Microsoft researchers present 18 papers at the International Conference on Machine Learning

Published

By Athima Chansanchai, Microsoft News Center Staff

Machine learning covers a lot of ground. At Microsoft, it’s being incorporated to detect lies, recognize human responses and forecast finances; as well as improve search, natural language processing, advertising, security and gaming. It’s a broad discipline that touches daily life through artificial intelligence and the cloud – and it’s growing by leaps and bounds.

John Langford

(opens in new tab) John Langford, Principal Researcher, Microsoft Research

Digital pathology helps decode tumor microenvironments for precision immunotherapy. GigaPath is a novel vision transformer that can scale to gigapixel whole-slide images by adapting dilated attention for digital pathology. In joint work with Providence and UW, we’re sharing Prov-GigaPath, the first whole-slide pathology foundation model pretrained on large-scale real-world data, for advancing clinical research and discovery.

GigaPath: Whole-Slide Foundation Model for Digital Pathology

Digital pathology helps decode tumor microenvironments for precision immunotherapy. In joint work with Providence and UW, we’re sharing Prov-GigaPath, the first whole-slide pathology foundation model, for advancing clinical research.

Read more

“Machine learning is working. It’s making a big difference in a lot of different applications that really matter for the future,” says John Langford (opens in new tab), an expert on machine learning at the Microsoft Research lab in New York City who is also the general chair for the International Conference on Machine Learning (opens in new tab), which has grown by 65 percent since last year thanks to the technology’s success.  “Figuring out how to use data to make decisions to help people is what machine learning is about.”

Machine learning saves time. A lot of it. Advanced analytics and data science resources make it possible for once-arduous tasks to get done quickly. It can speed up a multitude of normally time-consuming processes, such as vision recognition, causality, crowd sourcing and more.

“Now we’re seeing more work in neural networks and deep learning than in previous years,” says Langford of some prevalent themes in this year’s conference. “There’s quite a lot of people working on a lot of different subjects. By far, this is the largest ICML ever. The field is really growing fast.”

Focused on machine learning, algorithms and systems, ICML begins Sunday, June 19, and includes tutorials, presentations of accepted papers and workshops on more recent research. Nearly 3,000 participants are expected at the five-day conference.

“Microsoft has a longstanding role in this community. We’ve supported machine learning research for decades,” Langford says. The conference is so popular this year, they were in need of more space, he says. Luckily, the Microsoft Technology Center is next door to help handle the overflow.

While more than 1,300 papers were submitted, only 332 were accepted. Out of those, 18 are collaborations with Microsoft researchers.

One of them, “No Oops, You Won’t Do It Again: Mechanisms for Self-correction in Crowdsourcing (opens in new tab),” (by Nihar Shah at UC Berkeley and Dengyong Zhou of Microsoft Research) focuses on improving the quality of data using a self-correction mechanism. Another, “CryptoNets: Applying Neural Networks to Encrypted Data with High Throughput and Accuracy (opens in new tab),” (by Nathan Dowlin of Princeton; and Ran Gilad-Bachrach, Kim Laine, Kristin Lauter, Michael Naehrig and John Wernsing of Microsoft Research) looks at how machine learning can help maintain privacy and security with medical, financial and other sensitive data. Their work involves a method that allows a person to send their data in an encrypted form to a cloud service that hosts the network, which keeps the data confidential since the cloud does not have access to the keys needed to decrypt it.

And “Doubly Robust Off-policy Value Evaluation for Reinforcement Learning (opens in new tab),” (by Nan Jiang at the University of Michigan and Lihong Li of Microsoft Research) studies the problem of estimating the value of a new policy based on data collected by a different policy in reinforcement learning (RL). This problem is often a critical step when applying RL to real-world problems. Their research guarantees a lack of bias and can have a much lower variance than the popular importance sampling estimators.

The other accepted papers at ICML that feature Microsoft researchers are:

In addition to the papers, there are two workshops with Microsoft researchers: “Multi-View Representation Learning” with Xiaodong He and Scott Wen-tau Yih, and “Advances in non-convex analysis and optimization” by Praneeth Netrapalli.

Related:

Athima Chansanchai is a writer for the Microsoft News Center. Follow her on Twitter (opens in new tab).

Continue reading

See all blog posts

Research Areas