Artificial Intelligence
Building intelligent machines that complement human reasoning and extend what people can do.
Video
Opening remarks: Reinforcement Learning
Reward-based learning has been a foundational component in human psychology. With reinforcement learning, researchers are using reward systems to accelerate AI, where techniques for gaming, robotics, and autonomous systems are being created with an emphasis…
Video
Keynote: Key research challenges for real world reinforcement learning
Reinforcement learning has begun to have a broad impact on society with deployments and systems that make a real and tangible difference in how the world works. This is just scratching the surface though—in the…
Video
Research talk: Resource-efficient learning for large pretrained models
At Microsoft Research, we are approaching large-scale AI from many different perspectives, which include not only creating new, bigger models, but also developing unique ways of optimizing AI models from training to deployment. One of…
Video
Research talk: Successor feature sets: Generalizing successor representations across policies
Successor-style representations have many advantages for reinforcement learning. For example, they can help an agent generalize from experience to new goals. However, successor-style representations are not optimized to generalize across policies—typically, a limited-length list of…
Video
Research talk: Enhancing the robustness of massive language models via invariant risk minimization
Despite the dramatic recent progress in natural language processing (NLP) afforded by large pretrained language models, important limitations remain. A growing body of work demonstrates that such models are easily fooled by adversarial attacks and…
Video
Research talk: Causal learning: Discovering causal relations for out-of-distribution generalization
Machine learning models should be explainable and robust on out-of-distribution samples, especially on safety-critical tasks such as healthcare, and security. However, current models heavily rely on i.i.d assumption, and are therefore sensitive to OOD data.…
Video
Panel: Privacy preserving machine learning
In this panel, Microsoft’s Senior Researcher Shruti Tople will lead a discussion with prominent privacy and security researchers on the state of research in secure computation, differential privacy, and related technologies. The panel will explore…
Video
Panel: Causality in search and recommendation systems
With the scale of search and recommendation, real-time robust and explainable decision-making is at the heart of search and recommendation systems that work robustly even as the user-base changes, new content appears, and topics rise…
Video
Closing remarks: Responsible AI
As we’ve seen in countless media articles, AI systems can behave unfairly or unreliably. They can generate undesirable or harmful content, and they can reproduce or exacerbate existing social inequities. This track explores the complex…
Video
Research talk: Maia Chess: A human-like neural network chess engine
Even when machine learning surpasses human ability in a domain, there are many reasons why AI systems that capture human-like behavior would be desirable. For example, humans may want to learn and collaborate, or humans…
