Our current focus in computer vision includes image and video analysis, such as video editing and summarization, space-time analysis of video, the internal statistics of natural images and videos, image generation guided by both text and images, and deep internal learning, which can train a deep network on a single natural image or video. The vision models we are developing range from extraction of a scene’s physical properties such as depth maps and three-dimensional structure, motion estimation and image segmentation, to high-level understanding such as object recognition and scene analysis. Our models of high-level vision also study relationships of AI models with human cognition and the brain, and the reconstruction of visual data from fMRI brain activity. Related models deal with geometry and computer graphics, and include deep learning of geometric and irregular data, geometry processing, and discrete differential geometry. In robotics and motor control, research directions include control of humanoid robots, soft robotics and human-robot interaction, with applications to health care, including motion capture and analysis in various movement and neurological disorders.
In the theory of AI and ML, our current research focuses on theoretical machine learning, with emphasis on algorithms that combine practical efficiency and theoretical insight. Our theoretical studies also include research that aims to promote a foundational understanding of societal concerns in computation, such as security, privacy and algorithmic fairness. The methods we use combine computer science with statistical analysis of complex data and with statistical machine learning.
In work that combines theoretical analysis with engineering and applications of AI, our current research develops new technologies that can efficiently extract and process signals and information across a wide range of tasks. This research develops model-based deep learning, where the design of learning-based algorithms is based on prior domain knowledge. This approach allows to integrate models and physics of signals and other knowledge about the problem into both the architecture and training process of deep networks, resulting in efficient models in a variety of tasks Including super-resolution in imaging, optical microscopy and ultrasound, modern wireless communication systems, radar systems, signal sensing and separation, and more.
In the areas of computational biology, medicine, and health, we combine computational modeling with experimental work, in both basic research and medical applications. One large-scale ongoing research effort aims to develop personalized nutrition and personalized medicine using machine learning, computational biology, probabilistic modeling, and the analysis of heterogeneous high-throughput genomic and clinical data. This effort combines the development of AI models for biomedical and wellness applications with the creation of a novel large-scale data set of over 20,000 participants. The models draw inspiration from large language models but extend their capabilities to handle diverse clinical and multi-omics data modalities, including imaging, time-series, tabular, and sequencing-based data.
Another major research effort focuses on the study of epigenetics and single-cell genomics. It aims to understand tissue function from single cells by combining data analysis and experimental methods. Another example of current research combining AI and health care is an ongoing collaboration that combines different imaging techniques together with clinical data such as electronic health records in natural language for AI-based eyes and diabetes health care.
Recurring seminars and meetings
Local events and workshops
- Translational AI and Engineering in Health and Communication, December 2022
- Societal Concerns in Algorithms and Data Analysis, October-December 2018