Low-Variance Forward Gradients using Direct Feedback Alignment and Momentum
This work addresses the problem of developing efficient, scalable learning algorithms for neuromorphic hardware, representing an incremental improvement over existing forward gradient techniques.
The paper tackles the high variance issue in forward-mode automatic differentiation methods for deep neural networks, which limits their scalability and neuromorphic hardware compatibility, by proposing the Forward Direct Feedback Alignment algorithm that combines Activity-Perturbed Forward Gradients with Direct Feedback Alignment and momentum, achieving lower variance, faster convergence, and better performance compared to other local alternatives to backpropagation.
Supervised learning in deep neural networks is commonly performed using error backpropagation. However, the sequential propagation of errors during the backward pass limits its scalability and applicability to low-powered neuromorphic hardware. Therefore, there is growing interest in finding local alternatives to backpropagation. Recently proposed methods based on forward-mode automatic differentiation suffer from high variance in large deep neural networks, which affects convergence. In this paper, we propose the Forward Direct Feedback Alignment algorithm that combines Activity-Perturbed Forward Gradients with Direct Feedback Alignment and momentum. We provide both theoretical proofs and empirical evidence that our proposed method achieves lower variance than forward gradient techniques. In this way, our approach enables faster convergence and better performance when compared to other local alternatives to backpropagation and opens a new perspective for the development of online learning algorithms compatible with neuromorphic systems.