Feras Saad
|
|
I am an Assistant Professor in the Computer Science Department at CMU, where I am affiliated with the Principles of Programming and Artificial Intelligence groups.
My research develops new techniques for building scalable computing systems that enable flexible probabilistic modeling and sound probabilistic inference. By integrating ideas from programing languages and probabilistic AI, we seek to improve our ability to engineer powerful reasoning systems with a higher degree of automation, accuracy, and scale as compared to existing approaches. I explore applications in diverse areas such as automatic model discovery, time series analysis, and random variate generation; and maintain open-source software tools that help practitioners leverage these techniques in their own domains: see research and software for an overview.
Prior to joining CMU, I was a Visiting Research Scientist at Google. I received the PhD, MEng, and SB degrees in Electrical Engineering and Computer Science from MIT. My theses on probabilistic programming were recognized with the George M. Sprowls PhD Thesis Award in Artificial Intelligence + Decision Making (2023) and Charles & Jennifer Johnson MEng Thesis Award in Computer Science (2017).
Research Opportunities
I am actively recruiting self-motivated students and postdocs at all levels.- Prospective Students: If you are interested in working with me as a graduate student, please apply to the CMU CS PhD program and mention my name in your application.
- Current CMU Students: If you are already a graduate, masters, or undergraduate student at CMU, please send me an email and we can find a time to chat.
- Postdocs: Please send me an email with your CV, two representative papers, and a short description of your background and interests.
Probabilistic Programming.
Researchers in diverse areas (e.g., econometrics, cognitive science,
bioinformatics) work hard to develop computational models of uncertain data
and perform inference from observed data to learn about the underlying
data-generating process. Probabilistic programs, which produce outputs by
making random choices, are a unifying and uniquely expressive approach to
modeling and inference over these processes. Our focus lies in developing
specialized programmable systems that aim to rigorously formalize the
semantics of probabilistic models, automate the process of discovering
these models for complex data, and scale-up the very hard aspects of
probabilistic inference in high-dimensional settings.
[
Automated Probabilistic Model Discovery.
A central challenge in artificial intelligence is automating the process of
discovering accurate and interpretable probabilistic models that let us learn
and reason about data. We attack this problem from the lens of probabilistic program synthesis,
formalized as Bayesian structure learning over rich generative model families.
[
Statistical Estimators & Tests.
As probabilistic programs become a more widespread representation, we need
to develop new statistical analysis methods for estimating their properties
using the black-box computational inferences they expose. While static or
closed-form analysis is possible in some restricted cases, a more general
framework lies in simulation-based dynamic analyses that learn about
programs by analyzing their distributions over execution traces.
[
Random Sampling Algorithms.
This line of work explores fundamental computational limits of random
variate generation, which is a key operation that enables probabilistic
computation. Specific interests include new sampling algorithms that are
theoretically optimal or near-optimal (in entropy consumption and statistical
sampling error) while remaining extremely efficient in practice.
[
Software + Applications.
A central motivation of my work is to develop performant, freely available
