Machine Intelligence - Microsoft Research

Intelligent machines and intelligent software rely on algorithms that can reason about observed data to make predictions or decisions that are useful. Such systems rely on machine learning and artificial intelligence, combining computation, data, models, and algorithms. Our mission, in the Machine Intelligence theme at Microsoft Research Cambridge, is to expand the reach and efficiency of machine intelligence technology.

We research how to incorporate structured input data such as code and molecules effectively into deep learning models. We invent new methods so models can accurately quantify their uncertainty when making predictions. We build models that learn from small data that is corrupted or only partially observed. We develop deep learning algorithms that apply to interactive settings in gaming and in decision making task, where model predictions have consequences on future inputs.

Improving the performance of machine learning methods demands an ever-increasing scale in computation while retaining flexibility to develop new models. We research new AI compiler technology that will make it easier to express rich algorithms while effectively utilizing modern accelerators.

Projects

Deep Program Understanding

This project aims to teach machines to understand complex algorithms, combining methods from the programming languages, software engineering and the machine learning communities.

Deep Reinforcement Learning for Games

We aim to teach machines to understand complex algorithms, combining methods from the programming languages, software engineering and the machine learning communities.

Enterprise Knowledge

The aim of the Enterprise Knowledge project is to automatically extract business knowledge into a single, consistent knowledge base, made up of the entities that really matter to each organisation.

Infer.Net

Infer.NET is a .NET library for machine learning. It provides state-of-the-art algorithms for probabilistic inference from data. Infer.NET is open source software under the MIT license.

Project Azua

In this project we investigate how to best utilize AI algorithms to aid decision making while simultaneously minimizing data requirements (and, therefore, cost).

TrueMatch

The TrueMatch matchmaking system decides which people should play together in an online multiplayer game. The Coalition have announced that Gears 5 will use TrueMatch.

TrueSkill

The TrueSkill ranking system is a skill-based ranking system designed to overcome the limitations of existing ranking systems, and to ensure that interesting matches can be reliably arranged within a league.

Featured collaboration

Oxford University logo

University of Oxford

MSR PI: Katja Hofmann
University of Oxford PI: Shimon Whiteson
Joint Postdoctoral Researcher: Mingfei Sun

Reinforcement Learning for Gaming

Mingfei Sun

This project will focus on developing and analysing state-of-the-art reinforcement learning (RL) methods for application to video games. The project aims to tackle two key challenges. First, building effective game AI with RL requires dramatically scaling up existing tools for cooperative multi-agent RL, in which teams of agents must collaborate to complete tasks. Doing so requires new methods for performing multi-agent credit assignment and multi-agent exploration in large state and action spaces. Second, effective game AI must also be able to transfer effectively to new scenarios, such as new game levels and versions, without having to learn from scratch. Doing so requires new methods for transfer and meta-learning in RL that scale to the complexity of modern video games.

Industry collaborators

Ninja Theory logo

Ninja Theory

Ninja Theory was formed in 2004 by four partners, including current Directors Nina Kristensen (Chief Development Director), Tameem Antoniades (Chief Creative Director) and Jez San OBE (Non-Executive Director). The studio pride themselves on striving for the highest production values and continually pushing the boundaries of technology, art and design to create evermore exciting video game experiences.

Find out more about our collaboration with Ninja Theory on the Project Paidia page >

IGGI Centre for Doctoral Training

Industry Partner and Advisory Board Member of the IGGI Centre for Doctoral Training.

Academic Collaborations

Berkely AI Research

Learning to Collaborate with Human Players
Katja Hofmann (MSR Cambridge), Sam Devlin (MSR Cambridge), Kamil Ciosek (MSR Cambridge), Professor Anca Dragan (BAIR), Micah Carroll (PhD student)

Find out more on our Berkeley AI Research collaboration page >

Queen Mary University London

Malmo 2020 Multi-Agent Upgrade
Diego Perez Liebana
Microsoft’s Project Malmo platform enables users to create worlds and learning agents able to play multiple 3D games within Minecraft. In recent years, we have co-organised two international competitions. First on multi-agent learning and, secondly, on sample efficient reinforcement learning with human priors . These competitions have extended the features of the platform, but each introduced their own API, installation instructions and documentation, which has created an unnecessary barrier to researchers wanting to get started with the platform. The objective of this project is to unify the extensions from both competitions back into the original Malmo benchmark, to provide a common entry point for researchers.

PhD collaborations in EMEA

Reinforcement Learning for Enabling Next Generation Human-Machine PartnershipsMSR Supervisor: Sam DevlinExternal Supervisor: Adish Singla

Local Forward Model Learning for Sample-Efficient Sequential Decision Making in Open-World 3D GamesMSR Supervisor: Sam DevlinExternal Supervisor: Diego Perez Liebana

Deep Reinforcement Learning For Collaborative Game AI To Enhance Player ExperienceMSR Supervisor: Sam DevlinExternal Supervisor: TBC

Better Sample Efficiency of Reinforcement LearningMSR Supervisor: Kamil CiosekExternal Supervisor: Amos Storkey

Reinforcement Learning for Adaptive User InteractionMSR Supervisor: Katja HofmannExternal Supervisor: Shimon Whiteson

Intrinsically Motivated Exploration for Lifelong Deep Reinforcement Learning of Multiple TasksMSR Supervisor: Katja HofmannExternal Supervisor: Pierre-Yves Oudeyer