ML news

by Microsoft Research

How Synthetic Enterprises Build Trust in Microsoft Copilot

From synthetic enterprise data evaluations to real Microsoft Copilot AI confidence - Microsoft Research
...
...

How Magnetar Makes Zero-Knowledge Proofs Practical

Magnetar brings faster, simpler, quantum-ready zero-knowledge proofs
...
...

How AI Simulates Learners to Build Better Tutors

Simulating student learning for better AI tutoring
...
...

How Agent Lightning Trains Real AI Agents with RL

Agent Lightning: Lightweight RL agent-training framework
...
...

How AI Forecasts Space Weather Risks for Power Grids

Forecasting space weather risks on power grids
...
...

Why Robots Should Think Beyond Their Onboard GPUs

What if robots didn't need all their AI onboard?
...
...

How AI Plans Molecules: RetroChimera and Retrosynthesis

Improving synthesis prediction of small molecules at scale with RetroChimera - Microsoft Research
...
...

How Flash Models Scale Pathology AI for Discovery

Making pathology foundation models practical at scale
...
...

How AI Reasons About Topology—and Where It Fails

MindTopo puts AI's spatial reasoning to the test
...
...

How CARE-X Makes Radiology AI More Clinically Precise

A new approach to radiology AI
...
...

How Orchard Makes Agentic AI Scalable and Open

Orchard: An open framework for scalable agentic AI
...
...

Training Computer-Use Agents in Deep, Evolving Worlds

Deep, evolving environments for computer-use agents
...
...

How AI Learns from Experience Without Retraining

EvoLib: Teaching LLMs to learn from experience
...
...

Scaling cryptographic verification to boost computer security

Scaling cryptographic verification to boost computer security
...
...

Aurora 1.5: Extending open foundation models for weather and Earth-system applications - Microsoft Research

Aurora 1.5: Extending open foundation models for weather and Earth-system applications - Microsoft Research
...
...

AI can generate Charts. Flint helps generate better ones.

AI can generate Charts. Flint helps generate better ones.
...
...

SkillOpt turns AI agent skills into trainable assets

SkillOpt turns AI agent skills into trainable assets
...
...

How Memora Gives AI Agents Long-Term Memory

Memora scales agent memory to boost long-horizon productivity
...
...

Turning brain prediction models into testable explanations

Turning brain prediction models into testable explanations
...
...

Talos brings continuous genomic reanalysis to nearly 5,000 unsolved cases

Talos brings continuous genomic reanalysis to nearly 5,000 unsolved cases
...
...