A Probabilistic Framework for Modular Continual Learning
This addresses the problem of efficient module composition search in continual learning for AI systems, representing a novel method for a known bottleneck.
The paper tackles the challenge of searching through large discrete spaces of module compositions in modular continual learning by introducing PICLE, a probabilistic framework that cheaply evaluates compositions, enabling perceptual, few-shot, and latent transfer. It outperforms previous state-of-the-art modular approaches on long problem sequences.
Modular approaches that use a different composition of modules for each problem are a promising direction in continual learning (CL). However, searching through the large, discrete space of module compositions is challenging, especially because evaluating a composition's performance requires a round of neural network training. We address this challenge through a modular CL framework, PICLE, that uses a probabilistic model to cheaply compute the fitness of each composition, allowing PICLE to achieve both perceptual, few-shot and latent transfer. The model combines prior knowledge about good module compositions with dataset-specific information. We evaluate PICLE using two benchmark suites designed to assess different desiderata of CL techniques. Comparing to a wide range of approaches, we show that PICLE is the first modular CL algorithm to achieve perceptual, few-shot and latent transfer while scaling well to large search spaces, outperforming previous state-of-the-art modular CL approaches on long problem sequences.