Memory-Based Meta-Learning on Non-Stationary Distributions
This addresses the challenge of non-stationary distributions in natural language and partially observable environments, but it is incremental as it applies existing models to a specific scenario.
The paper tackled the problem of approximating Bayes-optimal predictors for piecewise stationary distributions with unobserved switching-points, showing that memory-based neural models like Transformers, LSTMs, and RNNs can accurately approximate known Bayes-optimal algorithms and behave as if performing Bayesian inference.
Memory-based meta-learning is a technique for approximating Bayes-optimal predictors. Under fairly general conditions, minimizing sequential prediction error, measured by the log loss, leads to implicit meta-learning. The goal of this work is to investigate how far this interpretation can be realized by current sequence prediction models and training regimes. The focus is on piecewise stationary sources with unobserved switching-points, which arguably capture an important characteristic of natural language and action-observation sequences in partially observable environments. We show that various types of memory-based neural models, including Transformers, LSTMs, and RNNs can learn to accurately approximate known Bayes-optimal algorithms and behave as if performing Bayesian inference over the latent switching-points and the latent parameters governing the data distribution within each segment.