Sepehr Dehdashtian

428 S Shaw Ln # 3208

East Lansing, MI 48824

Hi there! I’m Sepehr, currently pursuing a PhD in Computer Science at Michigan State University, with a Master’s degree in Electrical Engineering from Sharif University of Technology.

My work centers on reliability and representation learning in multimodal and generative models, as well as on understanding and mitigating bias and fairness issues in generative AI systems. I have investigated stereotype detection and steerability in diffusion and flow-matching text-to-image models, and I am currently exploring post-training policy enforcement for foundation models, applying both RL-based (e.g., RLHF) and non-RL (e.g., DPO) methods. Through these efforts, I aim to identify and address failure modes where models deviate from safety and alignment policies, ultimately contributing to more equitable and trustworthy AI systems.

If you’re interested in discussing my research or anything else, feel free to reach out via email or connect with me on social media. You can find more about my work on this website and in my CV.

I am actively seeking research internship opportunities for summer 2025 and would love to hear from you.

news

Feb 11, 2025	Our paper, OASIS, has been accepted as spotlight ✨ at ICLR 2025.
Jan 22, 2025	Our paper, OASIS Uncovers: High-Quality T2I Models, Same Old Stereotypes, has been accepted at ICLR 2025.
Dec 16, 2024	Happy to Share: I have joined Reality Defender as a Core AI Intern for the Spring 2025 semester.
Jul 11, 2024	Thrilled to announce that I've been awarded the STEAMpower Fellowship for 2024! 🎉
Apr 01, 2024	Our paper on The Dark Side of Dataset Scaling: Evaluating Racial Classification in Multimodal Models has been accepted to FAccT 2024.

selected publications

ICLR’25

OASIS Uncovers: High-Quality T2I Models, Same Old Stereotypes

Sepehr Dehdashtian, Gautam Sreekumar, and Vishnu Boddeti

In International Conference on Learning Representations (ICLR) [spotlight] , May 2025

Bib PDF Website

@inproceedings{dehdashtian2025oasis,
  title = {OASIS Uncovers: High-Quality T2I Models, Same Old Stereotypes},
  author = {Dehdashtian, Sepehr and Sreekumar, Gautam and Boddeti, Vishnu},
  booktitle = {International Conference on Learning Representations (ICLR) <span style="color: red;"><b> [spotlight] </b></span>},
  year = {2025},
  month = may,
  url = {https://openreview.net/forum?id=L6IgkJvcgV},
  dimensions = {true},
}

Preprint

Fairness and Bias Mitigation in Computer Vision: A Survey

Sepehr Dehdashtian*, Ruozhen He*, Yi Li, Guha Balakrishnan, Nuno Vasconcelos, Vicente Ordonez, and Vishnu Naresh Boddeti

Jul 2024

Bib PDF

@misc{dehdashtian2024fairnessbiasmitigationcomputer,
  title = {Fairness and Bias Mitigation in Computer Vision: A Survey},
  author = {Dehdashtian*, Sepehr and He*, Ruozhen and Li, Yi and Balakrishnan, Guha and Vasconcelos, Nuno and Ordonez, Vicente and Boddeti, Vishnu Naresh},
  year = {2024},
  eprint = {2408.02464},
  archiveprefix = {arXiv},
  primaryclass = {cs.CV},
  url = {https://arxiv.org/abs/2408.02464},
  month = jul,
  dimensions = {true},
}

FAccT’24

The Dark Side of Dataset Scaling: Evaluating Racial Classification in Multimodal Models

Abeba Birhane*, Sepehr Dehdashtian*, Vinay Uday Prabhu, and Vishnu Boddeti

In Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency , Jul 2024

Bib PDF Code Poster

@inproceedings{birhane2024thedark,
  title = {The Dark Side of Dataset Scaling: Evaluating Racial Classification in Multimodal Models},
  author = {Birhane*, Abeba and Dehdashtian*, Sepehr and Prabhu, Vinay Uday and Boddeti, Vishnu},
  booktitle = {Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency},
  year = {2024},
  month = jul,
  dimensions = {true},
}

CVPR’24

Utility-Fairness Trade-Offs and How to Find Them

Sepehr Dehdashtian, Bashir Sadeghi, and Vishnu Boddeti

In Conference on Computer Vision and Pattern Recognition (CVPR) 2024 , Jun 2024

Bib PDF Code Poster Slides Website

@inproceedings{dehdashtian2024utilityfairness,
  title = {Utility-Fairness Trade-Offs and How to Find Them},
  author = {Dehdashtian, Sepehr and Sadeghi, Bashir and Boddeti, Vishnu},
  booktitle = {Conference on Computer Vision and Pattern Recognition (CVPR) 2024},
  year = {2024},
  month = jun,
  dimensions = {true},
}

ICLR’24

FairerCLIP: Debiasing CLIP’s Zero-Shot Predictions using Functions in RKHSs

Sepehr Dehdashtian*, Lan Wang*, and Vishnu Boddeti

In International Conference on Learning Representations (ICLR) , May 2024

Bib PDF Code Poster Slides Website

@inproceedings{dehdashtian2024fairvlm,
  title = {FairerCLIP: Debiasing CLIP’s Zero-Shot Predictions using Functions in RKHSs},
  author = {Dehdashtian*, Sepehr and Wang*, Lan and Boddeti, Vishnu},
  booktitle = {International Conference on Learning Representations (ICLR)},
  year = {2024},
  month = may,
  url = {https://openreview.net/forum?id=HXoq9EqR9e},
  dimensions = {true},
}

TMLR

On Characterizing the Trade-off in Invariant Representation Learning

Bashir Sadeghi, Sepehr Dehdashtian, and Vishnu Boddeti

Transactions on Machine Learning Research (TMLR), May 2022

Featured Certification

Bib HTML PDF

@article{sadeghi2022on,
  title = {On Characterizing the Trade-off in Invariant Representation Learning},
  author = {Sadeghi, Bashir and Dehdashtian, Sepehr and Boddeti, Vishnu},
  journal = {Transactions on Machine Learning Research (TMLR)},
  issn = {2835-8856},
  year = {2022},
  url = {https://openreview.net/forum?id=3gfpBR1ncr},
  note = {Featured Certification},
  dimensions = {true},
}