Artificial Intelligence
6 Papers
From microstructure to performance optimization: Innovative applications of computer vision in materials science
This review article explores how rapid progress in computer vision (CV), especially deep learning (DL)-based methods, has revolutionized materials science by shifting from traditional experimental approaches to more efficient, automated analysis of material microstructures.
Multi-agent cooperation through in-context co-player inference
This paper addresses a core challenge in multi-agent reinforcement learning (MARL): enabling robust cooperation among self-interested agents without relying on hand-coded coordination, explicit opponent models, or artificial separations like naive learners vs. meta-learners.
Uni-Hema: Unified Model for Digital Hematopathology
This groundbreaking paper introduces Uni-Hema, the first unified, multi-task, multi-modal vision-language model specifically designed for digital hematopathology. It addresses the fragmentation in current AI approaches by enabling a single model to handle multiple tasks such as object detection (locating individual …
The Assistant Axis: Situating and Stabilizing the Default Persona of Language Models
Large language models (LLMs) are post-trained to adopt a default "helpful Assistant" persona, but this identity is fragile. The paper explores the internal "persona space" in model activations, discovering a dominant linear direction called the Assistant Axis the primary axis …
Training AI Co-Scientists Using Rubric Rewards
AI "co-scientists" (LLM-based assistants) can help researchers by generating detailed research plans from given goals and constraints. However, current models often produce plans that violate implicit requirements due to the open-ended nature of scientific planning and the lack of fast, …
Toward Training Superintelligent Software Agents through Self-Play SWE-RL
Current LLM-based software engineering agents rely heavily on human-curated data (e.g., GitHub issues, pull requests) and environments (e.g., test suites), which limits their path to superintelligence. The paper introduces Self-play SWE-RL (SSR), a reinforcement learning framework that trains a single …