Research

We aim to develop a systematic framework for robots to build models of the world and to use these to make effective and safe choices of actions to take in complex scenarios.

Robot scheduling via learning from demonstration for tasks in medicine and more

In order to be able to design synthetic organs that function autonomously, we will need to engineer artificial tissue homeostasis. To control the size of these artificial tissues, two major mechanisms will have to be engineered.

Automatic speech recognition (ASR) has been a grand challenge machine learning problem for decades. Our ongoing research in this area examines the use of deep learning models for distant and noisy recording conditions, multilingual, and low-resource scenarios.

The impending success of AV technology will create vehicles that collect sensor data at a high rate.

Adding domain knowledge to word embeddings.

We study the fundamentals of Bayesian optimization and develop efficient Bayesian optimization methods for global optimization of expensive black-box functions originated from a range of different applications.

Traffic is not just a nuisance for drivers: It’s also a public health hazard and bad news for the economy.

Developing techniques to allow self-driving cars and other AI-driven systems to explain behaviors and failures.

We develop image analysis algorithms for whole-heart cardiac MRI in patients with severe congenital heart disease.

Integrating multi-modal clinical data and using recurrent and convolution neural networks to predict when patients will need important interventions.

We develop novel algorithms for neuroimage segmentation, registration, and joint modeling of brain structure with clinical and genetic data.

Our main goal is to automatically search for relevant answers among many responses provided for a given question (Answer Selection), and search for relevant questions to reuse their existing answers (Question Retrieval).

Our goal is to design novel data compression techniques to accelerate popular machine learning algorithms in Big Data and streaming settings.

Building models that learn spoken language by seeing and hearing

Our team uses accelerometers and machine learning to help detect vocal disorders. We capture data about the motions of patient's vocal folds to determine if their vocal behavior is normal or abnormal.

We aim to understand theory and applications of diversity-inducing probabilities (and, more generally, "negative dependence") in machine learning, and develop fast algorithms based on their mathematical properties.

We use computational models of biological pathways in disease to estimate the range of possible responses to various hypothetical drugs, given patient variability and limited pathway-relevant data.

EQ-Radio can infer a person’s emotions using wireless signals. It transmits an RF signal and analyzes its reflections off a person’s body to recognize his emotional state (happy, sad, etc.).

Our main goal is to develop fact checking algorithms that can assess the credibility of claims mentioned in the textual statements and provide interpretable valid evidence that explains why a certain claim is considered as factually true or fake.

Existing methods for cloning and recombination of DNA enable construction of arbitrary sequences. However, the sequential nature of these techniques makes them time-consuming and expensive. Furthermore, while the transformation of an existing plasmid into a host strain can be reliable when a selection marker is used, there are many current limitations: the number of different plasmids that can be co-transformed is limited by the choice of markers and compatible origins of replication; plasmids are less stable than chromosomal DNA and are difficult to maintain indefinitely without mutation; and cistronic interactions cannot be designed since each new nucleotide sequence added is on an unconnected DNA molecule. To overcome these limitations, we are designing reconfigurable chromosomes consisting of both fixed and variable regions. While the fixed region is carefully optimized and tuned ahead of time, the variable region can be modified in the field, at the point-of-use, leading to rapid and on-demand realization of novel biocircuits with many different phenotypes.

Developing state-of-the-art tools that process 3D surfaces and volumes

Data often has geometric structure which can enable better inference; this project aims to scale up geometry-aware techniques for use in machine learning settings with lots of data, so that this structure may be utilized in practice.

Our goal is to help medical researchers and clincians understand the growing repositories of waveform and signal data collected from critically ill patients.