Bio
I'm a 5th year CS Ph.D. student. I work under supervision of Tom Goldstein at University of Maryland. My research is mostly focused on understanding the deep neural networks; their strength, their vulnerabilities, an how they can be improved. I have several publications regarding understanding deep neural networks. Please checkout the interactive blog . I am also interested in vulnerabilities regarding adversarial attacks. Regarding adversarial robustness, my research makes it possible for individuals with very limited resources/data/computations power to train robust netwroks.
Interests
- Adversarial Machine Learning
- Optimization
- Computer Vision
Education
- Ph.D. in Computer Science at University of Maryland (2017 - Present)
- M.Sc. in Computer Science at University of Maryland (2017 - 2019)
- B.Sc. in Software Engineering at Sharif University of Tech (2012 - 2017)
News
Publications
-
Feature Sonification: An Investigation on the Features Learned for Automated Speech Recognition (ASR)
Appeared in: VISxAI 2022 Blog
- Adversarial Training for Free!
- Adversarially Robust Transfer Learning
- Breaking Certifiable Defenses: Semantic Adversarial Examples with Spoofed Robustness Certificates
- Batch-wise Logit-Similarity: Generalizing Logit-Squeezing and Label-Smoothing
-
Strong Data Augmentation Sanitizes Poisoning and Backdoor Attacks Without an Accuracy Tradeoff
Appeared in: ICASSP 2021 PDF
-
DP-InstaHide: Provably defusing poisoning and backdoor attacks with differentially private data augmentations
Arxiv Paper: PDF
-
Towards Accurate Quantization and Pruning via Data-free Knowledge Transfer
Appreared in: SNN 2021 PDF
-
The Uncanny Similarity of Recurrence and Depth
Appeared in: ICLR 2022 PDF
-
Approximate Competitive Equilibrium with Generic Budget
Appeared in: SAGT2021 PDF
-
On the Efficiency and Equilibria of Rich Ads
Appeared in: IJCAI-2019 PDF
Contact Me!
Feel free to contact me. I would be more than happy to talk to you.
- Email: mohammad.amin.ghiasi[at]gmail.com
- Alt email: amin[at]cs.umd.edu