Meta-Learning Bidirectional Update Rules
This work proposes a new paradigm for neural network training that could impact machine learning by enabling gradient-free optimization, though it is still in early stages with domain-specific applications.
The paper tackles the problem of gradient-based backpropagation in neural networks by introducing a generalized framework with bidirectional Hebb-style update rules parameterized by a meta-learned genome, resulting in faster training on unseen tasks compared to gradient descent optimizers for computer vision and synthetic tasks.
In this paper, we introduce a new type of generalized neural network where neurons and synapses maintain multiple states. We show that classical gradient-based backpropagation in neural networks can be seen as a special case of a two-state network where one state is used for activations and another for gradients, with update rules derived from the chain rule. In our generalized framework, networks have neither explicit notion of nor ever receive gradients. The synapses and neurons are updated using a bidirectional Hebb-style update rule parameterized by a shared low-dimensional "genome". We show that such genomes can be meta-learned from scratch, using either conventional optimization techniques, or evolutionary strategies, such as CMA-ES. Resulting update rules generalize to unseen tasks and train faster than gradient descent based optimizers for several standard computer vision and synthetic tasks.