Self-adaptive reasoning for science
Microsoft is pioneering a vision for a self-adapting AI system that can adapt to the dynamic nature of scientific discovery, promoting deeper, more refined reasoning in complex scientific domains.
阿根廷发现史上最大恐龙：体长达37米（组图）
Self-adaptive reasoning for science
Microsoft is pioneering a vision for a self-adapting AI system that can adapt to the dynamic nature of scientific discovery, promoting deeper, more refined reasoning in complex scientific domains.
Recent Posts
-
-
Project Ire autonomously identifies malware at scale
August 5, 2025Designed to classify software without context, Project Ire replicates the gold standard in malware analysis through reverse engineering. It streamlines a complex, expert-driven process, making large-scale malware detection faster & more consistent.
-
VeriTrail: Detecting hallucination and tracing provenance in multi-step AI workflows
August 5, 2025 | Dasha MetropolitanskyVeriTrail, new from Microsoft Research, can detect AI-generated content that is not supported by the source text, trace the provenance of content from final output back to the source, and locate where errors were likely introduced.
-
Xinxing Xu bridges AI research and real-world impact at Microsoft Research Asia – Singapore
July 23, 2025Xinxing Xu is helping shape the work of Microsoft Research Asia – Singapore by turning advanced AI research into real-world solutions. Learn how he collaborates across sectors and disciplines to drive responsible innovation throughout Southeast Asia.
-
Technical approach for classifying human-AI interactions at scale
July 23, 2025Semantic Telemetry helps LLMs run efficiently, reliably, and in near real-time. Learn about the engineering behind that system, including the trade-offs and lessons learned along the way—from batching strategies to token optimization and orchestration.
-
CollabLLM: Teaching LLMs to collaborate with users
July 15, 2025Recipient of an ICML 2025 Outstanding Paper Award, CollabLLM improves how LLMs collaborate with users, including knowing when to ask questions and how to adapt tone and communication style to different situations. This approach helps move AI toward more user-centric and trustworthy systems.
-
PadChest-GR: A bilingual grounded radiology reporting benchmark for chest X-rays
June 26, 2025 | Daniel Coelho de Castro and Javier Alvarez-ValleThe world’s first multimodal, bilingual radiology dataset could reshape the way radiologists and AI systems make sense of X-rays. PadChest-GR, developed by the University of Alicante with Microsoft Research, has the potential to advance research across the field for years to come.
-
Learning from other domains to advance AI evaluation and testing
June 23, 2025 | Amanda Craig Deckard and Chad AtallaAs generative AI becomes more capable and widely deployed, familiar questions from the governance of other transformative technologies have resurfaced. Which opportunities, capabilities, risks, and impacts should be evaluated? Who should conduct evaluations, and at what stages of the technology lifecycle? What tests or measurements…
-
Breaking bonds, breaking ground: Advancing the accuracy of computational chemistry with deep learning
Microsoft researchers achieved a breakthrough in the accuracy of DFT, a method for predicting the properties of molecules and materials, by using deep learning. This work can lead to better batteries, green fertilizers, precision drug discovery, and more.
-
New methods boost reasoning in small and large language models
New techniques are reimagining how LLMs reason. By combining symbolic logic, mathematical rigor, and adaptive planning, these methods enable models to tackle complex, real-world problems across a variety of fields.
-
Rewriting SymCrypt in Rust to modernize Microsoft’s cryptographic library?
June 10, 2025We're rewriting parts of Microsoft's SymCrypt cryptographic library in Rust to improve memory safety and defend against side-channel attacks, enabling formal verification while maintaining backward compatibility via a Rust-to-C compiler.
-
BenchmarkQED: Automated benchmarking of RAG systems
BenchmarkQED is an open-source toolkit for benchmarking RAG systems using automated query generation, evaluation, and dataset prep. It shows that LazyGraphRAG outperforms standard methods, especially on complex, global queries.
Explore More
-
Events & conferences
Meet our community of researchers, learn about exciting research topics, and grow your network
-
Podcasts
Ongoing conversations at the cutting edge of research
-
Microsoft Research Forum
Join us for a continuous exchange of ideas about research in the era of general AI