IMITATION LEARNING USING THE FORWARD-FORWARD ALGORITHM

Insik Chung, Isaac Han, Kyung-Joong Kim

Keywords: Imitation Learning, Forward-forward algorithm

TL;DR: We evaluate the FF algorithm in the context of imitation learning

Abstract: The forward-forward (FF) algorithm has been recently introduced as a novel approach to training neural networks in a way that approximates the behavior of real neurons. Nevertheless, its application has been limited to visual domains and has not been investigated in the context of imitation learning or reinforcement learning. In this study, we evaluate the FF algorithm in the context of imitation learning. We implement a straightforward imitation model based on the FF algorithm and present a comparative analysis with the backpropagation (BP) model. Our findings indicate that the FF-based model exhibits comparable performance to the BP-based model on larger datasets but demonstrates inferior performance on smaller datasets.