Generative models have shown robust performance on speech enhancement and restoration tasks, but most prior approaches operate offline with high latency, making them unsuitable for streaming applications. In this work, we investigate the feasibility of a low-latency, real-time generative speech restoration system based on flow-matching…
Acoustic scene perception involves describing the type of sounds, their timing, their direction and distance, as well as their loudness and reverberation. While audio language models excel in sound recognition, single-channel input fundamentally limits spatial understanding. This work presents Sci-Phi, a spatial audio large language…
Retrieval-Augmented Generation (RAG) is an effective solution to overcome the limitations of Large Language Models (LLMs) in terms of specific-domain knowledge and timely information updates. However, current RAG methods typically respond to queries based on isolated segments, lacking the ability to integrate information within the…
In this work we consider the Metric Steiner Forest problem in the sublinear time model. Given a set V of n points in a metric space where distances are provided by means of query access to an n×n distance matrix, along with a set of…
Agentic workflows commonly coordinate multiple models and tools with complex control logic. They are quickly becoming the dominant paradigm for AI applications. However, serving them remains inefficient with today's frameworks. The key problem is that they expose workflows as opaque sequences of model and tool…
Background Predicting whether an organ offer will be accepted for transplantation remains challenging for several reasons, including large offer volumes, highly imbalanced observations (more declines than acceptances), and lack of information about the human decision-making process. Offer acceptance models are used for risk-adjusted program evaluations…
With the rapid adoption of generative AI in meetings, we anticipate a near-future where people prepare and deploy personalized AI-powered mimetic agents, or digital twins, to speak on their behalf. However, most people have limited experience in preparing for and reviewing conversations made by these…
Large Language Models (LLMs) exhibit strong multilingual capabilities, yet remain fundamentally constrained by the severe imbalance in global language resources. While over 7,000 languages are spoken worldwide, only a small subset (fewer than 100) has sufficient digital presence to meaningfully influence modern LLM training. This…
We introduce LEGOMem, a modular procedural memory framework for multi-agent large language model (LLM) systems in workflow automation. LEGOMem decomposes past task trajectories into reusable memory units and flexibly allocates them across orchestrators and task agents to support planning and execution. To explore the design…
The Machine Learning and Optimization (MLO) group in MSR-Redmond performs research in the intersection of optimization, machine learning and systems. Our focus right now is in combining Large Language Model (LLM) technology with optimization for efficient decision making. Example projects include training LLMs for algorithm…
Graph-based data structures have become powerful and ubiquitous tools for scalable approximate nearest-neighbor (ANN) search over the past decade. In spite of their apparent practical performance, there has only recently been progress on the worst-case performance of these data structures. Indeed, the influential work of Indyx and…
Diffusion-based policies have gained growing popularity in solving a wide range of decision-making tasks due to their superior expressiveness and controllable generation during inference. However, effectively training large diffusion policies using reinforcement learning (RL) remains challenging. Existing methods either suffer from unstable training due to…
