Continual learning agents experience a stream of (related) tasks.

Reinforcement learning (RL) promises to enable autonomous acquisition of complex behaviors for diverse agents.

That revelation is anticipated, for if the characteristics of previous data are remembered perfectly ( e.g., via realistic generative replay), no forgetting should be expected for lifelong learning.

By "well-behaved" we mean the shape within kernels (see L175; shared between source

This should be emphasised similar to the argument in the introduction.

Summary and Contributions: This manuscript addresses the problem of continual learning in RNN. The authors propose a new learning rule that allows to organize the dynamics for different tasks into orthogonal subspaces. Using a set of neuroscience tasks, they show how this learning rule allows to avoid catastrophic interferences between tasks. By analyzing the dynamics of trained networks they provide evidence for why their learning rule is successful, it also allows them to discuss the problem of transfer learning. Strengths: - propose a new original solution to the problem of continual learning, which also allows them to address and understand under which conditions learning in one task can be transfered to learning off another task.

Is "Six Seven" Really Brain Rot? The viral phrase is easy to dismiss, but its ubiquity suggests something crucial about human nature. Recently, my wife was texting with a friend who lives in Singapore. The news from the other side of the world turned out to be that kids there had discovered "six seven." On Halloween, our friend reported, a boy with a handmade "six seven" jersey had earned applause as he made his way through her neighborhood--a place that's a long way from Sixty-seventh Street in Philadelphia, which the rapper Skrilla may have been referencing in his song "Doot Doot (6 7)," which came out last December.

Reinforcement learning (RL) has emerged as a pivotal technique for fine-tuning large language models (LLMs) on specific tasks. However, prevailing RL fine-tuning methods predominantly rely on PPO and its variants. Though these algorithms are effective in general RL settings, they often exhibit suboptimal performance and vulnerability to distribution collapse when applied to the fine-tuning of LLMs.

This work was done when the second author was jointly with Google Research.

However, such "black-box" policies lack interpretability and reliability, and often,