Adarsh Kappiyath

World modeling and representation learning

prof_pic.jpg

26 BBO2, Alan Turing Building

University of Surrey

United Kingdom, GU2 7XH

Hi, I’m Adarsh Kappiyath, a PhD researcher at the University of Surrey with research interests in world modeling and representation learning. My primary supervisor is Dr. Sara Atito, and I am also supervised by Dr. Xiatian Zhu.

I am interested in how models can learn useful representations without explicit labels, and how those representations can support prediction and an understanding of the world. My current work examines the reliability of JEPA representations and how to preserve meaningful information while improving stability and generalization under unseen conditions.

Research directions

  • JEPA and predictive representations: studying representation reliability and corrective attribute recovery in joint embedding predictive architectures. My current project, CARE-JEPA, explores Corrective Attribute Recovery via Error.
  • Self-supervised learning: understanding and preventing representation collapse. I investigate score-based entropy maximization as a way to preserve informative representations during learning.
  • World modeling: exploring how predictive representations can capture useful semantic information and causal structure, with robustness under changing conditions as a central research interest.

I am actively seeking internships and research collaborations in world modeling and representation learning. Contact me at kadarsh22@gmail.com, or see my CV for publications, experience, and current projects.

news

Nov 17, 2026 📢 I am co-chairing the BMVA Symposium on World Models on 18 November 2026 in London!
Nov 17, 2026 🚀 Our new work “Self-Supervised Collapse Prevention with Score-Based Entropy Maximization” is now available on arXiv! We introduce Entropy-Based Regularization (EBReg), which uses a learned latent score model to prevent representation collapse in self-supervised learning. Explore the project page and alphaXiv discussion.
Nov 17, 2026 🎉 Our paper “MAPLE: Masked Adapter Prototype Learning for OOD generalization” has been accepted at EMNLP 2026!
Aug 29, 2025 🎊 Our Project has been awarded 144,000 GPU hours on NVIDIA A100 hardware through EuroHPC! We gratefully acknowledge the support of the “EuroHPC” in our scientific endeavours.
Jan 29, 2025 🎊 Our paper “SEBRA: Debiasing Through Self-Guided Bias Ranking” has been accepted at ICLR 2025!

selected publications

  1. ICLR2025
    SEBRA: Debiasing through Self-Guided Bias Ranking
    Adarsh Kappiyath, Abhra Chaudhuri, Ajay Jaiswal, and 4 more authors
    The Thirteenth International Conference on Learning Representations, 2025
  2. NeurIPS2024
    DeNetDM: Debiasing by Network Depth Modulation
    Silpa Vadakkeeveetil Sreelatha, Adarsh Kappiyath, Abhra Chaudhuri, and 1 more author
    The Thirty-eighth Annual Conference on Neural Information Processing Systems, 2024
  3. AAAI2022
    Self-Supervised Enhancement of Latent Discovery in GANs
    Adarsh Kappiyath, Silpa Vadakkeeveetil Sreelatha, and S. Sumitra
    Thirty-Sixth AAAI Conference on Artificial Intelligence, AAAI 2022, 2022