Experiments

[stfwn]

Stuff we can choose from

Datasets:

• TinyImageNet200
• Parity

Models:

• PonderNet, pondering.
• PonderNet, set number of ponder steps (no halting parameter).
• PonderNet, not pondering (stop after first step)
• Not pondering pondernet that takes the same time to pass fwd through as the pondering PonderNet

Step functions:

• MLP
• RNN: LSTM

Encoders:

• No encoder
• EfficientNetb0

Experiments we're gonna run

Dataset	Model	Encoder	Step function
TinyImageNet 200	PonderNet, pondering	EfficientNetb0	RNN: LSTM
TinyImageNet 200	PonderNet, not pondering	EfficientNetb0	RNN: LSTM