Thinking While Listening: Fast-Slow Recurrence for Long-Horizon Sequential Modeling
This addresses the problem of maintaining coherent representations over long sequences for applications in reinforcement learning and algorithmic tasks, representing an incremental improvement over existing methods.
The paper tackles long-horizon sequential modeling by extending latent recurrent modeling with fast-slow recurrence, which interleaves fast latent updates with slow observation updates to learn stable internal structures. This approach improves out-of-distribution generalization in reinforcement learning and algorithmic tasks compared to baselines like LSTM, state space models, and Transformers.
We extend the recent latent recurrent modeling to sequential input streams. By interleaving fast, recurrent latent updates with self-organizational ability between slow observation updates, our method facilitates the learning of stable internal structures that evolve alongside the input. This mechanism allows the model to maintain coherent and clustered representations over long horizons, improving out-of-distribution generalization in reinforcement learning and algorithmic tasks compared to sequential baselines such as LSTM, state space models, and Transformer variants.