Microsoft Research Cambridge UK is seeking an exceptional post-doctoral researcher to explore the impact of understanding the provenance of digital information on human perception, judgement, and trust. The research will inform new prototypes and designs for ensuring end users can effectively use and understand the…
Microsoft Research AI for Science seeks a highly motivated research engineer with expertise in computational materials science to join our density functional theory (DFT) team. The mission of the team is to enable highly accurate and scalable electronic structure calculations in molecules and materials with…
ColBERT introduced a late interaction mechanism that independently encodes queries and documents using BERT, and computes similarity via fine-grained interactions over token-level vector representations. This design enables expressive matching while allowing efficient computation of scores, as the multi-vector document representations could be pre-computed offline. ColBERT…
Conference services like Zoom, Microsoft Teams, and Google Meet facilitate millions of daily calls, yet ensuring high performance at low costs remains a significant challenge. This paper revisits the problem of packing calls across Media Processor (MP) servers that host the calls within individual datacenters…
In VS Code, GitHub Copilot Chat can access hundreds of tools through the Model Context Protocol (MCP) that range from codebase analysis tools to Azure-specific utilities. But giving an agent too many tools doesn’t always make it smarter. Sometimes it just makes it slower. If you’ve ever…
Large multimodal models (LMMs) demonstrate impressive capabilities in understanding images, videos, and audio beyond text. However, efficiently serving LMMs in production environments poses significant challenges due to their complex architectures and heterogeneous characteristics across their multi-stage inference pipelines. We present the first comprehensive systems analysis…
Joint automatic speech recognition (ASR) and speaker diarization aim to answer the question"who spoke what"in multi-speaker scenarios. In this paper, we present an end-to-end speech large language model (Speech-LLM) for Joint strEamable DIarization and aSr (JEDIS-LLM). The model is trained only on short audio under…
We are seeking Senior Applied Scientists and Principal Applied Scientists (Multiple Positions) with strong research skills and the desire to pursue the cutting edge in model development that pushes technological boundaries. We are looking for candidates with interest and experience in language model training, data…
Recent advances in large language models (LLMs) have accelerated their adoption in software engineering contexts. However, concerns persist about the structural quality of the code they produce. In particular, LLMs often replicate poor coding practices, introducing code smells (i.e., patterns that hinder readability, maintainability, or…
Computer-use agents are AI systems that autonomously navigate and interact with software applications through graphical user interfaces (GUIs), and they are emerging as a new capability in artificial intelligence. By navigating and manipulating the same visual interfaces that people use, they can perform complex tasks…
Multilingual LLMs have become so powerful, yet African languages remain misrepresented. Despite the progress in African NLP, driven by community-led initiatives, LLMs continue to misrepresent African languages and the meanings they convey. In this talk, Dr Shamsuddeen will discuss two decades of AfricaNLP and the…
