You were awarded an Honourable Mention for the 2019 AAAI / ACM SIGAI Doctoral Dissertation Award. What was the topic of your dissertation research, and what were the main contributions or findings? My PhD dissertation was on the topic of Visual Question Answering, called VQA. We proposed the task of open-ended and free-form VQA - a new way to benchmark computer vision models by asking them questions about images. We curated a large-scale dataset for researchers to train and test their models on this task.

This progress has led to a revival of robust statistics from an algorithmic perspective (see, e.g., [DK19, DKK+21] for surveys on the topic).

Hosted by Eleanor Drage and Kerry McInerney, The Good Robot is a podcast which explores the many complex intersections between gender, feminism and technology. In this episode, we talk to N. Katherine Hayles who's the distinguished research professor at the University of California Los Angeles (UCLA) and the James B. Duke Professor Emerita from Duke University. Her prolific research focuses on the relationship between science, literature and technology in the 20th and 21st centuries. We explore her newest book, Bacteria to AI: Human Futures with Our Nonhuman Symbionts, and discuss how the biological concept of symbiosis can inform the relationships we have with AI; how a neural network experiences the world; and whether ChatGPT can be conscious. N. Katherine Hayles is the Distinguished Research Professor at the University of California, Los Angeles, and the James B. Duke Professor Emerita from Duke University.

Sarah Masud is a fifth-year PhD scholar at the Laboratory for Computational Social Systems (LCS2) at the Indraprastha Institute of Information Technology, Delhi (IIIT-D). She holds the prestigious Google PhD Fellowship (2023-present) was previously awarded the Prime Minister's Doctoral Fellowship (2020-2023). As part of her PhD, she has authored publications in top-tier venues, addressing the analysis of hateful content in online forums. AI Membership Committee and is a Journal of Open Source Software reviewer. Before her academic pursuits, Sarah worked as a data scientist in developer tooling at Red Hat, Bangalore, for 2.5 years.

Researchers from MIT and elsewhere have developed a technique that enables a human to efficiently fine-tune a robot that failed to complete a desired task-- like picking up a unique mug-- with very little effort on the part of the human. Imagine purchasing a robot to perform household tasks. This robot was built and trained in a factory on a certain set of tasks and has never seen the items in your home. When you ask it to pick up a mug from your kitchen table, it might not recognize your mug (perhaps because this mug is painted with an unusual image, say, of MIT's mascot, Tim the Beaver). "Right now, the way we train these robots, when they fail, we don't really know why. So you would just throw up your hands and say, 'OK, I guess we have to start over.' A critical component that is missing from this system is enabling the robot to demonstrate why it is failing so the user can give it feedback," says Andi Peng, an electrical engineering and computer science (EECS) graduate student at MIT. Peng and her collaborators at MIT, New York University, and the University of California at Berkeley created a framework that enables humans to quickly teach a robot what they want it to do, with a minimal amount of effort.

From the gridiron to the battlefield, the study of traumatic brain injury has exploded in recent years. Crucial to understanding brain injury is the ability to model the mechanical forces that compress, stretch, and twist the brain tissue and causing damage that ranges from fleeting to fatal. Models discovered by the Constitutive Artificial Neural Network outperform existing models for brain tissue. Researchers at Stanford University now say they have tapped artificial intelligence to produce a profoundly more accurate model of how deformations translate into stresses in the brain and believe that their approach could reveal a more definitive understanding of when and why concussion sometimes leads to lasting brain damage, and other times not. "The problem in brain modeling to date is that the brain is not a homogeneous tissue – it's not the same in every part of the brain. Yet, trauma is often pervasive," said Ellen Kuhl, professor of mechanical engineering, director of the Living Matter Lab, and senior author of a new study appearing in the journal, Acta Biomaterialia.

Despite the infamous 2016 prophecy of deep learning expert Geoffrey Hinton, artificial intelligence has not yet replaced radiologists. And according to new data, it appears as though that prediction is still a long way off, as an AI "candidate" recently failed its mock radiology boards. The candidate's results were published this week in The BMJ and compared alongside 26 radiologists who had recently passed the rapid radiographic reporting component of the Fellowship of the Royal College of Radiologists (FRCR) examination. Out of ten mock exams, the AI candidate passed two, achieving an overall accuracy of 79.5%, suggesting that the candidate is not quite "ready to graduate." "Radiologists in the UK are required to pass the Fellowship of the Royal College of Radiologists (FRCR) examination before their completion of training, which allows them to practice independently as radiology consultants. For artificial intelligence to replace radiologists, ensuring that it too can pass the same examination would seem prudent," corresponding author Susan Cheng Shelmerdine, a consultant pediatric radiologist at the Great Ormond Street Hospital for Children in London, and colleagues suggested.

Searching the internet can reveal information a user would rather keep private. For instance, when someone looks up medical symptoms online, they could reveal their health conditions to Google, an online medical database like WebMD, and perhaps hundreds of these companies' advertisers and business partners. For decades, researchers have been crafting techniques that enable users to search for and retrieve information from a database privately, but these methods remain too slow to be effectively used in practice. MIT researchers have now developed a scheme for private information retrieval that is about 30 times faster than other comparable methods. Their technique enables a user to search an online database without revealing their query to the server.

Schmidt Futures, a philanthropic initiative co-founded by former Google CEO and Chairman Eric ... [ ] Schmidt and his wife Wendy, is expanding its investment in artificial intelligence research. Schmidt Futures announced today that it was investing $148 million to fund the Eric and Wendy Schmidt AI in Science Postdoctoral Fellowship, a program of Schmidt Futures. With this newest funding, Schmidt Futures, a philanthropic initiative co-founded by former Google CEO and Chairman Eric Schmidt and his wife Wendy, has now committed a total of $400 million to support the development of artificial intelligence (AI) for scientific discovery for other advances in technology and engineering fields. According to the announcement, the new funding will initially support about 160 postdoctoral fellows at nine universities around the world to learn and apply AI methods to their research. The fellowship is expected to expand to more institutions and countries in the future.

Accurate cell-type annotation from spatially resolved single cells is crucial to understand functional spatial biology that is the basis of tissue organization. However, current computational methods for annotating spatially resolved single-cell data are typically based on techniques established for dissociated single-cell technologies and thus do not take spatial organization into account. Here we present STELLAR, a geometric deep learning method for cell-type discovery and identification in spatially resolved single-cell datasets. STELLAR automatically assigns cells to cell types present in the annotated reference dataset and discovers novel cell types and cell states. STELLAR transfers annotations across different dissection regions, different tissues and different donors, and learns cell representations that capture higher-order tissue structures. We successfully applied STELLAR to CODEX multiplexed fluorescent microscopy data and multiplexed RNA imaging datasets. Within the Human BioMolecular Atlas Program, STELLAR has annotated 2.6 million spatially resolved single cells with dramatic time savings. STELLAR (spatial cell learning) is a geometric deep learning model that works with spatially resolved single-cell datasets to both assign cell types in unannotated datasets based on a reference dataset and discover new cell types.