Soheil Feizi
Associate Professor, CS @ UMD
Short Bio: Soheil Feizi is a faculty and the director of Reliable AI Lab in the Computer Science department at University of Maryland, College Park (UMD). Currently on leave from his academic position, he is the Founder and CEO of RELAI , a startup dedicated to advancing AI reliability. He holds a Ph.D. from MIT and completed postdoctoral research at Stanford University . He has published over 100 peer-reviewed papers and given more than 50 invited talks. He has received multiple awards for his work including the ONR's Young Investigator Award, the NSF CAREER Award, the ARO's Early Career Program Award, two best paper awards, the Ernst Guillemin Thesis Award, a Teaching Award, and more than fifteen research awards from national agencies such as NSF, DARPA, ARL, ONR, DOE, NIST as well as industry such as Meta, IBM, Amazon, Qualcomm and Capital One. His work has been featured by major outlets such as the Washington Post, BBC, MIT Technology Review, Bloomberg, and the Wire . Recently, he testified before the U.S. House's Bipartisan Task Force on AI, reflecting his commitment to ensuring AI is developed with safety, accuracy, and reliability in mind. He is committed to promoting diversity in STEM and has mentored several high school, undergraduate, and graduate students through various programs.
Research
My research is centered around developing reliable and trustworthy Artificial Intelligence (AI) with a focus on understanding its robustness (to natural and/or adversarial input variations), generalizability (to unforeseen data domains) and interpretability (of both test and training time predictions). I am interested in the reliability anslysis of both predictive and generative AI models.
Highlights
New: Launched: RELAI, , a startup whose goal is to make AI reliability accessible and achievable for all.
New: Testified before the U.S. House's Bipartisan Task Force on AI. .
New: Interviews with: Washington Post, Wired, MIT Technology Review, Bloomberg
ARO's Early Career Program Award Read More.
ONR Young Investigator Award Read More.
NSF CAREER AWARD. Read More.
Note to Prospective Students:
I am looking for students and post-docs interested in working in theoritical and practical aspects of AI/ML. Information for prospective students can be found here.
Update: I am currently on leave in AY24-25 and will not take new students.
For oppurtunities regarding RELAI, please visit Career Page .
For more info, see my profiles in
Google Scholar, DBLP, LinkedIn and
Twitter.
NEWS:
Launched RELAI
2024
Founded RELAI, a startup whose goal is to make AI reliability accessible and achievable for all. [Read more].
Testified in US Congress
2024
I restified before the U.S. House's Bipartisan Task Force on AI. [Read more].
Early Career Award
2023
I received ARO's Early Career Program Award to study robust dynamic AI systems . [Read more].
Talk at Google's GenAI Workshop
2023
I gave a talk on (un)reliability of AI-text detectors at Google's GenAI workshop [Read more]
AI4All Summer Camp
2023
We hosted a two-week long AI4All camp for high school students at UMD. [Read more]
NSF AI Institute
2023
Our NSF AI Institute on Trustworthy AI in Law & Society (TRAILS) got funded. [Read more].
Amazon Award
2023
I received an Amazon Research Award on understanding spurious correlations in deep learning. [Read more]
Three ICML Papers
2023
Three papers on poisoning robustness and interpretability accepted in ICML'23 [Read more]
Young Investigator Award
2022
I received ONR's Young Investigator Award on studying foundations of robust learning [Read more].
Five NeurIPS Papers
2022
Five papers on adv/distributional robustness, poisoning and Hard ImageNet accepted in NeurIPS'22 [Read more]
Two ICML Papers
2022
Two works (FOCUS, Improved poisoning Robustness) accepted in ICML'22 [Read more]
Plenary Talk
2022
I gave a plenary talk on generative models in the FinDer summer school [Read more]
Three ICLR Papers
2022
Three works (Salient Imagenet, Policy Smoothing, Improved L2 Robustness) accepted in ICLR'22 [Read more]
Two CVPR Papers
2021
Two works (RIVAL10 dataset and analysis, Patch defense for object detection) accepted in CVPR'22 [Read more]
Simons Talk
2022
I gave a talk on studying failure modes of deep learning at Simons/UC Berkeley [Watch here].
CISS Talks
2022
I give two talks on RL robustness and distributional robustness at CISS (Princeton) [Read more]
AISTATS Paper
2021
Our work on provable robustness against fractional threat models accepted in AISTATS'22 [Read more]
UCSD Talk
2021
I give a talk at UCSD's HDSI on studying failure models of deep learning. [Read more]
USC Talk
2021
I give a talk at USC's ML Symposium on studying failure models of deep learning.
UMD Talk
2021
I give a deptartment colloquium talk at UMD on studying failure models of deep learning. [Read more]
NeurIPS paper
2021
Our work on a new training procedure for DL interpretability has been accepted in NeurIPS. [Read more]
NIST Panel Organizer
2021
Organizer and moderator at NIST AI Measurement and Evaluation Workshop. [Read more]
Two ICML papers
2021
Two ICML papers including a long talk (among top 3% of submissions). [Read more]
AISTATS Paper
2021
Our Subadditive GANs work is in AISTATS'21 (oral presentation, among top 3% of submissions)) [Read more]
MIT/Harvard Talk
2021
I gave a talk on distributional robustness at a joint MIT/Harvard seminar.
UW/UT Austin Talk
2021
I gave a talk on generalizable adversarial robustness at a joint UW/UT Austin seminar.
EPFL Talk
2021
I gave a talk on foundations of robust learning at EPFL.
Five ICLR papers
2021
Five ICLR papers on adversarial robustness, GANs and influence functions. [Read more]
AAAI 2021 Paper
Our work on Lottery Tickets in Generative Models has been accepted in AAAI'21 [Read more]
NIST AWARD
2020
Received an award from National Institute of Standards and Technology supporting our research on robustness.
Best Paper Award
2020
from MIT-IBM Watson AI Lab at KDD's Adv ML workshop for our provable poisoning defense. [Read more]
Talk at Princeton's IAS
2020
On Generalizable Adversarial Robustness to Unforeseen Attacks. [Talk Video]
Talk at Capital One
2020
I gave a talk on Unsupervised Anomaly Detection at Capital One Modeling and Analytics Conference.
Three ICML Papers
2020
On curvature-based robustness certificates, smoothing-based robustness certificates, and influence functions. [Read more]
AWS ML Research Award
2020
For “Explainable Deep Learning: Accuracy, Robustness and Fairness”. [Read more]
UMD Research Excellence
2020
I was an honeree at 2020 Maryland Research Excellence Celebration. [Read more]
Deep Generative Model at ITA
2020
I organized a session on deep generative models at ITA 2020. [Read more]
Talk at NIST
2020
I gave a talk on certifiably robust method against adversarial examples at NIST.
Talk at NeurIPS
Dec, 2019
Gave a talk in the ML with Guarantees workshop at NeurIPS [Watch the Video]
Two AISTATS Papers
Dec, 2019
Two AISTATS papers on non-LP adv. robustness and flow-based generative models. [Read more]
Three NeurIPS papers
Three NeurIPS papers on GANs, interpretability and adversarial examples. [Read more]
Robustness Talk
Oct, 2019
I gave a talk on certifiably robust method against adversarial examples [Read more]
Teaching Award
I received the teaching award at UMD for my Fall 2018 and Spring 2019 courses. [Read more]
Deep Learning Workshop
Sept, 2019
I am attending a theory of deep learning workshop at IST, Austria.
ICML Paper
APR, 2019
Our work on deep learning interpretation [paper] has been accepted to ICML 2019.
Best Paper Award
APR, 2019
Our work on Multivariate Maximal Correlation [paper] has received TNSE's best paper award.
Awarded Simons-Berkeley Fellowship
I have received the Simons-Berkeley Research Fellowship on Deep Learning Foundations.
New Paper on arXiv
4
FEB, 2019
Our work on Normalized Wasserstein Distance [paper] is available on arXiv.
New Paper on arXiv
4
FEB, 2019
Our work on Deep Learning Interpretation [paper] is available on arXiv.
New Paper on arXiv
4
FEB, 2019
Our work on Robustness Certificates against Adversarial Examples [paper] is available on arXiv.
Talk at American University
4
FEB, 2019
I gave a talk at American University on generative models.
ICLR Paper
27
Dec, 2018
Our work on Inevitability of Adversarial Examples [paper] has been accepted in ICLR.
Talk at Capital One
17
Dec, 2018
I gave a talk at Capital One's Machine Learning center on Unsupervised Anomaly Detection.
Talk at NIH
30
Nov, 2018
I gave a talk at NIH on ML in biological applications.
Talk at IBM Research
26
Oct, 2018
I gave a talk at IBM research on a statistical approach to generative models.
Lecture at CS Honors
17
Oct, 2018
I gave a lecture at CS honors class on GANs.
New Paper on arXiv
3
OCT, 2018
Our work titled Entropic GANs meet VAEs [paper] is available on arXiv.
Paper Accepted!
29
SEPT, 2018
Our work on Spectral Alignment of Graphs [paper] has been accepted to IEEE Transactions on Network Science and Engineering.
Talk at Quantum Machine Learning Workshop
27
SEPT, 2018
I gave an invited talk titled Generative Adversarial Networks: Formulation, Design and Computation in the QML workshop.
Paper Accepted!
5
SEPT, 2018
Our work on understanding the Landscape of Neural Networks [paper] has been accepted to NeurIPS.
Paper Accepted!
20
JUL, 2018
Our work on Source Inference in Graphs [paper] has been accepted to IEEE Transactions on Network Science and Engineering.
Machine Learning Course
01
JUL, 2018
I am teaching CMSC 726 in Fall 2018. See the course webpage here.
First Day @ UMD
01
JUL, 2018
I am officially starting my faculty career at CS@UMD. I am also a member of UMIACS.
Talk @ Google Research
28
JUN, 2018
I am giving a talk titled "GANs: model-based or model-free?" in google research.