Soheil Feizi
Associate Professor
Leads the Reliable AI Lab, studying robustness, interpretability and detection of AI-generated content in deep learning models.
Soheil Feizi's lab at UMD works on trustworthy & safe ML, interpretability, evaluation & safety of LMs and recognition, detection & segmentation. 59 papers since 2023, 41 at top venues, mostly ICLR and NeurIPS. 12 PhD students. 13 alumni.
At a glance
- papers, 2023–2026
- 15/ year
- at T1 venues
- 69%
- PhD students
- 12
- MS students
- 0
- alumni listed
- 13
Papers by yearfrom DBLP, 2023–2026
- 8 papers2026 · T1 4 · collecting
- 16 papers2025 · T1 9
- 17 papers2024 · T1 14
- 18 papers2023 · T1 14
T1 venuesOtherVenues: ICLR, NeurIPS, ICML, EMNLP, AAAI
Research focuspaper titles that use each topic's words
- Interpretability, evaluation & safety of LMs2 papers
- Recognition, detection & segmentation2 papers
- ai-generated2 papers
- interpretability2 papers
- Trustworthy & safe ML0 papers
Paper list
From DBLP.
20268 papers
- EACLAttacker's Noise Can Manipulate Your Audio-based LLM in the Real World (opens in a new tab)
- EACLDecomposition-Enhanced Training for Post-Hoc Attributions in Language Models (opens in a new tab)
- ICLR (T1)GHOST: Hallucination-Inducing Image Generation for Multimodal LLMs (opens in a new tab)
- ICLR (T1)Revisiting the Past: Data Unlearning with Model State History (opens in a new tab)
- ACL (T1)Schoenfeld's Anatomy of Mathematical Reasoning by Language Models (opens in a new tab)
- CVPR (T1)SliderEdit: Continuous Image Editing with Fine-Grained Instruction Control (opens in a new tab)
- Trans. Mach. Learn. Res.SpurLens: Automatic Detection of Spurious Cues in Multimodal LLMs (opens in a new tab)
- Findings of ACLYour LLM Agents are Temporally Blind: The Misalignment Between Tool Use Decisions and Human Time Perception (opens in a new tab)
202516 papers
- Findings of EMNLPA Closer Look at Bias and Chain-of-Thought Faithfulness of Large (Vision) Language Models (opens in a new tab)
- NeurIPS (T1)A Technical Report on "Erasing the Invisible": The 2024 NeurIPS Competition on Stress Testing Image Watermarks (opens in a new tab)
- NeurIPS (T1)Adversarial Paraphrasing: A Universal Attack for Humanizing AI-Generated Text (opens in a new tab)
- Findings of ACLAlmost AI, Almost Human: The Challenge of Detecting AI-Polished Writing (opens in a new tab)
- Trans. Mach. Learn. Res.Can AI-Generated Text be Reliably Detected? Stress Testing AI Text Detectors Under Various Attacks (opens in a new tab)
- EMNLP (T1)DyePack: Provably Flagging Test Set Contamination in LLMs Using Backdoors (opens in a new tab)
- ICLR (T1)How Learnable Grids Recover Fine Detail in Low Dimensions: A Neural Tangent Kernel Analysis of Multigrid Parametric Encodings (opens in a new tab)
- NeurIPS (T1)Localizing Knowledge in Diffusion Transformers (opens in a new tab)
- PerCom WorkshopsMitigating Compositional Failures in Text-to-Image Models with Causal Text Embedding Refinement (opens in a new tab)
- SIGecom Exch.Online Advertisements with LLMs: Opportunities and Challenges (opens in a new tab)
- ACL (T1)RePanda: Pandas-powered Tabular Verification and Reasoning (opens in a new tab)
- Trans. Mach. Learn. Res.RESTOR: Knowledge Recovery in Machine Unlearning (opens in a new tab)
202417 papers
- WACVData-Centric Debugging: mitigating model failures via targeted image retrieval (opens in a new tab)
- NeurIPS (T1)Decomposing and Interpreting Image Representations via Text in ViTs Beyond CLIP (opens in a new tab)
- EMNLP (T1)Distilling Knowledge from Text-to-Image Generative Models Improves Visio-Linguistic Reasoning in CLIP (opens in a new tab)
- ICLR (T1)DRSM: De-Randomized Smoothing on Malware Classifier Providing Certified Robustness (opens in a new tab)
- ICML (T1)Fast Adversarial Attacks on Language Models In One GPU Minute (opens in a new tab)
- EMNLP (T1)IntCoOp: Interpretability-Aware Vision-Language Prompt Tuning (opens in a new tab)
- NeurIPS (T1)LLM-Check: Investigating Detection of Hallucinations in Large Language Models (opens in a new tab)
- ICLR (T1)Localizing and Editing Knowledge In Text-to-Image Generative Models (opens in a new tab)
- NeurIPS (T1)Loki: Low-rank Keys for Efficient Sparse Attention (opens in a new tab)
- AAAI (T1)Measuring Self-Supervised Representation Quality for Downstream Classification Using Discriminative Features (opens in a new tab)
- ICML (T1)On Mechanistic Knowledge Localization in Text-to-Image Generative Models (opens in a new tab)
- ICLR (T1)PRIME: Prioritizing Interpretability in Failure Mode Extraction (opens in a new tab)
202318 papers
- BMVCAdapting Self-Supervised Representations to Multi-Domain Setups (opens in a new tab)
- ICLR (T1)Certifiably Robust Policy Learning against Adversarial Multi-Agent Communication (opens in a new tab)
- CVPR (T1)CUDA: Convolution-Based Unlearnable Datasets (opens in a new tab)
- NeurIPS (T1)Diffused Redundancy in Pre-trained Representations (opens in a new tab)
- NeurIPS (T1)Exploring Geometry of Blind Spots in Vision models (opens in a new tab)
- AAAI (T1)Goal-Conditioned Q-learning as Knowledge Distillation (opens in a new tab)
- ICLR (T1)Hard-Meta-Dataset++: Towards Understanding Few-Shot Performance on Difficult Tasks (opens in a new tab)
- Found. Trends Priv. Secur.Identifying and Mitigating the Security Risks of Generative AI (opens in a new tab)
- ICML (T1)Identifying Interpretable Subspaces in Image Representations (opens in a new tab)
- TPAMI (T1)Interpolated Joint Space Adversarial Training for Robust and Generalizable Defenses (opens in a new tab)
- Trans. Mach. Learn. Res.Interpretable Mixture of Experts (opens in a new tab)
- ICLR (T1)Provable Robustness against Wasserstein Distribution Shifts via Input Randomization (opens in a new tab)
Where alumni went13 listed
Apply to UMDFall 2027
Not announced yet
UMD Computer Science PhD
- Round
- Single round
- Opens
- —
- GRE
- See program page
- English
- See program page
- Fee
- See program page
Not yet confirmed for this intake.
Similar labsnearest by research summary
- USUMD
Tom Goldstein
- Trustworthy & safe ML
- GBOxford
Philip Torr
- Trustworthy & safe ML
- GBImperial
Jiankang Deng
- Vision-language & multimodal
- USUSC
C.-C. Jay Kuo
- Video understanding
- USUC Berkeley
Trevor Darrell
- Vision-language & multimodal
- CNTsinghua
Wenming Yang
- Vision-language & multimodal
Your notes
Save this lab to keep notes.
Data
Checked against the lab homepage.
- Collected
- Verified
- Papers
- DBLP
- Members
- Lab homepage
- Alumni
- Lab homepage
Tiers are venue classifications, not lab rankings.