Unlikely research area reveals surprising twist in non-smooth optimization
Modern machine learning is characterized by two key features: high-dimensional models and very large datasets. Each of these features presents its own unique challenges, from basic issues such as storing and accessing all of the…
On the local Hessian in back-propagation
Machine Teaching Demo
Watch a demonstration of the Machine Teaching tool Platform for Interactive Concept Learning (PICL). People without machine learning expertise can build custom machine learning models on their own data. Starting with no labeled data, one…
Getting into a conversational groove: New approach encourages risk-taking in data-driven neural modeling
Microsoft Research’s Natural Language Processing group has set an ambitious goal for itself: to create a neural model that can engage in the full scope of conversational capabilities, providing answers to requests while also bringing…
Sonoma
The Sonoma project aims to develop an autonomous system that can operate indoor farms efficiently and reliably.