Microsoft Research

Minecraft Education Summit: Education

Spatial Transformer Networks - Max Jaderberg

If we funded and ran physical infrastructure the way we fund and run scholarly cyberinfrastructure, the lights and water would go out every grant cycle and we would have architects bidding to build

Symposium: Deep Learning; Emily Denton

In the closing keynote for Faculty Summit 2015, Peter Lee, corporate vice president of Microsoft Research, explores the topic of better ways to deliver innovation.

Poster Spotlights.

Chair: Courtney Soderberg, Statistical Consultant at the Center for Open Science, will lead a panel discussion of the challenges associated with organizing research data, disseminating and

Poster Spotlights.

Invited Talks: Computational Principles for Deep Neuronal Architectures.

A distributed algorithm for computing the majority between two populations.

Building Machines That Learn like Humans What is the essence of human intelligence — what makes any human child smarter than any artificial intelligence system that has ever been built?

A 3D animation of DNA strand displacement.

Gaming on phones, tablets and laptops is very popular.

We study the learning of fully connected neural networks for binary classification.

Bayesian models are rooted in Bayesian statistics, and easily benefit from the vast literature in the field. In contrast, deep learning lacks a solid mathematical grounding.

Towards Understandable Neural Networks for High Level AI Tasks; Part 2.

Overview of talk series: Current AI software relies increasingly on neural networks (NNs).

For several decades people have tried to write computer programs that can find proofs of mathematical statements.

Towards understandable neural networks for high level AI tasks - Part III Overview of talk series: Current AI software relies increasingly on neural networks (NNs).

Please help us welcome Michael Gough to Microsoft Research's studio99. Michael is corporate vice president of design for Microsoft's Applications and Services Group.

In this talk, we overview our contribution to data mining and more generally to the cross-fertilization between data mining and propositional satisfiability (

We present a novel technique for verifying that (recursive) heap manipulating programs terminate in polynomial time.

Many people delay or completely avoid updating the software on their devices. This can cause serious issues for security and software management.

Union-Find is a famous example of a simple data structure whose amortized asymptotic time complexity analysis is non-trivial.

The spherical harmonic transform (SHT), which returns spatial frequency components of data or distributions determined on the unit sphere, has found many applications in acoustics, such as spatial

In the field of computer networks, and undoubtedly in other engineering fields, a lot of what "everybody knows" is ... not true.

Online learning is one of the most important and well-established learning models in machine learning.

Speaker diarization consist of automatically partitioning an input audio stream into homogeneous segments (segmentation) and assigning these segments to the same speaker (speaker clustering).

In sequence recognition, the problem of long-span dependency in input sequences is typically tackled using recurrent neural network architectures, and robustness to sequential distortions is

This talk will review our ongoing work on unsupervised latent fault detection in large scale data centers, such as those used cloud services, supercomputers, and compute clusters.