Research

The project concerns algorithmic solutions for writing fast codes.

Alloy is a language for describing structures and a tool for exploring them. It has been used in a wide range of applications from finding holes in security mechanisms to designing telephone switching networks. Hundreds of projects have used Alloy for design analysis, for verification, for simulation, and as a backend for many other kinds of analysis and synthesis tools, and Alloy is currently being taught in courses worldwide.

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 are investigating the limits of computing on encrypted data, with a focus on the private outsourcing of computation over sensitive data.

To further parallelize co-prime sampling based sparse sensing, we introduce Diophantine Equation in different algebraic structures to build generalized lattice arrays.
With strong relationship to generalized Chinese Remainder Theorem, the geometry properties in the remainder code space, a special lattice space, are explored.

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.

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

We aim to study the causes and transmission modes of infectious diseases among members of a community in the presence of hidden, asymptomatic spreaders of the pathogen.

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.

Transitioning machine learning models across electronic health record (EHR) versions can be improved by mapping different EHR encodings to a common vocabulary.

To enable privacy preservation in decentralized optimization, differential privacy is the most commonly used approach. However, under such scenario, the trade-off between accuracy (even efficiency) and privacy is inevitable. In this project, distributed numerical optimization scheme incorporated with lightweight cryptographic information sharing are explored. The affect on the convergence rate from network topology is considered.

To explore how randomness in connectivity can improve the performance of secure multi-party computation (MPC) and the properties of communication graph to support MPC.

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.

We work towards a principled understanding of the current machine learning toolkit and making this toolkit be robust and reliable.

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

Starling is a scalable query execution engine built on cloud function services that computes at a fine granularity, helping people more easily match workload demand.

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