Alireza Kheirandish

Alireza Kheirandish

PhD Student. Georgia Tech ECE.

email / github / google scholar / cv

I am a PhD student in Electrical and Computer Engineering at Georgia Tech. My research focuses on generative models for inverse problems.

Recently, I have been working on localized out-of-distribution detection using diffusion priors, with applications in computational imaging, medical imaging, and visual anomaly detection. I also work on symbolic reinforcement learning and LLM-guided task decomposition for embodied decision-making.

News

  • 2026. Our paper KLIP was accepted to CVPR 2026.
  • 2026. Our paper GT-NLP was accepted to workshop on SemEval.
  • 2025. Our paper on LLM-augmented symbolic reinforcement learning appeared at ICASSP 2025.

Research

My work studies how generative priors can help solve inverse problems when the observation may contain unusual or out-of-distribution structure. I am especially interested in methods that do not only reconstruct an image, but also explain where and why the reconstruction may be unreliable.

More broadly, I am interested in diffusion models, large language models, reinforcement learning, and statistics.

Selected Publications

KLIP
CVPR 2026
KLIP: Localized Distribution Shift Detection via KL-Divergence with Diffusion Priors in Inverse Problems
Alireza Kheirandish, Jihoon Hong, Sara Fridovich-Keil.
CVPR 2026
paper code slides
Symbolic RL
ICASSP 2025
LLM-Augmented Symbolic RL with Landmark-Based Task Decomposition
Alireza Kheirandish, Duo Xu, Faramarz Fekri
IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2025
arxiv doi
SemEval
2025
GT-NLP at SemEval-2025 Task 11: EmoRationale, Evidence-Based Emotion Detection via Retrieval-Augmented Generation
Daniel Saeedi, Alireza Kheirandish, Sirwe Saeedi, Hossein Sahour, Aliakbar Panahi, Iman Ahmadi Naeeni
Proceedings of the 19th International Workshop on Semantic Evaluation (SemEval-2025), ACL, 2025
acl pdf code

Projects

KLIP

Localized out-of-distribution detection for inverse problems using diffusion posterior sampling and KL-based trajectory analysis.

project page
LLM-Augmented Symbolic RL

Landmark-based task decomposition that uses large language models to guide symbolic reinforcement learning in long-horizon environments.

project page
Wearable Glasses for EEG, EOG sensingt

Designing a wearable system integrated into a pair of eyeglasses to mitigate the risks associated with drowsy and distracted driving through early detection.

project page
EmoRationale

Retrieval-augmented emotion detection for SemEval-2025 Task 11, using retrieved multilingual examples and LLM reasoning for interpretable emotion classification.

project page
Multi Modal Protein Mutation Effect

This project was developed for the Generative and Geometric Deep Learning course.

notes

Contact

The best way to reach me is by email at akheirandish3@gatech.edu.