Self-Evolving Agents via Likelihood-Guided Tool-Space Optimization

LOTS uses observed tool outputs to learn reusable, task-specific tool selections that improve agent performance while reducing tool context.

LatentGym: A Testbed For Cross-Task Experiential Learning With Controllable Latent Structure

LatentGym uses environments with known latent structure to measure how LLM agents gather information and adapt across related tasks.

A Closer Look at Personalized Fine-Tuning in Heterogeneous Federated Learning

A study of federated feature distortion shows how linear probing followed by fine-tuning can improve the balance between client personalization and generalization.