Differentiable plasticity: training plastic neural networks with backpropagation
This work addresses the learning-to-learn problem for AI agents seeking efficient lifelong learning, offering a novel approach inspired by biological brains.
The authors tackled the problem of enabling agents to learn efficiently from new experiences after initial training by optimizing synaptic plasticity with gradient descent in large recurrent networks. They demonstrated that these plastic networks can memorize and reconstruct novel high-dimensional images, solve meta-learning tasks competitively, and outperform non-plastic networks in reinforcement learning settings, with results on tasks like Omniglot and maze exploration.
How can we build agents that keep learning from experience, quickly and efficiently, after their initial training? Here we take inspiration from the main mechanism of learning in biological brains: synaptic plasticity, carefully tuned by evolution to produce efficient lifelong learning. We show that plasticity, just like connection weights, can be optimized by gradient descent in large (millions of parameters) recurrent networks with Hebbian plastic connections. First, recurrent plastic networks with more than two million parameters can be trained to memorize and reconstruct sets of novel, high-dimensional 1000+ pixels natural images not seen during training. Crucially, traditional non-plastic recurrent networks fail to solve this task. Furthermore, trained plastic networks can also solve generic meta-learning tasks such as the Omniglot task, with competitive results and little parameter overhead. Finally, in reinforcement learning settings, plastic networks outperform a non-plastic equivalent in a maze exploration task. We conclude that differentiable plasticity may provide a powerful novel approach to the learning-to-learn problem.