in

Yale researchers win award for best machine learning paper

Yale researchers win award for best machine learning paper

Zoe Berg, Photography Editor

Alexander Tong GRD ’23, a computer science graduate student, and Smita Krishnaswamy, professor of genetics and computer science, won the award for best paper at the annual 2020 Machine Learning for Signal Processing conference, hosted by the Institute of Electrical and Electronics Engineers.

From Sept. 21 to Sept. 24, the MLSP conference was hosted virtually at Aalto University in Espoo, Finland. Tong and Krishnaswamy’s paper, “Fixing bias in reconstruction-based anomaly detection with Lipschitz discriminators,” won the best student paper award alongside two other teams.

The paper identified problems present in many machine learning based outlier detection models. The researchers found some cases where these systems do not work — and this occurs quite often for data types found in bigger data sets. The implications of this work are significant for identifying what kinds of anomalies current methods can detect well and what kinds they might miss.

“This paper is about trying to find outliers or ‘anomalies’ in high dimensional data without having any pre-identified anomalies to train with,” Krishnaswamy, senior author of the paper, wrote in an email to the News. “We propose a neural network that can detect anomalies and give them an anomaly score, given only access to a dataset that can even contain some anomalies itself.”

Over the course of the three-to-four-month research period, the scientists first trained a neural network called an autoencoder to encode the data into an “embedding” and to check whether miscellaneous points were missed in this process.

Because this approach did not entirely work, the team migrated to the idea of using a “discriminator,” which is an external network that decides if data is from the true distribution or an anomaly distribution. This method yielded satisfactory results — the team was able to prove that they were guaranteed to give a high anomaly score to data points that were at a significant distance from most other data points.

Krishnaswamy found this research significant to everyday situations.

“Anomalies can have a huge impact on real-world systems,” she wrote. “Take the hospital system, if a patient’s underlying condition and prescribed medication were at odds with each other (due to a mistake or human error), it can create adverse health effects, liability, lack of trust in the system. To be able to flag these as outliers, which are in some cases errors, is very important for the functioning of the system.”

Krishnaswamy initiated this project in her lab. Then, she and Tong brainstormed ideas and together came up with the discriminator proposal.

Tong, the paper’s first author, performed the coding experiments and proofs of the paper while collaborating on paper edits with Krishnaswamy and Guy Wolf, one of her colleagues at the University of Montreal.

“[This research] seemed like a place where I could contribute both with skills and background knowledge of the lab,” Tong wrote in an email to the News. “The process was a lot of me looking at different datasets and models trying to figure out what might work. … It was encouraging that some people found my work worthwhile and interesting.”

This year marked the MLSP conference’s 30th anniversary, and the official organizer of the conference was the IEEE Signal Processing Society. The conference serves as a workshop at the intersection of machine learning and signal processing.

The papers were reviewed by three double-blind anonymous reviewers. Final award decisions were made by the technical program’s chairs. People from diverse professional backgrounds in physics, statistics, biology and computer science attend the conference.

After being accepted by peer reviewers, the papers for this year’s conference were submitted in April and May and presented at the virtual conference. Simo Särkkä, a professor at Aalto University, led the 2020 conference management team.

“This year’s process was a bit special as we first had to prepare for a physical workshop in terms of local arrangements and venue bookings, but then everything was eventually changed to be virtual,” Särkkä wrote in an email to the News. “It indeed caused some extra work, but was an interesting experience.”

Due to COVID-19, the workshop was virtual because attendants would have to follow Finland quarantine precautions if the conference were in person. The presenters had to record their talks and upload these videos to a conference platform. The videos were then played over Zoom during the virtual conference dates, while live online Q&A sessions were organized for each paper.

The first MLSP conference took place in 1991.

Anjali Mangla | anjali.mangla@yale.edu

Source: yaledailynews.com

What do you think?

48 points
Upvote Downvote

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Artificial Neural Networks Based on Memristive Devices: From Device to System

Artificial Neural Networks Based on Memristive Devices: From Device to System

Artificial Intelligence research at University of Adelaide gets $20 million boost

Artificial Intelligence research at University of Adelaide gets $20 million boost