Artificial Intelligence
Building intelligent machines that complement human reasoning and extend what people can do.
Project
Enhancer-Promoter Pair
EP-pair (AKA gene regulation modeling): gene regulation leads to the creation of different cell types that possess different gene expression profiles from the same genome sequence in multicellular organisms. The process is essential for cellular…
Video
Research talk: The science behind semantic search: How AI from Microsoft Bing is powering Azure Cognitive Search
Azure Cognitive Search is a cloud search service that gives developers APIs and tools to build rich search experiences over private, heterogeneous content in web, mobile, and enterprise applications. As part of our AI at…
Video
Roundtable discussion: Beyond language models: Knowledge, multiple modalities, and more
In this roundtable discussion, we will discuss the current state-of-the-art in language models (LMs) and how the lack of external and commonsense knowledge is a limiting factor in several applications, including open domain question answering.…
Video
Closing remarks: Towards Human-Like Visual Learning and Reasoning
Big data-driven deep learning has helped significantly improve the performance of visual tasks in the past few years, but it has also exhibited limitations in scalability and adaptation to real-world scenarios. Researchers and practitioners are…
Video
Research talk: Causal ML and business
Using machine learning for causal inference can, in a subset of cases with rich data, replicate results from A/B experimentation. For other cases, like identifying the “average treatment effect for compliers” ML offers more limited…
Video
Research talk: Towards Self-Learning End-to-end Dialog Systems
At present, deep neural networks have become prevalent for building AI systems for vision, language and multimodality. However, how to build efficient and task-oriented models are still challenging problems for researchers. In these lightning talks,…
Video
Keynote: ReduNet: Deep (convolutional) networks from the principle of rate reduction
In this talk, we will offer an entirely white-box interpretation of deep (convolutional) networks from the perspective of data compression and group invariance. We’ll show how modern deep-layered architectures, linear (convolutional) operators and nonlinear activations,…
Video
Research talk: Learning and pretraining strategies for dense retrieval in search and beyond
In this talk, we’ll quickly go through our recent observations and findings with dense retrieval. First, we’ll recap the standard setup and training strategies for dense retrieval models in search. We’ll then share our recent…
Video
Research talk: An intelligent data-driven paradigm towards cloud reliability
Cloud systems are perhaps the most complicated computing systems developed so far, yet our daily lives depend heavily on their continuous and reliable operations. To achieve systematic cloud reliability, we propose an intelligent data-driven paradigm…
Video
Keynote: Extreme classification for dense retrieval and personalized recommendation
Extreme classification is a new research area pioneered by scientists at Microsoft dealing with classification problems involving millions, or even billions, of categories. In this keynote, partner researcher Manik Varma demonstrates how extreme classification can…
