DSPLab UMICH-Dearborn @UM_DSPLab

Our latest paper to appear @acm_ccs 2025, "PoisonSpot: Precise Spotting of Clean-Label Backdoors via Fine-Grained Training Provenance Tracking", brings provenance to the frontline of ML resilience in the face of stealthy backdoor attacks. Preprint & Code: github.com/um-dsp/PoisonS…

In his new trustworthy AI course, UM-Dearborn Associate Professor Birhanu Eshete is helping students go beyond the technical know-how and understand the social, political and economic contexts of the artificial intelligence boom. Find out more ➡️ myumi.ch/9gG5r

This semester, I am teaching a new course: Trustworthy AI/ML with focus on the need for holistic trustworthiness assessment of AI/ML used in high stakes applications. Course website (to be updated as we go): trustworthy-ml-course.github.io Course trailer 🙂 👇 youtube.com/watch?v=VUGG2w…

NIST just released a comprehensive taxonomy and terminology of attacks and mitigations for machine learning in adversarial settings. Very useful if you work in this space. Full report from @NIST 👇 nvlpubs.nist.gov/nistpubs/ai/NI…

Proud advisor moment: @AbderrahmenAmi2 presenting “DeResister: Toward Detection-Resistant Probing for Evasion of Internet Censorship” at @USENIXSecurity. #usesec23

Second day @USENIXSecurity: Excited to chair the session on "Adversarial Examples" At 1:30pm, stop by Platinum Salon 9–10 to listen to five talks on latest advances in making machine learning robust to adversarial examples. #usesec23

Today @USENIXSecurity, in the Censorship and Internet Freedom Session (9am), my PhD student Abe Amich (@AbderrahmenAmi2) will present DeResistor: a detection-resilient censorship evasion system that enables bypassing censorship while defusing censor's detection moves. #usesec23

First day of @USENIXSecurity going well with diverse set of interesting talks and meeting old colleagues and new ones! #usesec23

keyboard_arrow_left Previous keyboard_arrow_right Next

Excited to share that our paper "DeResistor: Toward Detection-Resistant Probing for Evasion of Internet Censorship" will be presented at @USENIXSecurity this week! Work led by @AbderrahmenAmi2. Done with Vinod Yegneswaran (@SRI_Intl) & @NP_tokumei (@UChicago). #usesec23

keyboard_arrow_left Previous keyboard_arrow_right Next

Our @PET_Symposium paper "MIAShield: Defending Membership Inference Attacks via Preemptive Exclusion of Members" will be presented in Session 6B: Privacy and AI/ML III on Thursday, July 13th. Camera Ready: petsymposium.org/popets/2023/po…

PETS @PET_Symposium

2 years ago

Our @PET_Symposium paper "MIAShield: Defending Membership Inference Attacks via Preemptive Exclusion of Members" will be presented in Session 6B: Privacy and AI/ML III on Thursday, July 13th. Camera Ready: petsymposium.org/popets/2023/po… https://t.co/xRR3BOJFXV

0 15 34 22K 1

Cybersecurity researcher Birhanu Eshete scores prestigious NSF CAREER award umdearborn.edu/news/cybersecu…

This week, we presented two papers at @acmcodaspy'22. Full talk videos: EG-Booster: dl.acm.org/doi/10.1145/35… DP-UTIL: dl.acm.org/doi/10.1145/35…

keyboard_arrow_left Previous keyboard_arrow_right Next

tinyurl.com/337ma3ja

📢New 📰 📢 Given f(x)=y, you obtain x'= x+δ (δ: feature-space perturbation). As a result, f(x')=y'≠y (f is fooled!). Q: is there correlation b/n δ and feature-based explanation of f(x') = y'? A: check our new SecureComm'21 paper 👇by @AbderrahmenAmi2 arxiv.org/abs/2106.15820

New paper 📢: We empirically explore the combination of additive secret sharing and differential privacy for privacy-preserving collaborative prediction with minimal loss in model accuracy. Paper: arxiv.org/abs/2102.09751 To appear at IWSPA'21 co-located with @acmcodaspy '21

Today, we make public our malware infection traffic characterization engine called MalEvol. Given a PCAP as input, MalEvol provides threat-centric details to assist security analysts make informed decisions. #malware github.com/um-dsp/MalEvol

Our latest work leverages representation mapping and cross-domain transferability for query-constrained black-box approximation of ML malware classifiers. Pre-print: www-personal.umd.umich.edu/~birhanu/files…

keyboard_arrow_left Previous keyboard_arrow_right Next

Two PhD students join @UM_DSPLab. Welcome aboard Ismat Jarin and Abderrahmen Amich!

Our latest work on behavioral alignment of natural language threat intelligence with kernel audit records for effective cyber threat hunting is accepted to @acm_ccs!

On April 1st, academia, industry, and government stakeholders will get together for a full-day event on Cybersecurity at @UM_Dearborn. @UMDearbornCECS @ChanceGrasso Event site: tinyurl.com/y9fg5ju3