1. collect training set data
   • 3 circle centers + variation
2. collect net params (optional)
   • hidden layer count
   • hidden layer sizes
   • learning rate
   • momentum
   • bias on/off
   • (?) propagation function
   • (?) initial weights
3. Learning
   • train 1 step
   • classify all input points
   • return to draw on screen
4. Testing
   • add 1 point and get it classified

Allow resetting

…

POST /simulation/
Send simulation params and initialize the neural network. Input: training set, network params Output: redirect to /step/1 (?) State: remember training set, create network, 1st eval, draw

GET /step/{number}
Trains the neural network 1 time with next point [chosen?]. Sends back the points after classification by current network.