rat-rnn
Simulate Rat neural activity in the maze
Attempt to re-produce results from ICLR2018: EMERGENCE OF GRID-LIKE REPRESENTATIONS BY TRAINING RECURRENT NEURAL NETWORKS TO PERFORM SPATIAL LOCALIZATION paper.
Results so far. Training with LSTM or CTRNN gives comparable results.
Below are LSTM results after training for 10000 iterations on 1 millions sampless:
- Rat did learn spatial localization (dead reckoning) task well
- Rat performs well under significant noise injected into the network
- However NO grid cells formed in the process :(
