidence scores, avoiding overconfidence in misleading task-id predictions. Extensive experiments demonstrate that CKAA outperforms existing PEFT-based CL methods. △ Less

Submitted 12 July, 2025; originally announced July 2025.

Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities

Authors: Gheorghe Comanici, Eric Bieber, Mike Schaekermann, Ice Pasupat, Noveen Sachdeva, Inderjit Dhillon, Marcel Blistein, Ori Ram, Dan Zhang, Evan Rosen, Luke Marris, Sam Petulla, Colin Gaffney, Asaf Aharoni, Nathan Lintz, Tiago Cardal Pais, Henrik Jacobsson, Idan Szpektor, Nan-Jiang Jiang, Krishna Haridasan, Ahmed Omran, Nikunj Saunshi, Dara Bahri, Gaurav Mishra, Eric Chu , et al. (3284 additional authors not shown)

Abstract: In this report, we introduce the Gemini 2.X model family: Gemini 2.5 Pro and Gemini 2.5 Flash, as well as our earlier Gemini 2.0 Flash and Flash-Lite models. Gemini 2.5 Pro is our most capable model yet, achieving SoTA performance on frontier coding and reasoning benchmarks. In addition to its incredible coding and reasoning skills, Gemini 2.5 Pro is a thinking model that excels at multimodal unde… ▽ More In this report, we introduce the Gemini 2.X model family: Gemini 2.5 Pro and Gemini 2.5 Flash, as well as our earlier Gemini 2.0 Flash and Flash-Lite models. Gemini 2.5 Pro is our most capable model yet, achieving SoTA performance on frontier coding and reasoning benchmarks. In addition to its incredible coding and reasoning skills, Gemini 2.5 Pro is a thinking model that excels at multimodal understanding and it is now able to process up to 3 hours of video content. Its unique combination of long context, multimodal and reasoning capabilities can be combined to unlock new agentic workflows. Gemini 2.5 Flash provides excellent reasoning abilities at a fraction of the compute and latency requirements and Gemini 2.0 Flash and Flash-Lite provide high performance at low latency and cost. Taken together, the Gemini 2.X model generation spans the full Pareto frontier of model capability vs cost, allowing users to explore the boundaries of what is possible with complex agentic problem solving. △ Less

Submitted 22 July, 2025; v1 submitted 7 July, 2025; originally announced July 2025.

Comments: 72 pages, 17 figures

Deconfounding Causal Inference through Two-Branch Framework with Early-Forking for Sensor-Based Cross-Domain Activity Recognition

Authors: Di Xiong, Lei Zhang, Shuoyuan Wang, Dongzhou Cheng, Wenbo Huang

Abstract: Recently, domain generalization (DG) has emerged as a promising solution to mitigate distribution-shift issue in sensor-based human activity recognition (HAR) scenario. However, most existing DG-based works have merely focused on modeling statistical dependence between sensor data and activity labels, neglecting the importance of intrinsic casual mechanism. Intuitively, every sensor input can be v… ▽ More Recently, domain generalization (DG) has emerged as a promising solution to mitigate distribution-shift issue in sensor-based human activity recognition (HAR) scenario. However, most existing DG-based works have merely focused on modeling statistical dependence between sensor data and activity labels, neglecting the importance of intrinsic casual mechanism. Intuitively, every sensor input can be viewed as a mixture of causal (category-aware) and non-causal factors (domain-specific), where only the former affects activity classification judgment. In this paper, by casting such DG-based HAR as a casual inference problem, we propose a causality-inspired representation learning algorithm for cross-domain activity recognition. To this end, an early-forking two-branch framework is designed, where two separate branches are respectively responsible for learning casual and non-causal features, while an independence-based Hilbert-Schmidt Information Criterion is employed to implicitly disentangling them. Additionally, an inhomogeneous domain sampling strategy is designed to enhance disentanglement, while a category-aware domain perturbation layer is performed to prevent representation collapse. Extensive experiments on several public HAR benchmarks demonstrate that our causality-inspired approach significantly outperforms eleven related state-of-the-art baselines under cross-person, cross-dataset, and cross-position settings. Detailed ablation and visualizations analyses reveal underlying casual mechanism, indicating its effectiveness, efficiency, and universality in cross-domain activity recognition scenario. △ Less

Submitted 5 July, 2025; originally announced July 2025.

Comments: Accepted by Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies (IMWUT)

Journal ref: Proc. ACM Interact. Mob. Wearable Ubiquitous Technol. 9, 2, Article 56 (June 2025)

Prompt Disentanglement via Language Guidance and Representation Alignment for Domain Generalization

Authors: De Cheng, Zhipeng Xu, Xinyang Jiang, Dongsheng Li, Nannan Wang, Xinbo Gao

Abstract: Domain Generalization (DG) seeks to develop a versatile model capable of performing effectively on unseen target domains. Notably, recent advances in pre-trained Visual Foundation Models (VFMs), such as CLIP, have demonstrated considerable potential in enhancing the generalization capabilities of deep learning models. Despite the increasing attention toward VFM-based domain prompt tuning within DG… ▽ More Domain Generalization (DG) seeks to develop a versatile model capable of performing effectively on unseen target domains. Notably, recent advances in pre-trained Visual Foundation Models (VFMs), such as CLIP, have demonstrated considerable potential in enhancing the generalization capabilities of deep learning models. Despite the increasing attention toward VFM-based domain prompt tuning within DG, the effective design of prompts capable of disentangling invariant features across diverse domains remains a critical challenge. In this paper, we propose addressing this challenge by leveraging the controllable and flexible language prompt of the VFM. Noting that the text modality of VFMs is naturally easier to disentangle, we introduce a novel framework for text feature-guided visual prompt tuning. This framework first automatically disentangles the text prompt using a large language model (LLM) and then learns domain-invariant visual representation guided by the disentangled text feature. However, relying solely on language to guide visual feature disentanglement has limitations, as visual features can sometimes be too complex or nuanced to be fully captured by descriptive text. To address this, we introduce Worst Explicit Representation Alignment (WERA), which extends text-guided visual prompts by incorporating an additional set of abstract prompts. These prompts enhance source domain diversity through stylized image augmentations, while alignment constraints ensure that visual representations remain consistent across both the original and augmented distributions. Experiments conducted on major DG datasets, including PACS, VLCS, OfficeHome, DomainNet, and TerraInc, demonstrate that our proposed method outperforms state-of-the-art DG methods. △ Less

Submitted 2 July, 2025; originally announced July 2025.

A computationally frugal open-source foundation model for thoracic disease detection in lung cancer screening programs

Authors: Niccolò McConnell, Pardeep Vasudev, Daisuke Yamada, Daryl Cheng, Mehran Azimbagirad, John McCabe, Shahab Aslani, Ahmed H. Shahin, Yukun Zhou, The SUMMIT Consortium, Andre Altmann, Yipeng Hu, Paul Taylor, Sam M. Janes, Daniel C. Alexander, Joseph Jacob

Abstract: Low-dose computed tomography (LDCT) imaging employed in lung cancer screening (LCS) programs is increasing in uptake worldwide. LCS programs herald a generational opportunity to simultaneously detect cancer and non-cancer-related early-stage lung disease. Yet these efforts are hampered by a shortage of radiologists to interpret scans at scale. Here, we present TANGERINE, a computationally frugal,… ▽ More Low-dose computed tomography (LDCT) imaging employed in lung cancer screening (LCS) programs is increasing in uptake worldwide. LCS programs herald a generational opportunity to simultaneously detect cancer and non-cancer-related early-stage lung disease. Yet these efforts are hampered by a shortage of radiologists to interpret scans at scale. Here, we present TANGERINE, a computationally frugal, open-source vision foundation model for volumetric LDCT analysis. Designed for broad accessibility and rapid adaptation, TANGERINE can be fine-tuned off the shelf for a wide range of disease-specific tasks with limited computational resources and training data. Relative to models trained from scratch, TANGERINE demonstrates fast convergence during fine-tuning, thereby requiring significantly fewer GPU hours, and displays strong label efficiency, achieving comparable or superior performance with a fraction of fine-tuning data. Pretrained using self-supervised learning on over 98,000 thoracic LDCTs, including the UK's largest LCS initiative to date and 27 public datasets, TANGERINE achieves state-of-the-art performance across 14 disease classification tasks, including lung cancer and multiple respiratory diseases, while generalising robustly across diverse clinical centres. By extending a masked autoencoder framework to 3D imaging, TANGERINE offers a scalable solution for LDCT analysis, departing from recent closed, resource-intensive models by combining architectural simplicity, public availability, and modest computational requirements. Its accessible, open-source lightweight design lays the foundation for rapid integration into next-generation medical imaging tools that could transform LCS initiatives, allowing them to pivot from a singular focus on lung cancer detection to comprehensive respiratory disease management in high-risk populations. △ Less

Submitted 15 July, 2025; v1 submitted 2 July, 2025; originally announced July 2025.

MPipeMoE: Memory Efficient MoE for Pre-trained Models with Adaptive Pipeline Parallelism

Authors: Zheng Zhang, Donglin Yang, Yaqi Xia, Liang Ding, Dacheng Tao, Xiaobo Zhou, Dazhao Cheng

Abstract: Recently, Mixture-of-Experts (MoE) has become one of the most popular techniques to scale pre-trained models to extraordinarily large sizes. Dynamic activation of experts allows for conditional computation, increasing the number of parameters of neural networks, which is critical for absorbing the vast amounts of knowledge available in many deep learning areas. However, despite the existing system… ▽ More Recently, Mixture-of-Experts (MoE) has become one of the most popular techniques to scale pre-trained models to extraordinarily large sizes. Dynamic activation of experts allows for conditional computation, increasing the number of parameters of neural networks, which is critical for absorbing the vast amounts of knowledge available in many deep learning areas. However, despite the existing system and algorithm optimizations, there are significant challenges to be tackled when it comes to the inefficiencies of communication and memory consumption. In this paper, we present the design and implementation of MPipeMoE, a high-performance library that accelerates MoE training with adaptive and memory-efficient pipeline parallelism. Inspired by that the MoE training procedure can be divided into multiple independent sub-stages, we design adaptive pipeline parallelism with an online algorithm to configure the granularity of the pipelining. Further, we analyze the memory footprint breakdown of MoE training and identify that activations and temporary buffers are the primary contributors to the overall memory footprint. Toward memory efficiency, we propose memory reusing strategies to reduce memory requirements by eliminating memory redundancies, and develop an adaptive selection component to determine the optimal strategy that considers both hardware capacities and model characteristics at runtime. We implement MPipeMoE upon PyTorch and evaluate it with common MoE models in a physical cluster consisting of 8 NVIDIA DGX A100 servers. Compared with the state-of-art approach, MPipeMoE achieves up to 2.8x speedup and reduces memory footprint by up to 47% in training large models. △ Less

Submitted 27 June, 2025; originally announced June 2025.

Comments: 11 pages, accepted at IPDPS 2023

Journal ref: 2023 IEEE International Parallel and Distributed Processing Symposium (IPDPS), pp. 167-177. IEEE, 2023

MCFuser: High-Performance and Rapid Fusion of Memory-Bound Compute-Intensive Operators

Authors: Zheng Zhang, Donglin Yang, Xiaobo Zhou, Dazhao Cheng

Abstract: Operator fusion, a key technique to improve data locality and alleviate GPU memory bandwidth pressure, often fails to extend to the fusion of multiple compute-intensive operators due to saturated computation throughput. However, the dynamicity of tensor dimension sizes could potentially lead to these operators becoming memory-bound, necessitating the generation of fused kernels, a task hindered by… ▽ More Operator fusion, a key technique to improve data locality and alleviate GPU memory bandwidth pressure, often fails to extend to the fusion of multiple compute-intensive operators due to saturated computation throughput. However, the dynamicity of tensor dimension sizes could potentially lead to these operators becoming memory-bound, necessitating the generation of fused kernels, a task hindered by limited search spaces for fusion strategies, redundant memory access, and prolonged tuning time, leading to sub-optimal performance and inefficient deployment. We introduce MCFuser, a pioneering framework designed to overcome these obstacles by generating high-performance fused kernels for what we define as memory-bound compute-intensive (MBCI) operator chains. Leveraging high-level tiling expressions to delineate a comprehensive search space, coupled with Directed Acyclic Graph (DAG) analysis to eliminate redundant memory accesses, MCFuser streamlines kernel optimization. By implementing guidelines to prune the search space and incorporating an analytical performance model with a heuristic search, MCFuser not only significantly accelerates the tuning process but also demonstrates superior performance. Benchmarked against leading compilers like Ansor on NVIDIA A100 and RTX3080 GPUs, MCFuser achieves up to a 5.9x speedup in kernel performance and outpaces other baselines while reducing tuning time by over 70-fold, showcasing its agility. △ Less

Submitted 27 June, 2025; originally announced June 2025.

Comments: 12 pages, accepted at SC 2024

Journal ref: SC24: International Conference for High Performance Computing, Networking, Storage and Analysis. IEEE, 2024

Reasoning with Exploration: An Entropy Perspective

Authors: Daixuan Cheng, Shaohan Huang, Xuekai Zhu, Bo Dai, Wayne Xin Zhao, Zhenliang Zhang, Furu Wei

Abstract: Balancing exploration and exploitation is a central goal in reinforcement learning (RL). Despite recent advances in enhancing language model (LM) reasoning, most methods lean toward exploitation, and increasingly encounter performance plateaus. In this work, we revisit entropy -- a signal of exploration in RL -- and examine its relationship to exploratory reasoning in LMs. Through empirical analys… ▽ More Balancing exploration and exploitation is a central goal in reinforcement learning (RL). Despite recent advances in enhancing language model (LM) reasoning, most methods lean toward exploitation, and increasingly encounter performance plateaus. In this work, we revisit entropy -- a signal of exploration in RL -- and examine its relationship to exploratory reasoning in LMs. Through empirical analysis, we uncover strong positive correlations between high-entropy regions and three types of exploratory reasoning actions: (1) pivotal tokens that determine or connect logical steps, (2) reflective actions such as self-verification and correction, and (3) rare behaviors under-explored by the base LMs. Motivated by this, we introduce a minimal modification to standard RL with only one line of code: augmenting the advantage function with an entropy-based term. Unlike traditional maximum-entropy methods which encourage exploration by promoting uncertainty, we encourage exploration by promoting longer and deeper reasoning chains. Notably, our method achieves significant gains on the Pass@K metric -- an upper-bound estimator of LM reasoning capabilities -- even when evaluated with extremely large K values, pushing the boundaries of LM reasoning. △ Less

Submitted 17 June, 2025; originally announced June 2025.

FinLMM-R1: Enhancing Financial Reasoning in LMM through Scalable Data and Reward Design

Authors: Kai Lan, Jiayong Zhu, Jiangtong Li, Dawei Cheng, Guang Chen, Changjun Jiang

Abstract: Large Multimodal Models (LMMs) demonstrate significant cross-modal reasoning capabilities. However, financial applications face challenges due to the lack of high-quality multimodal reasoning datasets and the inefficiency of existing training paradigms for reasoning enhancement. To address these issues, we propose an integrated framework, FinLMM-R1, combining an automated and scalable pipeline for… ▽ More Large Multimodal Models (LMMs) demonstrate significant cross-modal reasoning capabilities. However, financial applications face challenges due to the lack of high-quality multimodal reasoning datasets and the inefficiency of existing training paradigms for reasoning enhancement. To address these issues, we propose an integrated framework, FinLMM-R1, combining an automated and scalable pipeline for data construction with enhanced training strategies to improve the multimodal reasoning of LMM. The Automated and Scalable Pipeline (ASP) resolves textual-visual misalignment in financial reports through a separate paradigm of question-answer generation and image-question alignment, ensuring data integrity and extraction efficiency. Through ASP, we collect 89,378 aligned image-question pairs from 23,397 financial reports, covering tasks such as arithmetic reasoning, statistics reasoning, financial explanation, and financial knowledge. Moreover, we introduce the Thinking with Adversarial Reward in LMM (TAR-LMM), extending the prior two-stage training framework [1] with additional reward mechanisms. In the first stage, we focus on text-only tasks with format and accuracy rewards to guide the model in generating well-structured thinking contents. In the second stage, we construct multi-image contrastive samples with additional reward components including image selection, thinking content length, and adversarial reward to jointly optimize the LMM across visual perception, reasoning efficiency, and logical coherence. Extensive experiments on 7 benchmarks show ASP-derived dataset and training framework significantly improve answer accuracy and reasoning depth over existing reasoning LMMs in both general and financial multimodal contexts. △ Less

Submitted 15 June, 2025; originally announced June 2025.

Comments: 26 pages, 16 figures

CFBenchmark-MM: Chinese Financial Assistant Benchmark for Multimodal Large Language Model

Authors: Jiangtong Li, Yiyun Zhu, Dawei Cheng, Zhijun Ding, Changjun Jiang

Abstract: Multimodal Large Language Models (MLLMs) have rapidly evolved with the growth of Large Language Models (LLMs) and are now applied in various fields. In finance, the integration of diverse modalities such as text, charts, and tables is crucial for accurate and efficient decision-making. Therefore, an effective evaluation system that incorporates these data types is essential for advancing financial… ▽ More Multimodal Large Language Models (MLLMs) have rapidly evolved with the growth of Large Language Models (LLMs) and are now applied in various fields. In finance, the integration of diverse modalities such as text, charts, and tables is crucial for accurate and efficient decision-making. Therefore, an effective evaluation system that incorporates these data types is essential for advancing financial application. In this paper, we introduce CFBenchmark-MM, a Chinese multimodal financial benchmark with over 9,000 image-question pairs featuring tables, histogram charts, line charts, pie charts, and structural diagrams. Additionally, we develop a staged evaluation system to assess MLLMs in handling multimodal information by providing different visual content step by step. Despite MLLMs having inherent financial knowledge, experimental results still show limited efficiency and robustness in handling multimodal financial context. Further analysis on incorrect responses reveals the misinterpretation of visual content and the misunderstanding of financial concepts are the primary issues. Our research validates the significant, yet underexploited, potential of MLLMs in financial analysis, highlighting the need for further development and domain-specific optimization to encourage the enhanced use in financial domain. △ Less

Submitted 15 June, 2025; originally announced June 2025.

Comments: 22 pages, 9 figures

EKPC: Elastic Knowledge Preservation and Compensation for Class-Incremental Learning

Authors: Huaijie Wang, De Cheng, Lingfeng He, Yan Li, Jie Li, Nannan Wang, Xinbo Gao

Abstract: Class-Incremental Learning (CIL) aims to enable AI models to continuously learn from sequentially arriving data of different classes over time while retaining previously acquired knowledge. Recently, Parameter-Efficient Fine-Tuning (PEFT) methods, like prompt pool-based approaches and adapter tuning, have shown great attraction in CIL. However, these methods either introduce additional parameters… ▽ More Class-Incremental Learning (CIL) aims to enable AI models to continuously learn from sequentially arriving data of different classes over time while retaining previously acquired knowledge. Recently, Parameter-Efficient Fine-Tuning (PEFT) methods, like prompt pool-based approaches and adapter tuning, have shown great attraction in CIL. However, these methods either introduce additional parameters that increase memory usage, or rely on rigid regularization techniques which reduce forgetting but compromise model flexibility. To overcome these limitations, we propose the Elastic Knowledge Preservation and Compensation (EKPC) method, integrating Importance-aware Parameter Regularization (IPR) and Trainable Semantic Drift Compensation (TSDC) for CIL. Specifically, the IPR method assesses the sensitivity of network parameters to prior tasks using a novel parameter-importance algorithm. It then selectively constrains updates within the shared adapter according to these importance values, thereby preserving previously acquired knowledge while maintaining the model's flexibility. However, it still exhibits slight semantic differences in previous knowledge to accommodate new incremental tasks, leading to decision boundaries confusion in classifier. To eliminate this confusion, TSDC trains a unified classifier by compensating prototypes with trainable semantic drift. Extensive experiments on five CIL benchmarks demonstrate the effectiveness of the proposed method, showing superior performances to existing state-of-the-art methods. △ Less

Submitted 14 June, 2025; originally announced June 2025.

How to Train a Model on a Cheap Cluster with Low Cost using Block Coordinate Descent

Authors: Zeyu Liu, Yunquan Zhang, Boyang Zhang, Guoyong Jiang, Daning Cheng

Abstract: Training large language models typically demands extensive GPU memory and substantial financial investment, which poses a barrier for many small- to medium-sized teams. In this paper, we present a full-parameter pre-training framework based on block coordinate descent (BCD), augmented with engineering optimizations, to efficiently train large models on affordable RTX 4090 GPU clusters. BCD ensures… ▽ More Training large language models typically demands extensive GPU memory and substantial financial investment, which poses a barrier for many small- to medium-sized teams. In this paper, we present a full-parameter pre-training framework based on block coordinate descent (BCD), augmented with engineering optimizations, to efficiently train large models on affordable RTX 4090 GPU clusters. BCD ensures model convergence based on block coordinate descent theory and performs gradient computation and update at the level of parameter blocks. Experiments show that 1) Lower cost of Same-Device: BCD significantly reduces pre-training cost. For the 7B model, under identical hardware settings, BCD lowers training costs to approximately 33% on A100,A800 clusters on 7B model averagely and to approximately 2.6% on RTX 4090 clusters on 7B model, compared to traditional full-parameter training. 2) Cross-Device Transfer: By leveraging BCD, large-scale models previously trainable only on high-end A100 clusters can be seamlessly migrated and pre-trained on 4090 clusters-whose hourly cost is only one-quarter that of A100-without requiring expensive hardware. 3) Accuracy Retention: In both scenarios, BCD training achieves the same level of model accuracy as full-parameter pre-training. △ Less

Submitted 22 May, 2025; originally announced June 2025.

Comments: under review

Semi-Tensor-Product Based Convolutional Neural Networks

Authors: Daizhan Cheng

Abstract: The semi-tensor product (STP) of vectors is a generalization of conventional inner product of vectors, which allows the factor vectors to of different dimensions. This paper proposes a domain-based convolutional product (CP). Combining domain-based CP with STP of vectors, a new CP is proposed. Since there is no zero or any other padding, it can avoid the junk information caused by padding. Using i… ▽ More The semi-tensor product (STP) of vectors is a generalization of conventional inner product of vectors, which allows the factor vectors to of different dimensions. This paper proposes a domain-based convolutional product (CP). Combining domain-based CP with STP of vectors, a new CP is proposed. Since there is no zero or any other padding, it can avoid the junk information caused by padding. Using it, the STP-based convolutional neural network (CNN) is developed. Its application to image and third order signal identifications is considered. △ Less

Submitted 12 June, 2025; originally announced June 2025.

Language Model-Enhanced Message Passing for Heterophilic Graph Learning

Authors: Wenjun Wang, Dawei Cheng

Abstract: Traditional graph neural networks (GNNs), which rely on homophily-driven message passing, struggle with heterophilic graphs where connected nodes exhibit dissimilar features and different labels. While existing methods address heterophily through graph structure refinement or adaptation of neighbor aggregation functions, they often overlook the semantic potential of node text, rely on suboptimal m… ▽ More Traditional graph neural networks (GNNs), which rely on homophily-driven message passing, struggle with heterophilic graphs where connected nodes exhibit dissimilar features and different labels. While existing methods address heterophily through graph structure refinement or adaptation of neighbor aggregation functions, they often overlook the semantic potential of node text, rely on suboptimal message representation for propagation and compromise performance on homophilic graphs. To address these limitations, we propose a novel language model (LM)-enhanced message passing approach for heterophilic graph leaning (LEMP4HG). Specifically, in the context of text-attributed graph, we provide paired node texts for LM to generate their connection analysis, which are encoded and then fused with paired node textual embeddings through a gating mechanism. The synthesized messages are semantically enriched and adaptively balanced with both nodes' information, which mitigates contradictory signals when neighbor aggregation in heterophilic regions. Furthermore, we introduce an active learning strategy guided by our heuristic MVRD (Modulated Variation of Reliable Distance), selectively enhancing node pairs suffer most from message passing, reducing the cost of analysis generation and side effects on homophilic regions. Extensive experiments validate that our approach excels on heterophilic graphs and performs robustly on homophilic ones, with a graph convolutional network (GCN) backbone and a practical budget. △ Less

Submitted 26 May, 2025; originally announced May 2025.

Can LLMs Alleviate Catastrophic Forgetting in Graph Continual Learning? A Systematic Study

Authors: Ziyang Cheng, Zhixun Li, Yuhan Li, Yixin Song, Kangyi Zhao, Dawei Cheng, Jia Li, Jeffrey Xu Yu

Abstract: Nowadays, real-world data, including graph-structure data, often arrives in a streaming manner, which means that learning systems need to continuously acquire new knowledge without forgetting previously learned information. Although substantial existing works attempt to address catastrophic forgetting in graph machine learning, they are all based on training from scratch with streaming data. With… ▽ More Nowadays, real-world data, including graph-structure data, often arrives in a streaming manner, which means that learning systems need to continuously acquire new knowledge without forgetting previously learned information. Although substantial existing works attempt to address catastrophic forgetting in graph machine learning, they are all based on training from scratch with streaming data. With the rise of pretrained models, an increasing number of studies have leveraged their strong generalization ability for continual learning. Therefore, in this work, we attempt to answer whether large language models (LLMs) can mitigate catastrophic forgetting in Graph Continual Learning (GCL). We first point out that current experimental setups for GCL have significant flaws, as the evaluation stage may lead to task ID leakage. Then, we evaluate the performance of LLMs in more realistic scenarios and find that even minor modifications can lead to outstanding results. Finally, based on extensive experiments, we propose a simple-yet-effective method, Simple Graph Continual Learning (SimGCL), that surpasses the previous state-of-the-art GNN-based baseline by around 20% under the rehearsal-free constraint. To facilitate reproducibility, we have developed an easy-to-use benchmark LLM4GCL for training and evaluating existing GCL methods. The code is available at: https://github.com/ZhixunLEE/LLM4GCL. △ Less

Submitted 24 May, 2025; originally announced May 2025.

A Novel Generative Model with Causality Constraint for Mitigating Biases in Recommender Systems

Authors: Jianfeng Deng, Qingfeng Chen, Debo Cheng, Jiuyong Li, Lin Liu, Shichao Zhang

Abstract: Accurately predicting counterfactual user feedback is essential for building effective recommender systems. However, latent confounding bias can obscure the true causal relationship between user feedback and item exposure, ultimately degrading recommendation performance. Existing causal debiasing approaches often rely on strong assumptions-such as the availability of instrumental variables (IVs) o… ▽ More Accurately predicting counterfactual user feedback is essential for building effective recommender systems. However, latent confounding bias can obscure the true causal relationship between user feedback and item exposure, ultimately degrading recommendation performance. Existing causal debiasing approaches often rely on strong assumptions-such as the availability of instrumental variables (IVs) or strong correlations between latent confounders and proxy variables-that are rarely satisfied in real-world scenarios. To address these limitations, we propose a novel generative framework called Latent Causality Constraints for Debiasing representation learning in Recommender Systems (LCDR). Specifically, LCDR leverages an identifiable Variational Autoencoder (iVAE) as a causal constraint to align the latent representations learned by a standard Variational Autoencoder (VAE) through a unified loss function. This alignment allows the model to leverage even weak or noisy proxy variables to recover latent confounders effectively. The resulting representations are then used to improve recommendation performance. Extensive experiments on three real-world datasets demonstrate that LCDR consistently outperforms existing methods in both mitigating bias and improving recommendation accuracy. △ Less

Submitted 22 May, 2025; originally announced May 2025.

Comments: 11 pages

StPR: Spatiotemporal Preservation and Routing for Exemplar-Free Video Class-Incremental Learning

Authors: Huaijie Wang, De Cheng, Guozhang Li, Zhipeng Xu, Lingfeng He, Jie Li, Nannan Wang, Xinbo Gao

Abstract: Video Class-Incremental Learning (VCIL) seeks to develop models that continuously learn new action categories over time without forgetting previously acquired knowledge. Unlike traditional Class-Incremental Learning (CIL), VCIL introduces the added complexity of spatiotemporal structures, making it particularly challenging to mitigate catastrophic forgetting while effectively capturing both frame-… ▽ More Video Class-Incremental Learning (VCIL) seeks to develop models that continuously learn new action categories over time without forgetting previously acquired knowledge. Unlike traditional Class-Incremental Learning (CIL), VCIL introduces the added complexity of spatiotemporal structures, making it particularly challenging to mitigate catastrophic forgetting while effectively capturing both frame-shared semantics and temporal dynamics. Existing approaches either rely on exemplar rehearsal, raising concerns over memory and privacy, or adapt static image-based methods that neglect temporal modeling. To address these limitations, we propose Spatiotemporal Preservation and Routing (StPR), a unified and exemplar-free VCIL framework that explicitly disentangles and preserves spatiotemporal information. First, we introduce Frame-Shared Semantics Distillation (FSSD), which identifies semantically stable and meaningful channels by jointly considering semantic sensitivity and classification contribution. These important semantic channels are selectively regularized to maintain prior knowledge while allowing for adaptation. Second, we design a Temporal Decomposition-based Mixture-of-Experts (TD-MoE), which dynamically routes task-specific experts based on their temporal dynamics, enabling inference without task ID or stored exemplars. Together, StPR effectively leverages spatial semantics and temporal dynamics, achieving a unified, exemplar-free VCIL framework. Extensive experiments on UCF101, HMDB51, and Kinetics400 show that our method outperforms existing baselines while offering improved interpretability and efficiency in VCIL. Code is available in the supplementary materials. △ Less

Submitted 20 May, 2025; originally announced May 2025.

Semantic-Aligned Learning with Collaborative Refinement for Unsupervised VI-ReID

Authors: De Cheng, Lingfeng He, Nannan Wang, Dingwen Zhang, Xinbo Gao

Abstract: Unsupervised visible-infrared person re-identification (USL-VI-ReID) seeks to match pedestrian images of the same individual across different modalities without human annotations for model learning. Previous methods unify pseudo-labels of cross-modality images through label association algorithms and then design contrastive learning framework for global feature learning. However, these methods ove… ▽ More Unsupervised visible-infrared person re-identification (USL-VI-ReID) seeks to match pedestrian images of the same individual across different modalities without human annotations for model learning. Previous methods unify pseudo-labels of cross-modality images through label association algorithms and then design contrastive learning framework for global feature learning. However, these methods overlook the cross-modality variations in feature representation and pseudo-label distributions brought by fine-grained patterns. This insight results in insufficient modality-shared learning when only global features are optimized. To address this issue, we propose a Semantic-Aligned Learning with Collaborative Refinement (SALCR) framework, which builds up optimization objective for specific fine-grained patterns emphasized by each modality, thereby achieving complementary alignment between the label distributions of different modalities. Specifically, we first introduce a Dual Association with Global Learning (DAGI) module to unify the pseudo-labels of cross-modality instances in a bi-directional manner. Afterward, a Fine-Grained Semantic-Aligned Learning (FGSAL) module is carried out to explore part-level semantic-aligned patterns emphasized by each modality from cross-modality instances. Optimization objective is then formulated based on the semantic-aligned features and their corresponding label space. To alleviate the side-effects arising from noisy pseudo-labels, we propose a Global-Part Collaborative Refinement (GPCR) module to mine reliable positive sample sets for the global and part features dynamically and optimize the inter-instance relationships. Extensive experiments demonstrate the effectiveness of the proposed method, which achieves superior performances to state-of-the-art methods. Our code is available at \href{https://github.com/FranklinLingfeng/code-for-SALCR}. △ Less

Submitted 5 May, 2025; v1 submitted 27 April, 2025; originally announced April 2025.

Comments: Accepted by IJCV 2025

On Dimension-Free Transformer: An Application of STP to AI

Authors: Daizhan Cheng

Abstract: The matrix expressions for every parts of a transformer are firstly described. Based on semi-tensor product (STP) of matrices the hypervectors are reconsidered and the linear transformation over hypervectors is constructed by using projection. Its properties and calculating formulas are obtained. Using projection-based transformation of hypervector (PBTH), the framework of dimension-free transform… ▽ More The matrix expressions for every parts of a transformer are firstly described. Based on semi-tensor product (STP) of matrices the hypervectors are reconsidered and the linear transformation over hypervectors is constructed by using projection. Its properties and calculating formulas are obtained. Using projection-based transformation of hypervector (PBTH), the framework of dimension-free transformer (DFT) is proposed by verifying each linear transformation in a transformer and replacing it by a proper PBTH, which allows the inputs and outputs being of arbitrary dimensions. Using balanced information about all entries, DFT must be more efficient in dealing with signals. △ Less

Submitted 20 April, 2025; originally announced April 2025.

Stochastic Gradient Descent in Non-Convex Problems: Asymptotic Convergence with Relaxed Step-Size via Stopping Time Methods

Authors: Ruinan Jin, Difei Cheng, Hong Qiao, Xin Shi, Shaodong Liu, Bo Zhang

Abstract: Stochastic Gradient Descent (SGD) is widely used in machine learning research. Previous convergence analyses of SGD under the vanishing step-size setting typically require Robbins-Monro conditions. However, in practice, a wider variety of step-size schemes are frequently employed, yet existing convergence results remain limited and often rely on strong assumptions. This paper bridges this gap by i… ▽ More Stochastic Gradient Descent (SGD) is widely used in machine learning research. Previous convergence analyses of SGD under the vanishing step-size setting typically require Robbins-Monro conditions. However, in practice, a wider variety of step-size schemes are frequently employed, yet existing convergence results remain limited and often rely on strong assumptions. This paper bridges this gap by introducing a novel analytical framework based on a stopping-time method, enabling asymptotic convergence analysis of SGD under more relaxed step-size conditions and weaker assumptions. In the non-convex setting, we prove the almost sure convergence of SGD iterates for step-sizes $ \{ ε_t \}_{t \geq 1} $ satisfying $\sum_{t=1}^{+\infty} ε_t = +\infty$ and $\sum_{t=1}^{+\infty} ε_t^p < +\infty$ for some $p > 2$. Compared with previous studies, our analysis eliminates the global Lipschitz continuity assumption on the loss function and relaxes the boundedness requirements for higher-order moments of stochastic gradients. Building upon the almost sure convergence results, we further establish $L_2$ convergence. These significantly relaxed assumptions make our theoretical results more general, thereby enhancing their applicability in practical scenarios. △ Less

Submitted 16 April, 2025; originally announced April 2025.

Comments: 42 pages

MSC Class: 40G15 ACM Class: G.1.0

Can the capability of Large Language Models be described by human ability? A Meta Study

Authors: Mingrui Zan, Yunquan Zhang, Boyang Zhang, Fangming Liu, Daning Cheng

Abstract: Users of Large Language Models (LLMs) often perceive these models as intelligent entities with human-like capabilities. However, the extent to which LLMs' capabilities truly approximate human abilities remains a topic of debate. In this paper, to characterize the capabilities of LLMs in relation to human capabilities, we collected performance data from over 80 models across 37 evaluation benchmark… ▽ More Users of Large Language Models (LLMs) often perceive these models as intelligent entities with human-like capabilities. However, the extent to which LLMs' capabilities truly approximate human abilities remains a topic of debate. In this paper, to characterize the capabilities of LLMs in relation to human capabilities, we collected performance data from over 80 models across 37 evaluation benchmarks. The evaluation benchmarks are categorized into 6 primary abilities and 11 sub-abilities in human aspect. Then, we then clustered the performance rankings into several categories and compared these clustering results with classifications based on human ability aspects. Our findings lead to the following conclusions: 1. We have confirmed that certain capabilities of LLMs with fewer than 10 billion parameters can indeed be described using human ability metrics; 2. While some abilities are considered interrelated in humans, they appear nearly uncorrelated in LLMs; 3. The capabilities possessed by LLMs vary significantly with the parameter scale of the model. △ Less

Submitted 13 April, 2025; originally announced April 2025.

Gemma 3 Technical Report

Authors: Gemma Team, Aishwarya Kamath, Johan Ferret, Shreya Pathak, Nino Vieillard, Ramona Merhej, Sarah Perrin, Tatiana Matejovicova, Alexandre Ramé, Morgane Rivière, Louis Rouillard, Thomas Mesnard, Geoffrey Cideron, Jean-bastien Grill, Sabela Ramos, Edouard Yvinec, Michelle Casbon, Etienne Pot, Ivo Penchev, Gaël Liu, Francesco Visin, Kathleen Kenealy, Lucas Beyer, Xiaohai Zhai, Anton Tsitsulin , et al. (191 additional authors not shown)

Abstract: We introduce Gemma 3, a multimodal addition to the Gemma family of lightweight open models, ranging in scale from 1 to 27 billion parameters. This version introduces vision understanding abilities, a wider coverage of languages and longer context - at least 128K tokens. We also change the architecture of the model to reduce the KV-cache memory that tends to explode with long context. This is achie… ▽ More We introduce Gemma 3, a multimodal addition to the Gemma family of lightweight open models, ranging in scale from 1 to 27 billion parameters. This version introduces vision understanding abilities, a wider coverage of languages and longer context - at least 128K tokens. We also change the architecture of the model to reduce the KV-cache memory that tends to explode with long context. This is achieved by increasing the ratio of local to global attention layers, and keeping the span on local attention short. The Gemma 3 models are trained with distillation and achieve superior performance to Gemma 2 for both pre-trained and instruction finetuned versions. In particular, our novel post-training recipe significantly improves the math, chat, instruction-following and multilingual abilities, making Gemma3-4B-IT competitive with Gemma2-27B-IT and Gemma3-27B-IT comparable to Gemini-1.5-Pro across benchmarks. We release all our models to the community. △ Less

Submitted 25 March, 2025; originally announced March 2025.

Leveraging Deep Neural Networks for Aspect-Based Sentiment Classification

Authors: Chen Li, Debo Cheng, Yasuhiko Morimoto

Abstract: Aspect-based sentiment analysis seeks to determine sentiment with a high level of detail. While graph convolutional networks (GCNs) are commonly used for extracting sentiment features, their straightforward use in syntactic feature extraction can lead to a loss of crucial information. This paper presents a novel edge-enhanced GCN, called EEGCN, which improves performance by preserving feature inte… ▽ More Aspect-based sentiment analysis seeks to determine sentiment with a high level of detail. While graph convolutional networks (GCNs) are commonly used for extracting sentiment features, their straightforward use in syntactic feature extraction can lead to a loss of crucial information. This paper presents a novel edge-enhanced GCN, called EEGCN, which improves performance by preserving feature integrity as it processes syntactic graphs. We incorporate a bidirectional long short-term memory (Bi-LSTM) network alongside a self-attention-based transformer for effective text encoding, ensuring the retention of long-range dependencies. A bidirectional GCN (Bi-GCN) with message passing then captures the relationships between entities, while an aspect-specific masking technique removes extraneous information. Extensive evaluations and ablation studies on four benchmark datasets show that EEGCN significantly enhances aspect-based sentiment analysis, overcoming issues with syntactic feature extraction and advancing the field's methodologies. △ Less

Submitted 17 March, 2025; originally announced March 2025.

Exploring Interpretability for Visual Prompt Tuning with Hierarchical Concepts

Authors: Yubin Wang, Xinyang Jiang, De Cheng, Xiangqian Zhao, Zilong Wang, Dongsheng Li, Cairong Zhao

Abstract: Visual prompt tuning offers significant advantages for adapting pre-trained visual foundation models to specific tasks. However, current research provides limited insight into the interpretability of this approach, which is essential for enhancing AI reliability and enabling AI-driven knowledge discovery. In this paper, rather than learning abstract prompt embeddings, we propose the first framewor… ▽ More Visual prompt tuning offers significant advantages for adapting pre-trained visual foundation models to specific tasks. However, current research provides limited insight into the interpretability of this approach, which is essential for enhancing AI reliability and enabling AI-driven knowledge discovery. In this paper, rather than learning abstract prompt embeddings, we propose the first framework, named Interpretable Visual Prompt Tuning (IVPT), to explore interpretability for visual prompts, by introducing hierarchical concept prototypes. Specifically, visual prompts are linked to human-understandable semantic concepts, represented as a set of category-agnostic prototypes, each corresponding to a specific region of the image. Then, IVPT aggregates features from these regions to generate interpretable prompts, which are structured hierarchically to explain visual prompts at different granularities. Comprehensive qualitative and quantitative evaluations on fine-grained classification benchmarks show its superior interpretability and performance over conventional visual prompt tuning methods and existing interpretable methods. △ Less

Submitted 8 March, 2025; originally announced March 2025.

Comments: 10 pages, 9 figures

An Empirical Study on Eliciting and Improving R1-like Reasoning Models

Authors: Zhipeng Chen, Yingqian Min, Beichen Zhang, Jie Chen, Jinhao Jiang, Daixuan Cheng, Wayne Xin Zhao, Zheng Liu, Xu Miao, Yang Lu, Lei Fang, Zhongyuan Wang, Ji-Rong Wen

Abstract: In this report, we present the third technical report on the development of slow-thinking models as part of the STILL project. As the technical pathway becomes clearer, scaling RL training has become a central technique for implementing such reasoning models. We systematically experiment with and document the effects of various factors influencing RL training, conducting experiments on both base m… ▽ More In this report, we present the third technical report on the development of slow-thinking models as part of the STILL project. As the technical pathway becomes clearer, scaling RL training has become a central technique for implementing such reasoning models. We systematically experiment with and document the effects of various factors influencing RL training, conducting experiments on both base models and fine-tuned models. Specifically, we demonstrate that our RL training approach consistently improves the Qwen2.5-32B base models, enhancing both response length and test accuracy. Furthermore, we show that even when a model like DeepSeek-R1-Distill-Qwen-1.5B has already achieved a high performance level, it can be further refined through RL training, reaching an accuracy of 39.33% on AIME 2024. Beyond RL training, we also explore the use of tool manipulation, finding that it significantly boosts the reasoning performance of large reasoning models. This approach achieves a remarkable accuracy of 86.67% with greedy search on AIME 2024, underscoring its effectiveness in enhancing model capabilities. We release our resources at the STILL project website: https://github.com/RUCAIBox/Slow_Thinking_with_LLMs. △ Less

Submitted 6 March, 2025; originally announced March 2025.

Comments: Technical Report on Slow Thinking with LLMs: Part III

Effective High-order Graph Representation Learning for Credit Card Fraud Detection

Authors: Yao Zou, Dawei Cheng

Abstract: Credit card fraud imposes significant costs on both cardholders and issuing banks. Fraudsters often disguise their crimes, such as using legitimate transactions through several benign users to bypass anti-fraud detection. Existing graph neural network (GNN) models struggle with learning features of camouflaged, indirect multi-hop transactions due to their inherent over-smoothing issues in deep mul… ▽ More Credit card fraud imposes significant costs on both cardholders and issuing banks. Fraudsters often disguise their crimes, such as using legitimate transactions through several benign users to bypass anti-fraud detection. Existing graph neural network (GNN) models struggle with learning features of camouflaged, indirect multi-hop transactions due to their inherent over-smoothing issues in deep multi-layer aggregation, presenting a major challenge in detecting disguised relationships. Therefore, in this paper, we propose a novel High-order Graph Representation Learning model (HOGRL) to avoid incorporating excessive noise during the multi-layer aggregation process. In particular, HOGRL learns different orders of \emph{pure} representations directly from high-order transaction graphs. We realize this goal by effectively constructing high-order transaction graphs first and then learning the \emph{pure} representations of each order so that the model could identify fraudsters' multi-hop indirect transactions via multi-layer \emph{pure} feature learning. In addition, we introduce a mixture-of-expert attention mechanism to automatically determine the importance of different orders for jointly optimizing fraud detection performance. We conduct extensive experiments in both the open source and real-world datasets, the result demonstrates the significant improvements of our proposed HOGRL compared with state-of-the-art fraud detection baselines. HOGRL's superior performance also proves its effectiveness in addressing high-order fraud camouflage criminals. △ Less

Submitted 3 March, 2025; originally announced March 2025.

Comments: 9 pages, 5 figures, accepted at IJCAI 2024

MSC Class: 68T07; 91B06 ACM Class: I.2.6; H.2.8

Journal ref: Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence (IJCAI 2024), pages 7581-7589

FinTSB: A Comprehensive and Practical Benchmark for Financial Time Series Forecasting

Authors: Yifan Hu, Yuante Li, Peiyuan Liu, Yuxia Zhu, Naiqi Li, Tao Dai, Shu-tao Xia, Dawei Cheng, Changjun Jiang

Abstract: Financial time series (FinTS) record the behavior of human-brain-augmented decision-making, capturing valuable historical information that can be leveraged for profitable investment strategies. Not surprisingly, this area has attracted considerable attention from researchers, who have proposed a wide range of methods based on various backbones. However, the evaluation of the area often exhibits th… ▽ More Financial time series (FinTS) record the behavior of human-brain-augmented decision-making, capturing valuable historical information that can be leveraged for profitable investment strategies. Not surprisingly, this area has attracted considerable attention from researchers, who have proposed a wide range of methods based on various backbones. However, the evaluation of the area often exhibits three systemic limitations: 1. Failure to account for the full spectrum of stock movement patterns observed in dynamic financial markets. (Diversity Gap), 2. The absence of unified assessment protocols undermines the validity of cross-study performance comparisons. (Standardization Deficit), and 3. Neglect of critical market structure factors, resulting in inflated performance metrics that lack practical applicability. (Real-World Mismatch). Addressing these limitations, we propose FinTSB, a comprehensive and practical benchmark for financial time series forecasting (FinTSF). To increase the variety, we categorize movement patterns into four specific parts, tokenize and pre-process the data, and assess the data quality based on some sequence characteristics. To eliminate biases due to different evaluation settings, we standardize the metrics across three dimensions and build a user-friendly, lightweight pipeline incorporating methods from various backbones. To accurately simulate real-world trading scenarios and facilitate practical implementation, we extensively model various regulatory constraints, including transaction fees, among others. Finally, we conduct extensive experiments on FinTSB, highlighting key insights to guide model selection under varying market conditions. Overall, FinTSB provides researchers with a novel and comprehensive platform for improving and evaluating FinTSF methods. The code is available at https://github.com/TongjiFinLab/FinTSBenchmark. △ Less

Submitted 26 February, 2025; originally announced February 2025.

A General Error-Theoretical Analysis Framework for Constructing Compression Strategies

Authors: Boyang Zhang, Daning Cheng, Yunquan Zhang, Meiqi Tu, Fangmin Liu, Jiake Tian

Abstract: The exponential growth in parameter size and computational complexity of deep models poses significant challenges for efficient deployment. The core problem of existing compression methods is that different layers of the model have significant differences in their tolerance to compression levels. For instance, the first layer of a model can typically sustain a higher compression level compared to… ▽ More The exponential growth in parameter size and computational complexity of deep models poses significant challenges for efficient deployment. The core problem of existing compression methods is that different layers of the model have significant differences in their tolerance to compression levels. For instance, the first layer of a model can typically sustain a higher compression level compared to the last layer without compromising performance. Thus, the key challenge lies in how to allocate compression levels across layers in a way that minimizes performance loss while maximizing parameter reduction. To address this challenge, we propose a Compression Error Theory (CET) framework, designed to determine the optimal compression level for each layer. Taking quantization as an example, CET leverages differential expansion and algebraic geometry to reconstruct the quadratic form of quantization error as ellipsoids and hyperbolic paraboloids, and utilizes their geometric structures to define an error subspace. To identify the error subspace with minimal performance loss, by performing orthogonal decomposition of the geometric space, CET transforms the optimization process of the error subspace into a complementary problem. The final theoretical analysis shows that constructing the quantization subspace along the major axis results in minimal performance degradation. Through experimental verification of the theory, CET can greatly retain performance while compressing. Specifically, on the ResNet-34 model, CET achieves nearly 11$\times$ parameter compression while even surpassing performance comparable to the original model. △ Less

Submitted 19 February, 2025; originally announced February 2025.

Comments: Under Review

V-SQL: A View-based Two-stage Text-to-SQL Framework

Authors: Zeshun You, Jiebin Yao, Dong Cheng, Zhiwei Wen, Zhiliang Lu, Xianyi Shen

Abstract: The text-to-SQL task aims to convert natural language into Structured Query Language (SQL) without bias. Recently, text-to-SQL methods based on large language models (LLMs) have garnered significant attention. The core of mainstream text-to-SQL frameworks is schema linking, which aligns user queries with relevant tables and columns in the database. Previous methods focused on schema linking while… ▽ More The text-to-SQL task aims to convert natural language into Structured Query Language (SQL) without bias. Recently, text-to-SQL methods based on large language models (LLMs) have garnered significant attention. The core of mainstream text-to-SQL frameworks is schema linking, which aligns user queries with relevant tables and columns in the database. Previous methods focused on schema linking while neglecting to enhance LLMs' understanding of database schema. The complex coupling relationships between tables in the database constrain the SQL generation capabilities of LLMs. To tackle this issue, this paper proposes a simple yet effective strategy called view-based schema. This strategy aids LLMs in understanding the database schema by decoupling tightly coupled tables into low-coupling views. We then introduce V-SQL, a view-based two-stage text-to-SQL framework. V-SQL involves the view-based schema strategy to enhance LLMs' understanding of database schema. Results on the authoritative datasets Bird indicate that V-SQL achieves competitive performance compared to existing state-of-the-art methods. △ Less

Submitted 16 December, 2024; originally announced February 2025.

Comments: 10 pages,5 figures,

ActionPiece: Contextually Tokenizing Action Sequences for Generative Recommendation

Authors: Yupeng Hou, Jianmo Ni, Zhankui He, Noveen Sachdeva, Wang-Cheng Kang, Ed H. Chi, Julian McAuley, Derek Zhiyuan Cheng

Abstract: Generative recommendation (GR) is an emerging paradigm where user actions are tokenized into discrete token patterns and autoregressively generated as predictions. However, existing GR models tokenize each action independently, assigning the same fixed tokens to identical actions across all sequences without considering contextual relationships. This lack of context-awareness can lead to suboptima… ▽ More Generative recommendation (GR) is an emerging paradigm where user actions are tokenized into discrete token patterns and autoregressively generated as predictions. However, existing GR models tokenize each action independently, assigning the same fixed tokens to identical actions across all sequences without considering contextual relationships. This lack of context-awareness can lead to suboptimal performance, as the same action may hold different meanings depending on its surrounding context. To address this issue, we propose ActionPiece to explicitly incorporate context when tokenizing action sequences. In ActionPiece, each action is represented as a set of item features. Given the action sequence corpora, we construct the vocabulary by merging feature patterns as new tokens, based on their co-occurrence frequency both within individual sets and across adjacent sets. Considering the unordered nature of feature sets, we further introduce set permutation regularization, which produces multiple segmentations of action sequences with the same semantics. Our code is available at: https://github.com/google-deepmind/action_piece. △ Less

Submitted 6 June, 2025; v1 submitted 19 February, 2025; originally announced February 2025.

Comments: ICML 2025 (Spotlight)

MasRouter: Learning to Route LLMs for Multi-Agent Systems

Authors: Yanwei Yue, Guibin Zhang, Boyang Liu, Guancheng Wan, Kun Wang, Dawei Cheng, Yiyan Qi

Abstract: Multi-agent systems (MAS) powered by Large Language Models (LLMs) have been demonstrated to push the boundaries of LLM capabilities, yet they often incur significant costs and face challenges in dynamic LLM selection. Current LLM routing methods effectively reduce overhead in single-agent scenarios by customizing LLM selection for each query, but they overlook the critical decisions regarding coll… ▽ More Multi-agent systems (MAS) powered by Large Language Models (LLMs) have been demonstrated to push the boundaries of LLM capabilities, yet they often incur significant costs and face challenges in dynamic LLM selection. Current LLM routing methods effectively reduce overhead in single-agent scenarios by customizing LLM selection for each query, but they overlook the critical decisions regarding collaboration modes and agent roles in MAS. In response to this challenge, we first introduce the problem of Multi-Agent System Routing (MASR), which integrates all components of MAS into a unified routing framework. Toward this goal, we propose MasRouter, the first high-performing, cost-effective, and inductive MASR solution. MasRouter employs collaboration mode determination, role allocation, and LLM routing through a cascaded controller network, progressively constructing a MAS that balances effectiveness and efficiency. Extensive experiments demonstrate that MasRouter is (1) high-performing, achieving a $1.8\%\sim8.2\%$ improvement over the state-of-the-art method on MBPP; (2) economical, reducing overhead by up to $52.07\%$ compared to SOTA methods on HumanEval; and (3) plug-and-play, seamlessly integrating with mainstream MAS frameworks, reducing overhead by $17.21\%\sim28.17\%$ via customized routing. The code is available at https://github.com/yanweiyue/masrouter. △ Less

Submitted 16 February, 2025; originally announced February 2025.

FinMamba: Market-Aware Graph Enhanced Multi-Level Mamba for Stock Movement Prediction

Authors: Yifan Hu, Peiyuan Liu, Yuante Li, Dawei Cheng, Naiqi Li, Tao Dai, Jigang Bao, Shu-Tao Xia

Abstract: Recently, combining stock features with inter-stock correlations has become a common and effective approach for stock movement prediction. However, financial data presents significant challenges due to its low signal-to-noise ratio and the dynamic complexity of the market, which give rise to two key limitations in existing methods. First, the relationships between stocks are highly influenced by m… ▽ More Recently, combining stock features with inter-stock correlations has become a common and effective approach for stock movement prediction. However, financial data presents significant challenges due to its low signal-to-noise ratio and the dynamic complexity of the market, which give rise to two key limitations in existing methods. First, the relationships between stocks are highly influenced by multifaceted factors including macroeconomic market dynamics, and current models fail to adaptively capture these evolving interactions under specific market conditions. Second, for the accuracy and timeliness required by real-world trading, existing financial data mining methods struggle to extract beneficial pattern-oriented dependencies from long historical data while maintaining high efficiency and low memory consumption. To address the limitations, we propose FinMamba, a Mamba-GNN-based framework for market-aware and multi-level hybrid stock movement prediction. Specifically, we devise a dynamic graph to learn the changing representations of inter-stock relationships by integrating a pruning module that adapts to market trends. Afterward, with a selective mechanism, the multi-level Mamba discards irrelevant information and resets states to skillfully recall historical patterns across multiple time scales with linear time costs, which are then jointly optimized for reliable prediction. Extensive experiments on U.S. and Chinese stock markets demonstrate the effectiveness of our proposed FinMamba, achieving state-of-the-art prediction accuracy and trading profitability, while maintaining low computational complexity. The code is available at https://github.com/TROUBADOUR000/FinMamba. △ Less

Submitted 10 February, 2025; originally announced February 2025.

Demystifying Catastrophic Forgetting in Two-Stage Incremental Object Detector

Authors: Qirui Wu, Shizhou Zhang, De Cheng, Yinghui Xing, Di Xu, Peng Wang, Yanning Zhang

Abstract: Catastrophic forgetting is a critical chanllenge for incremental object detection (IOD). Most existing methods treat the detector monolithically, relying on instance replay or knowledge distillation without analyzing component-specific forgetting. Through dissection of Faster R-CNN, we reveal a key insight: Catastrophic forgetting is predominantly localized to the RoI Head classifier, while regres… ▽ More Catastrophic forgetting is a critical chanllenge for incremental object detection (IOD). Most existing methods treat the detector monolithically, relying on instance replay or knowledge distillation without analyzing component-specific forgetting. Through dissection of Faster R-CNN, we reveal a key insight: Catastrophic forgetting is predominantly localized to the RoI Head classifier, while regressors retain robustness across incremental stages. This finding challenges conventional assumptions, motivating us to develop a framework termed NSGP-RePRE. Regional Prototype Replay (RePRE) mitigates classifier forgetting via replay of two types of prototypes: coarse prototypes represent class-wise semantic centers of RoI features, while fine-grained prototypes model intra-class variations. Null Space Gradient Projection (NSGP) is further introduced to eliminate prototype-feature misalignment by updating the feature extractor in directions orthogonal to subspace of old inputs via gradient projection, aligning RePRE with incremental learning dynamics. Our simple yet effective design allows NSGP-RePRE to achieve state-of-the-art performance on the Pascal VOC and MS COCO datasets under various settings. Our work not only advances IOD methodology but also provide pivotal insights for catastrophic forgetting mitigation in IOD. Code is available at \href{https://github.com/fanrena/NSGP-RePRE}{https://github.com/fanrena/NSGP-RePRE} . △ Less

Submitted 29 May, 2025; v1 submitted 8 February, 2025; originally announced February 2025.

Comments: Accepted in ICML2025

Efficient Adaptive Label Refinement for Label Noise Learning

Authors: Wenzhen Zhang, Debo Cheng, Guangquan Lu, Bo Zhou, Jiaye Li, Shichao Zhang

Abstract: Deep neural networks are highly susceptible to overfitting noisy labels, which leads to degraded performance. Existing methods address this issue by employing manually defined criteria, aiming to achieve optimal partitioning in each iteration to avoid fitting noisy labels while thoroughly learning clean samples. However, this often results in overly complex and difficult-to-train models. To addres… ▽ More Deep neural networks are highly susceptible to overfitting noisy labels, which leads to degraded performance. Existing methods address this issue by employing manually defined criteria, aiming to achieve optimal partitioning in each iteration to avoid fitting noisy labels while thoroughly learning clean samples. However, this often results in overly complex and difficult-to-train models. To address this issue, we decouple the tasks of avoiding fitting incorrect labels and thoroughly learning clean samples and propose a simple yet highly applicable method called Adaptive Label Refinement (ALR). First, inspired by label refurbishment techniques, we update the original hard labels to soft labels using the model's predictions to reduce the risk of fitting incorrect labels. Then, by introducing the entropy loss, we gradually `harden' the high-confidence soft labels, guiding the model to better learn from clean samples. This approach is simple and efficient, requiring no prior knowledge of noise or auxiliary datasets, making it more accessible compared to existing methods. We validate ALR's effectiveness through experiments on benchmark datasets with artificial label noise (CIFAR-10/100) and real-world datasets with inherent noise (ANIMAL-10N, Clothing1M, WebVision). The results show that ALR outperforms state-of-the-art methods. △ Less

Submitted 1 February, 2025; originally announced February 2025.

TimeFilter: Patch-Specific Spatial-Temporal Graph Filtration for Time Series Forecasting

Authors: Yifan Hu, Guibin Zhang, Peiyuan Liu, Disen Lan, Naiqi Li, Dawei Cheng, Tao Dai, Shu-Tao Xia, Shirui Pan

Abstract: Time series forecasting methods generally fall into two main categories: Channel Independent (CI) and Channel Dependent (CD) strategies. While CI overlooks important covariate relationships, CD captures all dependencies without distinction, introducing noise and reducing generalization. Recent advances in Channel Clustering (CC) aim to refine dependency modeling by grouping channels with similar c… ▽ More Time series forecasting methods generally fall into two main categories: Channel Independent (CI) and Channel Dependent (CD) strategies. While CI overlooks important covariate relationships, CD captures all dependencies without distinction, introducing noise and reducing generalization. Recent advances in Channel Clustering (CC) aim to refine dependency modeling by grouping channels with similar characteristics and applying tailored modeling techniques. However, coarse-grained clustering struggles to capture complex, time-varying interactions effectively. To address these challenges, we propose TimeFilter, a GNN-based framework for adaptive and fine-grained dependency modeling. After constructing the graph from the input sequence, TimeFilter refines the learned spatial-temporal dependencies by filtering out irrelevant correlations while preserving the most critical ones in a patch-specific manner. Extensive experiments on 13 real-world datasets from diverse application domains demonstrate the state-of-the-art performance of TimeFilter. The code is available at https://github.com/TROUBADOUR000/TimeFilter. △ Less

Submitted 20 May, 2025; v1 submitted 22 January, 2025; originally announced January 2025.

Training Data Attribution (TDA): Examining Its Adoption & Use Cases

Authors: Deric Cheng, Juhan Bae, Justin Bullock, David Kristofferson

Abstract: This report investigates Training Data Attribution (TDA) and its potential importance to and tractability for reducing extreme risks from AI. First, we discuss the plausibility and amount of effort it would take to bring existing TDA research efforts from their current state, to an efficient and accurate tool for TDA inference that can be run on frontier-scale LLMs. Next, we discuss the numerous r… ▽ More This report investigates Training Data Attribution (TDA) and its potential importance to and tractability for reducing extreme risks from AI. First, we discuss the plausibility and amount of effort it would take to bring existing TDA research efforts from their current state, to an efficient and accurate tool for TDA inference that can be run on frontier-scale LLMs. Next, we discuss the numerous research benefits AI labs will expect to see from using such TDA tooling. Then, we discuss a key outstanding bottleneck that would limit such TDA tooling from being accessible publicly: AI labs' willingness to disclose their training data. We suggest ways AI labs may work around these limitations, and discuss the willingness of governments to mandate such access. Assuming that AI labs willingly provide access to TDA inference, we then discuss what high-level societal benefits you might see. We list and discuss a series of policies and systems that may be enabled by TDA. Finally, we present an evaluation of TDA's potential impact on mitigating large-scale risks from AI systems. △ Less

Submitted 22 January, 2025; originally announced January 2025.

Text-to-Edit: Controllable End-to-End Video Ad Creation via Multimodal LLMs

Authors: Dabing Cheng, Haosen Zhan, Xingchen Zhao, Guisheng Liu, Zemin Li, Jinghui Xie, Zhao Song, Weiguo Feng, Bingyue Peng

Abstract: The exponential growth of short-video content has ignited a surge in the necessity for efficient, automated solutions to video editing, with challenges arising from the need to understand videos and tailor the editing according to user requirements. Addressing this need, we propose an innovative end-to-end foundational framework, ultimately actualizing precise control over the final video content… ▽ More The exponential growth of short-video content has ignited a surge in the necessity for efficient, automated solutions to video editing, with challenges arising from the need to understand videos and tailor the editing according to user requirements. Addressing this need, we propose an innovative end-to-end foundational framework, ultimately actualizing precise control over the final video content editing. Leveraging the flexibility and generalizability of Multimodal Large Language Models (MLLMs), we defined clear input-output mappings for efficient video creation. To bolster the model's capability in processing and comprehending video content, we introduce a strategic combination of a denser frame rate and a slow-fast processing technique, significantly enhancing the extraction and understanding of both temporal and spatial video information. Furthermore, we introduce a text-to-edit mechanism that allows users to achieve desired video outcomes through textual input, thereby enhancing the quality and controllability of the edited videos. Through comprehensive experimentation, our method has not only showcased significant effectiveness within advertising datasets, but also yields universally applicable conclusions on public datasets. △ Less

Submitted 10 January, 2025; originally announced January 2025.

Comments: 16pages conference

Comparison Study: Glacier Calving Front Delineation in Synthetic Aperture Radar Images With Deep Learning

Authors: Nora Gourmelon, Konrad Heidler, Erik Loebel, Daniel Cheng, Julian Klink, Anda Dong, Fei Wu, Noah Maul, Moritz Koch, Marcel Dreier, Dakota Pyles, Thorsten Seehaus, Matthias Braun, Andreas Maier, Vincent Christlein

Abstract: Calving front position variation of marine-terminating glaciers is an indicator of ice mass loss and a crucial parameter in numerical glacier models. Deep Learning (DL) systems can automatically extract this position from Synthetic Aperture Radar (SAR) imagery, enabling continuous, weather- and illumination-independent, large-scale monitoring. This study presents the first comparison of DL systems… ▽ More Calving front position variation of marine-terminating glaciers is an indicator of ice mass loss and a crucial parameter in numerical glacier models. Deep Learning (DL) systems can automatically extract this position from Synthetic Aperture Radar (SAR) imagery, enabling continuous, weather- and illumination-independent, large-scale monitoring. This study presents the first comparison of DL systems on a common calving front benchmark dataset. A multi-annotator study with ten annotators is performed to contrast the best-performing DL system against human performance. The best DL model's outputs deviate 221 m on average, while the average deviation of the human annotators is 38 m. This significant difference shows that current DL systems do not yet match human performance and that further research is needed to enable fully automated monitoring of glacier calving fronts. The study of Vision Transformers, foundation models, and the inclusion and processing strategy of more information are identified as avenues for future research. △ Less

Submitted 9 January, 2025; originally announced January 2025.

ACM Class: I.4.6; I.5.4

Counterfactual Samples Constructing and Training for Commonsense Statements Estimation

Authors: Chong Liu, Zaiwen Feng, Lin Liu, Zhenyun Deng, Jiuyong Li, Ruifang Zhai, Debo Cheng, Li Qin

Abstract: Plausibility Estimation (PE) plays a crucial role for enabling language models to objectively comprehend the real world. While large language models (LLMs) demonstrate remarkable capabilities in PE tasks but sometimes produce trivial commonsense errors due to the complexity of commonsense knowledge. They lack two key traits of an ideal PE model: a) Language-explainable: relying on critical word se… ▽ More Plausibility Estimation (PE) plays a crucial role for enabling language models to objectively comprehend the real world. While large language models (LLMs) demonstrate remarkable capabilities in PE tasks but sometimes produce trivial commonsense errors due to the complexity of commonsense knowledge. They lack two key traits of an ideal PE model: a) Language-explainable: relying on critical word segments for decisions, and b) Commonsense-sensitive: detecting subtle linguistic variations in commonsense. To address these issues, we propose a novel model-agnostic method, referred to as Commonsense Counterfactual Samples Generating (CCSG). By training PE models with CCSG, we encourage them to focus on critical words, thereby enhancing both their language-explainable and commonsense-sensitive capabilities. Specifically, CCSG generates counterfactual samples by strategically replacing key words and introducing low-level dropout within sentences. These counterfactual samples are then incorporated into a sentence-level contrastive training framework to further enhance the model's learning process. Experimental results across nine diverse datasets demonstrate the effectiveness of CCSG in addressing commonsense reasoning challenges, with our CCSG method showing 3.07% improvement against the SOTA methods. △ Less

Submitted 29 December, 2024; originally announced December 2024.

Comments: 14 pages, 4 figures

Hybrid Local Causal Discovery

Authors: Zhaolong Ling, Honghui Peng, Yiwen Zhang, Debo Cheng, Xingyu Wu, Peng Zhou, Kui Yu

Abstract: Local causal discovery aims to learn and distinguish the direct causes and effects of a target variable from observed data. Existing constraint-based local causal discovery methods use AND or OR rules in constructing the local causal skeleton, but using either rule alone is prone to produce cascading errors in the learned local causal skeleton, and thus impacting the inference of local causal rela… ▽ More Local causal discovery aims to learn and distinguish the direct causes and effects of a target variable from observed data. Existing constraint-based local causal discovery methods use AND or OR rules in constructing the local causal skeleton, but using either rule alone is prone to produce cascading errors in the learned local causal skeleton, and thus impacting the inference of local causal relationships. On the other hand, directly applying score-based global causal discovery methods to local causal discovery may randomly return incorrect results due to the existence of local equivalence classes. To address the above issues, we propose a Hybrid Local Causal Discovery algorithm, called HLCD. Specifically, HLCD initially utilizes a constraint-based approach combined with the OR rule to obtain a candidate skeleton and then employs a score-based method to eliminate redundant portions in the candidate skeleton. Furthermore, during the local causal orientation phase, HLCD distinguishes between V-structures and equivalence classes by comparing the local structure scores between the two, thereby avoiding orientation interference caused by local equivalence classes. We conducted extensive experiments with seven state-of-the-art competitors on 14 benchmark Bayesian network datasets, and the experimental results demonstrate that HLCD significantly outperforms existing local causal discovery algorithms. △ Less

Submitted 12 May, 2025; v1 submitted 27 December, 2024; originally announced December 2024.

Comments: This paper has been accepted for publication in the Proceedings of the 34th International Joint Conference on Artificial Intelligence (IJCAI 2025)

Semi-supervised Credit Card Fraud Detection via Attribute-Driven Graph Representation

Authors: Sheng Xiang, Mingzhi Zhu, Dawei Cheng, Enxia Li, Ruihui Zhao, Yi Ouyang, Ling Chen, Yefeng Zheng

Abstract: Credit card fraud incurs a considerable cost for both cardholders and issuing banks. Contemporary methods apply machine learning-based classifiers to detect fraudulent behavior from labeled transaction records. But labeled data are usually a small proportion of billions of real transactions due to expensive labeling costs, which implies that they do not well exploit many natural features from unla… ▽ More Credit card fraud incurs a considerable cost for both cardholders and issuing banks. Contemporary methods apply machine learning-based classifiers to detect fraudulent behavior from labeled transaction records. But labeled data are usually a small proportion of billions of real transactions due to expensive labeling costs, which implies that they do not well exploit many natural features from unlabeled data. Therefore, we propose a semi-supervised graph neural network for fraud detection. Specifically, we leverage transaction records to construct a temporal transaction graph, which is composed of temporal transactions (nodes) and interactions (edges) among them. Then we pass messages among the nodes through a Gated Temporal Attention Network (GTAN) to learn the transaction representation. We further model the fraud patterns through risk propagation among transactions. The extensive experiments are conducted on a real-world transaction dataset and two publicly available fraud detection datasets. The result shows that our proposed method, namely GTAN, outperforms other state-of-the-art baselines on three fraud detection datasets. Semi-supervised experiments demonstrate the excellent fraud detection performance of our model with only a tiny proportion of labeled data. △ Less

Submitted 24 December, 2024; originally announced December 2024.

Comments: 9 pages, 5 figures, AAAI 2023, code: https://github.com/AI4Risk/antifraud

Journal ref: Proceedings of the AAAI Conference on Artificial Intelligence. Vol. 37. No. 12. 2023

Attack by Yourself: Effective and Unnoticeable Multi-Category Graph Backdoor Attacks with Subgraph Triggers Pool

Authors: Jiangtong Li, Dungy Liu, Dawei Cheng, Changchun Jiang

Abstract: \textbf{G}raph \textbf{N}eural \textbf{N}etworks~(GNNs) have achieved significant success in various real-world applications, including social networks, finance systems, and traffic management. Recent researches highlight their vulnerability to backdoor attacks in node classification, where GNNs trained on a poisoned graph misclassify a test node only when specific triggers are attached. These stu… ▽ More \textbf{G}raph \textbf{N}eural \textbf{N}etworks~(GNNs) have achieved significant success in various real-world applications, including social networks, finance systems, and traffic management. Recent researches highlight their vulnerability to backdoor attacks in node classification, where GNNs trained on a poisoned graph misclassify a test node only when specific triggers are attached. These studies typically focus on single attack categories and use adaptive trigger generators to create node-specific triggers. However, adaptive trigger generators typically have a simple structure, limited parameters, and lack category-aware graph knowledge, which makes them struggle to handle backdoor attacks across multiple categories as the number of target categories increases. We address this gap by proposing a novel approach for \textbf{E}ffective and \textbf{U}nnoticeable \textbf{M}ulti-\textbf{C}ategory~(EUMC) graph backdoor attacks, leveraging subgraph from the attacked graph as category-aware triggers to precisely control the target category. To ensure the effectiveness of our method, we construct a \textbf{M}ulti-\textbf{C}ategory \textbf{S}ubgraph \textbf{T}riggers \textbf{P}ool~(MC-STP) using the subgraphs of the attacked graph as triggers. We then exploit the attachment probability shifts of each subgraph trigger as category-aware priors for target category determination. Moreover, we develop a ``select then attach'' strategy that connects suitable category-aware trigger to attacked nodes for unnoticeability. Extensive experiments across different real-world datasets confirm the efficacy of our method in conducting multi-category graph backdoor attacks on various GNN models and defense strategies. △ Less

Submitted 22 December, 2024; originally announced December 2024.

Comments: 13 pages, 5 figures

How to Synthesize Text Data without Model Collapse?

Authors: Xuekai Zhu, Daixuan Cheng, Hengli Li, Kaiyan Zhang, Ermo Hua, Xingtai Lv, Ning Ding, Zhouhan Lin, Zilong Zheng, Bowen Zhou

Abstract: Model collapse in synthetic data indicates that iterative training on self-generated data leads to a gradual decline in performance. With the proliferation of AI models, synthetic data will fundamentally reshape the web data ecosystem. Future GPT-$\{n\}$ models will inevitably be trained on a blend of synthetic and human-produced data. In this paper, we focus on two questions: what is the impact o… ▽ More Model collapse in synthetic data indicates that iterative training on self-generated data leads to a gradual decline in performance. With the proliferation of AI models, synthetic data will fundamentally reshape the web data ecosystem. Future GPT-$\{n\}$ models will inevitably be trained on a blend of synthetic and human-produced data. In this paper, we focus on two questions: what is the impact of synthetic data on language model training, and how to synthesize data without model collapse? We first pre-train language models across different proportions of synthetic data, revealing a negative correlation between the proportion of synthetic data and model performance. We further conduct statistical analysis on synthetic data to uncover distributional shift phenomenon and over-concentration of n-gram features. Inspired by the above findings, we propose token editing on human-produced data to obtain semi-synthetic data. As a proof of concept, we theoretically demonstrate that token-level editing can prevent model collapse, as the test error is constrained by a finite upper bound. We conduct extensive experiments on pre-training from scratch, continual pre-training, and supervised fine-tuning. The results validate our theoretical proof that token-level editing improves model performance. △ Less

Submitted 28 May, 2025; v1 submitted 19 December, 2024; originally announced December 2024.

Comments: Accepted at ICML 2025

Efficient Dynamic Attributed Graph Generation

Authors: Fan Li, Xiaoyang Wang, Dawei Cheng, Cong Chen, Ying Zhang, Xuemin Lin

Abstract: Data generation is a fundamental research problem in data management due to its diverse use cases, ranging from testing database engines to data-specific applications. However, real-world entities often involve complex interactions that cannot be effectively modeled by traditional tabular data. Therefore, graph data generation has attracted increasing attention recently. Although various graph gen… ▽ More Data generation is a fundamental research problem in data management due to its diverse use cases, ranging from testing database engines to data-specific applications. However, real-world entities often involve complex interactions that cannot be effectively modeled by traditional tabular data. Therefore, graph data generation has attracted increasing attention recently. Although various graph generators have been proposed in the literature, there are three limitations: i) They cannot capture the co-evolution pattern of graph structure and node attributes. ii) Few of them consider edge direction, leading to substantial information loss. iii) Current state-of-the-art dynamic graph generators are based on the temporal random walk, making the simulation process time-consuming. To fill the research gap, we introduce VRDAG, a novel variational recurrent framework for efficient dynamic attributed graph generation. Specifically, we design a bidirectional message-passing mechanism to encode both directed structural knowledge and attribute information of a snapshot. Then, the temporal dependency in the graph sequence is captured by a recurrence state updater, generating embeddings that can preserve the evolution pattern of early graphs. Based on the hidden node embeddings, a conditional variational Bayesian method is developed to sample latent random variables at the neighboring timestep for new snapshot generation. The proposed generation paradigm avoids the time-consuming path sampling and merging process in existing random walk-based methods, significantly reducing the synthesis time. Finally, comprehensive experiments on real-world datasets are conducted to demonstrate the effectiveness and efficiency of the proposed model. △ Less

Submitted 11 December, 2024; originally announced December 2024.

Comments: 14 pages,10 figures. Accepted by IEEE ICDE2025

Fast Track to Winning Tickets: Repowering One-Shot Pruning for Graph Neural Networks

Authors: Yanwei Yue, Guibin Zhang, Haoran Yang, Dawei Cheng

Abstract: Graph Neural Networks (GNNs) demonstrate superior performance in various graph learning tasks, yet their wider real-world application is hindered by the computational overhead when applied to large-scale graphs. To address the issue, the Graph Lottery Hypothesis (GLT) has been proposed, advocating the identification of subgraphs and subnetworks, \textit{i.e.}, winning tickets, without compromising… ▽ More Graph Neural Networks (GNNs) demonstrate superior performance in various graph learning tasks, yet their wider real-world application is hindered by the computational overhead when applied to large-scale graphs. To address the issue, the Graph Lottery Hypothesis (GLT) has been proposed, advocating the identification of subgraphs and subnetworks, \textit{i.e.}, winning tickets, without compromising performance. The effectiveness of current GLT methods largely stems from the use of iterative magnitude pruning (IMP), which offers higher stability and better performance than one-shot pruning. However, identifying GLTs is highly computationally expensive, due to the iterative pruning and retraining required by IMP. In this paper, we reevaluate the correlation between one-shot pruning and IMP: while one-shot tickets are suboptimal compared to IMP, they offer a \textit{fast track} to tickets with a stronger performance. We introduce a one-shot pruning and denoising framework to validate the efficacy of the \textit{fast track}. Compared to current IMP-based GLT methods, our framework achieves a double-win situation of graph lottery tickets with \textbf{higher sparsity} and \textbf{faster speeds}. Through extensive experiments across 4 backbones and 6 datasets, our method demonstrates $1.32\% - 45.62\%$ improvement in weight sparsity and a $7.49\% - 22.71\%$ increase in graph sparsity, along with a $1.7-44 \times$ speedup over IMP-based methods and $95.3\%-98.6\%$ MAC savings. △ Less

Submitted 10 December, 2024; originally announced December 2024.

Comments: AAAI 2025

Compression for Better: A General and Stable Lossless Compression Framework

Authors: Boyang Zhang, Daning Cheng, Yunquan Zhang, Fangmin Liu, Wenguang Chen

Abstract: This work focus on how to stabilize and lossless model compression, aiming to reduce model complexity and enhance efficiency without sacrificing performance due to compression errors. A key challenge is effectively leveraging compression errors and defining the boundaries for lossless compression to minimize model loss. i.e., compression for better. Currently, there is no systematic approach to de… ▽ More This work focus on how to stabilize and lossless model compression, aiming to reduce model complexity and enhance efficiency without sacrificing performance due to compression errors. A key challenge is effectively leveraging compression errors and defining the boundaries for lossless compression to minimize model loss. i.e., compression for better. Currently, there is no systematic approach to determining this error boundary or understanding its specific impact on model performance. We propose a general \textbf{L}oss\textbf{L}ess \textbf{C}ompression theoretical framework (\textbf{LLC}), which further delineates the compression neighborhood and higher-order analysis boundaries through the total differential, thereby specifying the error range within which a model can be compressed without loss. To verify the effectiveness of LLC, we apply various compression techniques, including quantization and decomposition. Specifically, for quantization, we reformulate the classic quantization search problem as a grouped knapsack problem within the lossless neighborhood, achieving lossless quantization while improving computational efficiency. For decomposition, LLC addresses the approximation problem under low-rank constraints, automatically determining the rank for each layer and producing lossless low-rank models. We conduct extensive experiments on multiple neural network architectures on different datasets. The results show that without fancy tricks, LLC can effectively achieve lossless model compression. Our code will be made publicly. △ Less

Submitted 9 December, 2024; originally announced December 2024.

Comments: Under Review

Lossless Model Compression via Joint Low-Rank Factorization Optimization

Authors: Boyang Zhang, Daning Cheng, Yunquan Zhang, Fangmin Liu, Jiake Tian

Abstract: Low-rank factorization is a popular model compression technique that minimizes the error $δ$ between approximated and original weight matrices. Despite achieving performances close to the original models when $δ$ is optimized, a performance discrepancy remains due to the separate optimization processes for low-rank factorization and model performance, resulting in unavoidable losses. We address th… ▽ More Low-rank factorization is a popular model compression technique that minimizes the error $δ$ between approximated and original weight matrices. Despite achieving performances close to the original models when $δ$ is optimized, a performance discrepancy remains due to the separate optimization processes for low-rank factorization and model performance, resulting in unavoidable losses. We address this issue by introducing a novel joint optimization strategy for lossless low-rank weight factorization, which, for the first time, enhances the model's performance beyond the original. Our approach begins with a theoretical analysis of the relationship between low-rank factorization and model optimization objectives, establishing a precise perturbation range for matrix factorization errors on model performance. This challenge is then reformulated as a numerical rank deficiency problem with inequality constraints and develop a joint objective that simultaneously addresses factorization error and model performance. Based on the above analysis, we propose two optimization algorithms: \textbf{a lossless optimization algorithm} that maximizes model accuracy while ensuring compression, and \textbf{a compact optimization algorithm} that minimizes model size while preserving performance. These algorithms do not require fine-tuning and can directly compress numerous deep models to achieve lossless results. Our methods demonstrate robust efficacy across various vision and language tasks. For example, the compressed model reduced by 70\% on ResNext50 outperforms the original. Our code will be made public. △ Less

Submitted 9 December, 2024; originally announced December 2024.

Comments: Under Review

FP=xINT:A Low-Bit Series Expansion Algorithm for Post-Training Quantization

Authors: Boyang Zhang, Daning Cheng, Yunquan Zhang, Fangmin Liu

Abstract: Post-Training Quantization (PTQ) converts pre-trained Full-Precision (FP) models into quantized versions without training. While existing methods reduce size and computational costs, they also significantly degrade performance and quantization efficiency at extremely low settings due to quantization noise. We introduce a deep model series expansion framework to address this issue, enabling rapid a… ▽ More Post-Training Quantization (PTQ) converts pre-trained Full-Precision (FP) models into quantized versions without training. While existing methods reduce size and computational costs, they also significantly degrade performance and quantization efficiency at extremely low settings due to quantization noise. We introduce a deep model series expansion framework to address this issue, enabling rapid and accurate approximation of unquantized models without calibration sets or fine-tuning. This is the first use of series expansion for neural network quantization. Specifically, our method expands the FP model into multiple low-bit basis models. To ensure accurate quantization, we develop low-bit basis model expansions at different granularities (tensor, layer, model), and theoretically confirm their convergence to the dense model, thus restoring FP model accuracy. Additionally, we design AbelianAdd/Mul operations between isomorphic models in the low-bit expansion, forming an Abelian group to ensure operation parallelism and commutativity. The experiments show that our algorithm achieves state-of-the-art performance in low-bit settings; for example, 4-bit quantization of ResNet-50 surpasses the original accuracy, reaching 77.03%. The code will be made public. △ Less

Submitted 9 December, 2024; originally announced December 2024.

Comments: Under Review

Evaluating Hallucination in Text-to-Image Diffusion Models with Scene-Graph based Question-Answering Agent

Authors: Ziyuan Qin, Dongjie Cheng, Haoyu Wang, Huahui Yi, Yuting Shao, Zhiyuan Fan, Kang Li, Qicheng Lao

Abstract: Contemporary Text-to-Image (T2I) models frequently depend on qualitative human evaluations to assess the consistency between synthesized images and the text prompts. There is a demand for quantitative and automatic evaluation tools, given that human evaluation lacks reproducibility. We believe that an effective T2I evaluation metric should accomplish the following: detect instances where the gener… ▽ More Contemporary Text-to-Image (T2I) models frequently depend on qualitative human evaluations to assess the consistency between synthesized images and the text prompts. There is a demand for quantitative and automatic evaluation tools, given that human evaluation lacks reproducibility. We believe that an effective T2I evaluation metric should accomplish the following: detect instances where the generated images do not align with the textual prompts, a discrepancy we define as the `hallucination problem' in T2I tasks; record the types and frequency of hallucination issues, aiding users in understanding the causes of errors; and provide a comprehensive and intuitive scoring that close to human standard. To achieve these objectives, we propose a method based on large language models (LLMs) for conducting question-answering with an extracted scene-graph and created a dataset with human-rated scores for generated images. From the methodology perspective, we combine knowledge-enhanced question-answering tasks with image evaluation tasks, making the evaluation metrics more controllable and easier to interpret. For the contribution on the dataset side, we generated 12,000 synthesized images based on 1,000 composited prompts using three advanced T2I models. Subsequently, we conduct human scoring on all synthesized images and prompt pairs to validate the accuracy and effectiveness of our method as an evaluation metric. All generated images and the human-labeled scores will be made publicly available in the future to facilitate ongoing research on this crucial issue. Extensive experiments show that our method aligns more closely with human scoring patterns than other evaluation metrics. △ Less

Submitted 7 December, 2024; originally announced December 2024.