Learn vanilla AE

[vlad17]

(Try to find existing imple if possible)

autoencoders reading

diff prediction

[vlad17]

On top of vanilla/regularized AE consider using a CAE (might need jacobian code as performed here, or use a trace of the jacobian (?) .

DAE - dae might be useful, considering its relation to ladder nets.

Use HW3 conv net for the lower stack (also collect training from hw3 random agent).

[vlad17]

• [x] setup repo #7 #8
• [x] slow transition dataset-gathering implementation #10
• [x] fast transition dataset-gathering implementation (vectorized env) #12 #13
• [ ] vanilla AE (w/ sparsity regularizer)