View this email in your browser
Dear members of the MIT-IBM community,

As the year comes to close, we'd like to thank you for your contributions to the MIT-IBM Watson AI Lab and share some news. This joint initiative was designed to foster long-term relationships and a cross-pollination of ideas. We are now extending the collaboration to a group of innovative companies that will help us push the boundaries of AI even further. 

We are proud to welcome Samsung, Boston Scientific, Nexplore and Refinitiv as new members. They represent a broad swathe of the economy — consumer technology, health care, construction and finance — and each has demonstrated a commitment to applying AI to their respective industries. These companies will help us to broaden our research and maximize its real-world impact. 

We’d also like to congratulate our faculty and researchers on an exceptional year. The Lab had a strong showing this month at the Conference on Neural Information Processing Systems (NeurIPS) and the International Conference on Computer Vision (ICCV), with MIT and IBM researchers presenting more than a dozen co-authored papers. We highlight some of that work below and on our refreshed Research page.

We wish you a joyful holiday season and thank you for your support.

Antonio Torralba, MIT Director
David Cox, IBM Director
Aude Oliva, MIT Executive Director
Lisa Amini, IBM Research Cambridge Director 


New Dataset Aims to Spur Next-Generation AI

ObjectNet is unlike other datasets. Household objects are tipped on their side and shot at odd angles. MIT and IBM researchers unveiled the new dataset at NeurIPS in an effort to challenge the field to develop better object-detection models.  “We need better, smarter algorithms,” says MIT's Boris Katz in The Verge. 

Visualizing AI Models’ Blind Spots

A new tool reveals what GANs, or generative adversarial networks, leave out when recreating a scene. "When GANs encounter objects they can’t generate, they seem to imagine what the scene would look like without them,” says IBM's Hendrik Strobelt, who presented the study at ICCV. 

Making It Easier for AI Models to Understand Video 

A team led by MIT's Song Han and IBM’s Chuang Gan is teaching artificial intelligence to process more videos while using less power, which could make it easier to apply AI to large amounts of video. The work was presented at ICCV. “Video understanding is so important,” Han told Wired. “But the amount of computation is prohibitive."
What a Little More Computing Power Can Do

The climate is warming and seas are rising. How high will they go? With the help of IBM-donated cloud services, MIT researchers are training a pair of GANs to mimic satellite images and eventually visualize real-world sea-level rise. It’s one of many AI projects that IBM has made possible with in-kind gifts to MIT.


AI may not kill your job—just change it. That was the headline in Wired summarizing a recent study by MIT and IBM researchers analyzing 170 million U.S. job postings for shifts in salary and required skills.
As AI moves into applications, the need for tinier models is growing. The Lab is developing methods to shrink models so they run faster and consume less energy. MIT Technology Review has this profile of the work.


AI Latin America SumMIT

Jan. 21-23, 2020, MIT Media Lab

The Lab is a proud sponsor of an upcoming conference aimed at speeding up the adoption of AI in Latin America. “We believe that AI plays a key role now, and in the future development of the region, if it’s used in the right way,” Omar Costilla Reyes (left), one of four students organizing the event, told MIT News.
MIT-IBM Watson AI Lab, 75 Binney Street
Cambridge, Mass., 02142

This email was sent to *|EMAIL|*
why did I get this?    unsubscribe from this list    update subscription preferences