Machine learning

A recent review by MIT researchers finds that “only about 23% of machine learning studies in health care used multiple datasets to establish their results, compared to 80% in the adjacent field of computer vision, and 58% in natural language processing,” writes Casey Ross for STAT. “If the performance results are not reproduced in clinical care to the standard that was used during [a study], then we risk approving algorithms that we can’t trust,” says graduate student Matthew McDermott. “They may actually end up worsening patient care.”

Times Higher Education reporter Simon Baker writes that Media Lab researchers have developed a new machine learning model that can predict research studies that will have the highest impact. The tool has the potential to “aid funders and research evaluators in making better decisions and avoiding the kind of biases and gaming that occurred with simpler metric assessments.”

Principal research scientist Leo Anthony Celi speaks with STAT reporter Katie Palmer about the importance of open data sharing in medical research, his new role as editor of PLOS Digital Health, and the challenges facing machine learning in medicine. “With digitization, we’re hoping each country will have an opportunity to create their own medical knowledge system,” says Celi.

In a forthcoming book, photographer Jessica Wynne spotlights the chalkboards of mathematicians, including Professor Alexei Borodin’s and Associate Professor Ankur Moitra’s, reports Clara Moskowitz for Scientific American. 

Wired reporter Will Knight spotlights how MIT researchers have showed that “an AI program trained to verify that code will run safely can be deceived by making a few careful changes, like substituting certain variables, to create a harmful program.”

Graduate student Joy Buolamwini joins Kara Swisher on The New York Times' “Sway” podcast to discuss her crusade against bias in facial recognition technologies. “If you have a face, you have a place in this conversation,” says Buolamwini.

MIT researchers are applying machine learning algorithms typically used for natural language processing to identify coronavirus variants, reports Brian Browdie for Quartz. “Besides being able to quantify the potential for mutations to escape, the research may pave the way for vaccines that broaden the body’s defenses against variants or that protect recipients against more than one virus, such as flu and the novel coronavirus, in a single shot,” writes Browdie. 

CSAIL researchers have developed a new material with embedded sensors that can track a person’s movement, reports Mashable. The clothing could “track things like posture or give feedback on how you’re walking.”

Fast Company reporter Elizabeth Segran spotlights how CSAIL researchers have crafted a new smart fabric embedded with sensors that can sense pressure from the person wearing it. “Sensors in this new material can be used to gather data about people’s posture and body movements,” writes Segran. “This could be useful in a variety of settings, including athletic training, monitoring the health of elderly patients, and identifying whether someone has fallen over.”

Wired reporter Will Knight writes that MIT researchers have found that many of the key AI data sets used to train algorithms could contain many errors. “What this work is telling the world is that you need to clean the errors out,” says graduate student Curtis Northcutt. “Otherwise the models that you think are the best for your real-world business problem could actually be wrong.”

TechCrunch reporter Brian Heater spotlights how MIT researchers have devised a neural network to help optimize sensor placement on soft robots to help give them a better picture of their environment.

Graduate student Joy Buolamwini speaks with Molly Wood of Marketplace about her work uncovering bias in AI systems and her calls for greater oversight of facial recognition systems. “We need the laws, we need the regulations, we need an external pressure, and that’s when companies respond,” says Buolamwini. “But the change will not come from within alone because the incentives are not aligned.”

Writing for Forbes, research affiliate Tom Davenport spotlights how Stitch Fix “uses AI algorithms and human stylists working in combination to make recommendations to clients of items of clothing, shoes, or accessories.”

Research scientist Andreas Mershin speaks with Noam Hassenfeld of Vox about his work developing a new AI system that could be used to detect disease using smell.

A new AI-powered system developed by researchers from MIT and other institutions can detect prostate cancer in urine samples as accurately as dogs can, reports Tanya Lewis and Prachi Patel for Scientific American. “We found we could repeat the training you use for dogs on the machines until we can’t tell the difference between the two,” says research scientist Andreas Mershin.