Reinforcement learning, the driving force behind AI advancements, has been supercharged by a blend of backpropagation (BP) and Equilibrium Propagation (EP). BP has a long history of training neural networks, mastering image classification, speech recognition, and even reinforcement learning tasks. However, doubts loom over its biological plausibility as a neural learning mechanism. Enter EP, a more biologically realistic alternative that nails the CIFAR-10 image classification challenge. This game-changing study presents an EP-based reinforcement learning framework called Actor-Critic, where the actor network is trained using EP. Mind you, this model effortlessly tackles standard control tasks used to benchmark BP-driven models. What’s remarkable is that our trained EP model exhibits consistently high-reward behavior compared to a similar model trained exclusively with BP. It’s like combining the best traits of your favorite superhero and topping it with a cherry on top! Dive into the fascinating realm of this cutting-edge research to see how Reinforcement Learning can reach new heights.

Backpropagation (BP) has been used to train neural networks for many years, allowing them to solve a wide variety of tasks like image classification, speech recognition, and reinforcement learning tasks. But the biological plausibility of BP as a mechanism of neural learning has been questioned. Equilibrium Propagation (EP) has been proposed as a more biologically plausible alternative and achieves comparable accuracy on the CIFAR-10 image classification task. This study proposes the first EP-based reinforcement learning architecture: an Actor-Critic architecture with the actor network trained by EP. We show that this model can solve the basic control tasks often used as benchmarks for BP-based models. Interestingly, our trained model demonstrates more consistent high-reward behavior than a comparable model trained exclusively by BP.

Read Full Article (External Site)