Project

Recursive Reasoning with Tiny Networks (TRM in Keras)

My own implementation of the TRM paper “Less is More: Recursive Reasoning with Tiny Networks” in Keras.

Context

The paper “Less is More: Recursive Reasoning with Tiny Networks” shows that a very small network can solve hard reasoning tasks by applying the same layers repeatedly. This is exactly the direction that interests me: fewer parameters, more recursion.

Contribution

I rebuilt the architecture from the paper myself in Keras to understand the structure of the recursive loop and the training in detail.

Approach and result

Measurements and comparison figures are not yet cleared for this page and are therefore not stated. The code is not public yet.

Back to all projects