Machine Learning Researcher @TorontoMet Trustworthy AI and Machine Learning | Uncertainty Quantification in Deep Learning
Toronto, ON, Canada
Joined October 2013
- Tweets17
- Following234
- Followers50
- Likes91
1/7 🧵 Thrilled to share that our paper "LLMs Uncertainty Quantification via Adaptive Conformal Semantic Entropy" was accepted at #IJCAI2026, and I had the opportunity to present it through both oral & poster presentations at IJCAI-ECAI 2026 in Bremen.
📄arxiv.org/abs/2605.04295
6/7
We finally apply conformal calibration on these adjusted uncertainty scores at both prompt and response levels, enabling principled accept/abstain decisions with finite-sample, distribution-free guarantees. Across multiple LLMs & tasks, ACSE outperformed strong UQ baselines.
7/7
Many thanks to my co-author Vaishali Meyappan and my supervisor/co-author Prof. @RezaSamavi for their collaboration, insights, and support.
#TrustworthyAI #LLM #LargeLanguageModels #UncertaintyQuantification #ConformalPrediction #GenerativeAI #MachineLearning #AIResearch
Robust AI should not only resist adversarial perturbations; it should provide certifiable guarantees that its predictions remain stable under carefully crafted attacks.
Our paper, “CEAR: Certified Ensemble Adversarial Robustness in DNNs,” was accepted at Canadian AI 2026.
1/5
CEAR combines model diversity, ensemble consensus, and randomized-smoothing-based certification to improve robustness beyond purely empirical defenses.
Our experiments show improved certified accuracy, larger robustness radii, and reduced adversarial transferability.
4/5
Many thanks to my co-authors and Prof. @RezaSamavi for the excellent collaboration.
Paper: arxiv.org/abs/2606.01437
Code: github.com/tailabTMU/CEAR
#TrustworthyAI #RobustAI #AdversarialRobustness #CertifiedRobustness #DeepLearning #MachineLearning
5/5
Hamed Karimi retweeted
Conformal Prediction (CP) is a nice and (almost) assumption-free framework for quantifying uncertainty by constructing prediction sets with finite-sample coverage guarantees. @thehdmx
Hamed Karimi retweeted
Another well-organized and successful ML Security and Privacy workshop. Active TAILab participation with posters on Uncertainty quantification (@thehdmx @Hirad_Daneshvar) and Certified robustness (@MrezaMaleki_95) . Thanx to @VectorInst @thegautamkamath @TMU_ecbe
Hamed Karimi retweeted
🧵1/n
Good question! My process is usually:
1) Skim the paper, take some notes;
2) Do an in-depth read and start writing strengths/weaknesses/questions (at this point I usually know the score);
3) Sleep on it and re-visit after a few days to write the final review.
Thrilled to present our paper "Evidential Uncertainty Sets in Deep Classifiers Using Conformal Prediction" with @RezaSamavi at COPA2024 in @polimi, Milan! Grateful for the opportunity to discuss our work on Uncertainty Quantification in #DeepLearning.
#ConformalPrediction #AI
Special thanks to Dr. Simone Vantini and @DrMatteoFontana
Paper Link: arxiv.org/pdf/2406.10787
The 13th Symposium on Conformal and Probabilistic Prediction with Applications (COPA2024)
copa-conference.com/