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.

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.

Clinical concept extraction (CCE) of named entities - such as problems, tests, and treatments - aids in forming an understanding of notes and provides a foundation for many downstream clinical decision-making tasks. Historically, this task has been posed as a standard named entity recognition (NER) sequence tagging problem, and solved with feature-based methods using hand-engineered domain knowledge. Recent advances, however, have demonstrated the efficacy of LSTM-based models for NER tasks, including CCE. This work presents CliNER 2.0, a simple-to-install, open-source tool for extracting concepts from clinical text. CliNER 2.0 uses a word- and character- level LSTM model, and achieves state-of-the-art performance. For ease of use, the tool also includes pre-trained models available for public use.

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 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.

Multi-robot path planning for robot swarms that can both fly and drive

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.).

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

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

From audio recordings of clinician-subject interactions, we determine the spoken language bio-markers that are associated with health outcomes, such as dementia and depression.

A "precision medicine" approach for finding relevant cancer treatments in clinical literature and eligible trials. For a given patient with associated demographics (age, gender) and disease (cancer type, genetic variants), we query a database of all pubmed articles and clinicaltrials.gov trials using NLP techniques to find the most useful and relevant treatments for the patient. Our ensemble-based system performed very well in the TREC 2016 Precision Medicine challenge.

Our goal is to develop better metagenomic binning by origin species of fragments of sequenced environmental DNA.

We develop algorithms for fetal MRI interpretation, to enable noninvasive fetal monitoring from MRI.

The MOOC Learner Project provides learning scientists, instructional designers and online education specialists with open source software that enables them to efficiently extract teaching and learning insights from the data collected when students learn on the edX or open edX platform.