Preprint
Diffusion-based Prompt Mapping for LLM Evaluation
Under submission
Hi, I'm Ali
I am a fifth-year Ph.D. student in the College of Information Sciences and Technology at Penn State, advised by Prof. Suhang Wang. I also received my M.S. and B.S. (Honors) in Computer Science at Penn State.
My research is on AI safety, centered on the gap between what language models can do and what they can be trusted to do. I build methods to uncover their hidden risks, from private data leaks to rare failures that standard testing misses, and study how to build in safer behavior, from curated pretraining to inference that stays grounded in context and avoids hallucination.
Before my Ph.D., I spent fifteen years building production systems, including a national data platform, and founded and scaled an e-commerce company.
Research interests
Interested in collaborating? Reach out at aha112 at psu dot edu.
News
- 10/2026Two new co-first-author papers with Elsayed Eshra: Diffusion-based Prompt Mapping for LLM Evaluation (under submission) and Quantifying Behavioral Tails in Black-Box Language Models [Paper].
- 09/2026Released ali-llm v0.1, a 2B model pretrained from scratch on a children’s curriculum (pre-K to grade 5) to study how early training data shapes later behavior, toward safer pretraining. [🤗 Model]
- 05/2026BLUFF, a benchmark for detecting false and synthetic content across 58 low-resource languages, accepted at KDD 2026. [Paper]
- 05/2026Position paper accepted to ICML 2026. [Paper]
- 04/2026Two papers accepted to ACL 2026. DIA-HARM received the Social Impact Paper Award 🏆 [Paper]
- 03/2026Will be interning as an AI Engineer at Capital One, San Jose this summer.
- 02/2026Our Penn State team was selected for the Amazon Nova Trusted Software Agents Challenge. [Story]
- 01/2026New paper: A Systemic Evaluation of Multimodal RAG Privacy. We evaluate VLM privacy risk to MIA and retrieval attacks under novel settings, such as transformed images. [Paper]
Older news
- 10/2025Authored the Privacy-preserving Cloud–Edge Collaboration section in a new survey on SLM–LLM collaboration. [Paper]
- 08/2025Two papers accepted at EMNLP 2025 Findings, including graph-based RAG ICL selection for molecular learning. [Paper]
- 06/2025Started as an Applied Scientist Intern at Amazon, Seattle.
- 11/2024Paper accepted at COLING 2025. We introduce self-correcting iterative prompting to generate a SQL2Text dataset, and benchmark a graph-based RAG ICL selection mechanism. [Paper]
- 11/2024Invited talk on WildGraph at the Interdisciplinary Scientific Computing Laboratory. [Slides]
- 10/2024Presented WildGraph at SIGSPATIAL 2024. An encoder–decoder with attention over graph-based embeddings learns and generates long-horizon trajectories from very few samples. [Paper]
- 10/2023Presented a workshop paper on VAE-based trajectory generation. [Paper]
- 05/2023Visiting intern at L3S Research Center, Hannover, Germany.
- 08/2022Began my Ph.D. at Penn State.
Publications
†equal contribution. Full list on Google Scholar.
Preprints
arXiv
arXiv
arXiv
A Survey on Collaborating Small and Large Language Models for Performance, Cost-Effectiveness, Cloud-Edge Privacy, and Trustworthiness
arXiv preprint
I authored the privacy-preserving cloud–edge collaboration section.
2026
ICML 2026
Position: LLM Benchmark Datasets Should Be Contamination-Resistant
International Conference on Machine Learning, Position Paper Track (to appear)
ACL 2026
Do Multimodal RAG Systems Leak Data? A Comprehensive Evaluation of Membership Inference and Image Caption Retrieval Attacks
Findings of the Association for Computational Linguistics (to appear)
We evaluate VLM privacy risk to membership inference and retrieval attacks under novel settings, such as transformed images.
ACL 2026
DIA-HARM: Dialectal Disparities in Harmful Content Detection Across 50 English Dialects
Annual Meeting of the Association for Computational Linguistics (to appear)
KDD 2026
BLUFF: Benchmarking the Detection of False and Synthetic Content across 58 Low-Resource Languages
ACM SIGKDD Conference on Knowledge Discovery and Data Mining
2025
EMNLP 2025
GAMIC: Graph-Aligned Molecular In-context Learning for Molecule Analysis via LLMs
Findings of Empirical Methods in Natural Language Processing
Graph-based RAG in-context example selection for molecular learning with LLMs.
EMNLP 2025
Chain-of-Interactions: Multi-step Iterative ICL Framework for Abstractive Task-Oriented Dialogue Summarization of Conversational AI Interactions
Findings of Empirical Methods in Natural Language Processing
COLING 2025
Semantic Captioning: Benchmark Dataset and Graph-Aware Few-Shot In-Context Learning for SQL2Text
International Conference on Computational Linguistics
Self-correcting iterative prompting builds a SQL2Text dataset, and a graph-based RAG ICL selection mechanism is benchmarked on it.
2024
SIGSPATIAL 2024
WildGraph: Realistic Long-Horizon Trajectory Generation with Limited Sample Size
ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems
An encoder–decoder with attention over graph-based embeddings generates long-horizon trajectories from very few samples.
2023
CIKM 2023 Workshop
WildGEN: Long-horizon Trajectory Generation for Wildlife
CIKM Workshop on Knowledge Extraction and Management for Wildlife Conservation (InfoWild)
Experience
- 2022 – present Graduate Researcher Penn State, College of IST Advised by Prof. Suhang Wang
- Summer 2026 AI Engineer Intern Capital One San Jose, CA
- Summer 2025 Applied Scientist Intern Amazon Seattle, WA
- Summer 2023 Visiting Research Intern L3S Research Center Hannover, Germany
Education
- 2022 – present Ph.D., Information Sciences and Technology The Pennsylvania State University
- M.S., Computer Science The Pennsylvania State University
- B.S. (Honors), Computer Science The Pennsylvania State University
Teaching
Teaching Assistant, Penn State
- DS 340W: Applied Data ScienceSpring 2026
- DS 340W: Applied Data ScienceFall 2025
- DS/CMPSC 442: Artificial IntelligenceSpring 2025
- DS 410: Data Analytics at ScaleFall 2024
- IST 402: Emerging Technologies: Data and AI ColonialismSpring 2024
- IST 402: Emerging Technologies: The Question Concerning TechnologyFall 2023
Service
- Reviewer: KDD, RecSys, SDM, TIST, ARR, COLM