Profile banner
Aditya Taparia

Aditya Taparia

PhD student and ML researcher focused on generative vision

San Francisco Bay Area
Joined August 2026
Favicon imageaditya-taparia.github.ioFavicon imagegithub.comFavicon imagelinkedin.comFavicon imagehuggingface.coFavicon imagesearch.asu.eduFavicon imagedl.acm.org

Network

1.6K connections

Summary

Active machine learning researcher focused on generative vision, explainable AI, and reinforcement learning. Aditya's publication and preprint record centers on concept-based XAI, vision-language generation, and uncertainty in large models. google+2
Combines academic research with applied research internships in industry and national labs. Roles include research assistant and teaching responsibilities at Arizona State University, a student research internship at Lawrence Livermore National Laboratory, and a research intern position at Bosch. asu+1
Maintains open-source code and reproducible research artifacts. Public repositories (e.g., RLPO and LLM-Uncertainty) and a Hugging Face profile show an emphasis on releasing code and papers alongside academic publications. github+2
Regular contributor to peer-reviewed conferences and workshops. Publications and proceedings entries (ICML, PMLR, AAAI proceedings listings and arXiv preprints) indicate active participation in the ML research community. mlr+2

Work

Education

Projects

Writing

The Anatomy of Uncertainty in LLMs

January 1, 2026

Proposes an uncertainty decomposition framework for LLMs that separates uncertainty into input ambiguity, knowledge gaps, and decoding randomness, enabling targeted interventions for auditing LLM reliability and detecting hallucinations.

Favicon imagearxiv.org

Explainable Concept Generation through Vision-Language Preference Learning for Understanding Neural Networks' Internal Representations

January 1, 2024

Frames concept image set creation as an image generation problem and introduces RLPO (reinforcement learning-based preference optimization) to fine-tune vision-language generative models to produce meaningful concept sets for concept-based explanations.

Favicon imagearxiv.org

Failures are fated, but can be faded: Characterizing and mitigating unwanted behaviors in large-scale vision and language models

January 1, 2024

Paper on characterizing and mitigating unwanted behaviors in large-scale vision and language models (listed in publication record).

Favicon imagescholar.google.com