Demystifying Scientific Problem-Solving in LLMs by Probing Knowledge and Reasoning

Scientific problem solving poses unique challenges for LLMs, requiring both deep domain knowledge and the ability to apply such knowledge through complex reasoning. While automated scientific reasoners hold great promise for assisting human scientists, there is currently no widely adopted holistic benchmark for evaluating scientific reasoning, and few approaches systematically disentangle the distinct roles of knowledge and reasoning in these tasks. To address these gaps, we introduce SciReas, a diverse suite of existing benchmarks for scientific reasoning tasks, and SciReas-Pro, a selective subset that requires more complex reasoning. Our holistic evaluation surfaces insights about scientific reasoning performance that remain hidden when relying on individual benchmarks alone. We then propose KRUX, a probing framework for studying the distinct roles of reasoning and knowledge in scientific tasks. Combining the two, we conduct an in-depth analysis that yields several key findings: (1) Retrieving task-relevant knowledge from model parameters is a critical bottleneck for LLMs in scientific reasoning; (2) Reasoning models consistently benefit from external knowledge added in-context on top of the reasoning enhancement; (3) Enhancing verbalized reasoning improves LLMs' ability to surface task-relevant knowledge. Finally, we conduct a lightweight analysis, comparing our science-focused data composition with concurrent efforts on long CoT SFT, and release SciLit01, a strong 8B baseline for scientific reasoning.

YaleNLP @ PerAnsSumm 2025: Multi-Perspective Integration via Mixture-of-Agents for Enhanced Healthcare QA Summarization

Automated summarization of healthcare community question-answering forums is challenging due to diverse perspectives presented across multiple user responses to each question. The PerAnsSumm Shared Task was therefore proposed to tackle this challenge by identifying perspectives from different answers and then generating a comprehensive answer to the question. In this study, we address the PerAnsSumm Shared Task using two complementary paradigms: (i) a training-based approach through QLoRA fine-tuning of LLaMA-3.3-70B-Instruct, and (ii) agentic approaches including zero- and few-shot prompting with frontier LLMs (LLaMA-3.3-70B-Instruct and GPT-4o) and a Mixture-of-Agents (MoA) framework that leverages a diverse set of LLMs by combining outputs from multi-layer feedback aggregation. For perspective span identification/classification, GPT-4o zero-shot achieves an overall score of 0.57, substantially outperforming the 0.40 score of the LLaMA baseline. With a 2-layer MoA configuration, we were able to improve LLaMA performance up by 28 percent to 0.51. For perspective-based summarization, GPT-4o zero-shot attains an overall score of 0.42 compared to 0.28 for the best LLaMA zero-shot, and our 2-layer MoA approach boosts LLaMA performance by 32 percent to 0.37. Furthermore, in few-shot setting, our results show that the sentence-transformer embedding-based exemplar selection provides more gain than manually selected exemplars on LLaMA models, although the few-shot prompting is not always helpful for GPT-4o. The YaleNLP team's approach ranked the overall second place in the shared task.

Survey on Evaluation of LLM-based Agents

The emergence of LLM-based agents represents a paradigm shift in AI, enabling autonomous systems to plan, reason, use tools, and maintain memory while interacting with dynamic environments. This paper provides the first comprehensive survey of evaluation methodologies for these increasingly capable agents. We systematically analyze evaluation benchmarks and frameworks across four critical dimensions: (1) fundamental agent capabilities, including planning, tool use, self-reflection, and memory; (2) application-specific benchmarks for web, software engineering, scientific, and conversational agents; (3) benchmarks for generalist agents; and (4) frameworks for evaluating agents. Our analysis reveals emerging trends, including a shift toward more realistic, challenging evaluations with continuously updated benchmarks. We also identify critical gaps that future research must address-particularly in assessing cost-efficiency, safety, and robustness, and in developing fine-grained, and scalable evaluation methods. This survey maps the rapidly evolving landscape of agent evaluation, reveals the emerging trends in the field, identifies current limitations, and proposes directions for future research.

Targeted Hard Sample Synthesis Based on Estimated Pose and Occlusion Error for Improved Object Pose Estimation

6D Object pose estimation is a fundamental component in robotics enabling efficient interaction with the environment. It is particularly challenging in bin-picking applications, where objects may be textureless and in difficult poses, and occlusion between objects of the same type may cause confusion even in well-trained models. We propose a novel method of hard example synthesis that is model-agnostic, using existing simulators and the modeling of pose error in both the camera-to-object viewsphere and occlusion space. Through evaluation of the model performance with respect to the distribution of object poses and occlusions, we discover regions of high error and generate realistic training samples to specifically target these regions. With our training approach, we demonstrate an improvement in correct detection rate of up to 20% across several ROBI-dataset objects using state-of-the-art pose estimation models.

Multi-View Keypoints for Reliable 6D Object Pose Estimation

6D Object pose estimation is a fundamental component in robotics enabling efficient interaction with the environment. It is particularly challenging in bin-picking applications, where many objects are low-feature and reflective, and self-occlusion between objects of the same type is common. We propose a novel multi-view approach leveraging known camera transformations from an eye-in-hand setup to combine heatmap and keypoint estimates into a probability density map over 3D space. The result is a robust approach that is scalable in the number of views. It relies on a confidence score composed of keypoint probabilities and point-cloud alignment error, which allows reliable rejection of false positives. We demonstrate an average pose estimation error of approximately 0.5mm and 2 degrees across a variety of difficult low-feature and reflective objects in the ROBI dataset, while also surpassing the state-of-art correct detection rate, measured using the 10% object diameter threshold on ADD error.