Research

Our goal is to develop a socially intelligent team coacher agent that helps humans communicate, strategize, and work together efficiently.

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.

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

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

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.

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.

Our system for generating ad hoc “cache hierarchies” increases processing speed while reducing energy consumption

The creation of low-power circuits capable of speech recognition and speaker verification will enable spoken interaction on a wide variety of devices in the era of Internet of Things.

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

We develop, parameterize, and validate a model for tumor growth inhibition using in vivo mouse data and study the effects of modeling uncertainty and inter-individual variability on drug candidate efficacy predictions.

We are developing a general framework that enforces privacy transparently enabling different kinds of machine learning to be developed that are automatically privacy preserving.

We develop algorithms, systems and software architectures for automating reconstruction of accurate representations of neural tissue structures, such as nanometer-scale neurons' morphology and synaptic connections in the mammalian cortex.

Using adversarial signals and a cycle-consistency based regularization, we can supplement paired regression tasks with unpaired data to improve regression performance.

This project uses computational biology and chemistry tools to design a pharmaceutical to combat the Zika Virus.

What if we could control robots more intuitively, using just hand gestures and brainwaves?

A polyhedral compiler for expressing image processing, DNN, and linear/tensor algebra applications

We aim to develop a context-aware data-driven functional genomics framework that can characterize tissue-specific gene representations, provide context-aware genotype to phenotype mapping, and enable network-based exploration of disease genetics.

Uhura is an autonomous system that collaborates with humans in planning and executing complex tasks, especially under over-subscribed and risky situations.

All humans process vast quantities of unannotated speech and manage to learn phonetic inventories, word boundaries, etc., and can use these abilities to acquire new word. Why can't ASR technology have similar capabilities? Our goal in this research project is to build speech technology using unannotated speech corpora.

Using medical knowledge to improve learned representations of patient health trajectories.

We enable independent navigation for visually impaired users through a wearable feedback system using time-of-light sensors or depth cameras.

The goal of this project is to develop and test a wearable ultrasonic echolocation aid for people who are blind and visually impaired. We combine concepts from engineering, acoustic physics, and neuroscience to make echolocation accessible as a research tool and mobility aid.