Baris Askin

Pittsburgh, PA, USA

baskin [at] andrew.cmu.edu

Hi, I am Baris 👋

I am a fifth-year PhD candidate in the Department of Electrical and Computer Engineering at Carnegie Mellon University, advised by Gauri Joshi and Carlee Joe-Wong. My research work spans large language models (LLMs), including alignment, interpretability, post-training, and model routing, as well as federated/distributed learning. I have also been a research intern at NVIDIA and a quant research intern at IMC Trading.

I graduated as the valedictorian from my undergraduate studies in the Department of Electrical and Electronics Engineering at Bilkent University, where I worked with Tolga Çukur on deep learning for medical imaging.

news

Sep 2026	3 first-authored papers were accepted to NeurIPS 2026! Emergent and subliminal LLM misalignment [paper] LLM routing in federated settings [paper] Asynchronous FL with data-free knowledge distillation [paper]
Jun 2026	Our work on GRPO with off-policy coordination on public data in FL was accepted to the ICML 2026 Workshop on RL from World Feedback! [paper]
Jun 2026	Started a Quant Research internship at IMC Trading for Summer 2026.
Jan 2026	Our work on internal planning of language models was accepted to ICLR 2026! [paper]
Sep 2025	Our work on LoRA fine-tuning in FL was accepted to NeurIPS 2025! [paper]
Jun 2025	Started a research internship at NVIDIA for Summer 2025, working on asynchronous federated learning algorithms with knowledge distillation.
Jan 2025	Our work on federated multi-objective optimization was accepted to AISTATS 2025! [paper]
Apr 2024	Our work on asynchronous multi-model FL was accepted to UAI 2024! [paper]
Sep 2023	Honored to receive the Ben Cook Presidential Graduate Fellowship at CMU.

selected publications

NeurIPS ICML FoGen

Emergent and Subliminal Misalignment Through the Lens of Data-Mediated Transfer

B. Askin^*, M. Ustaomeroglu^*, A. Nayak^*, G. Joshi, G. Qu, and C. Joe-Wong

In Conference on Neural Information Processing Systems (NeurIPS), 2026

arXiv Bib

@misc{askin2026emergentsubliminalmisalignmentlens,
      title={Emergent and Subliminal Misalignment Through the Lens of Data-Mediated Transfer},
      author={Baris Askin and Muhammed Ustaomeroglu and Anupam Nayak and Gauri Joshi and Guannan Qu and Carlee Joe-Wong},
      year={2026},
      eprint={2605.12798},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2605.12798},
}

ICLR

Internal Planning in Language Models: Characterizing Horizon and Branch Awareness

M. Ustaomeroglu^*, B. Askin^*, G. Joshi, C. Joe-Wong, and G. Qu

In International Conference on Learning Representations (ICLR), 2026

arXiv Bib Poster

@inproceedings{
ustaomeroglu2026internal,
title={Internal Planning in Language Models: Characterizing Horizon and Branch Awareness},
author={Muhammed Ustaomeroglu and Baris Askin and Gauri Joshi and Carlee Joe-Wong and Guannan Qu},
booktitle={The Fourteenth International Conference on Learning Representations},
year={2026},
url={https://openreview.net/forum?id=dqGWQdFdTC}
}

NeurIPS

Federate the Router: Learning LM Routers with Sparse and Decentralized Evaluations

B. Askin^*, S. Patel^*, A. Nayak^*, A. Vigano, J. Woo, G. Joshi, and C. Joe-Wong

In Conference on Neural Information Processing Systems (NeurIPS), 2026

arXiv Bib PROJECT PAGE Code

@misc{askin2026federaterouterlearninglanguage,
      title={Federate the Router: Learning Language Model Routers with Sparse and Decentralized Evaluations},
      author={Baris Askin and Shivam Patel and Anupam Nayak and Andrea Vigano and Jiin Woo and Gauri Joshi and Carlee Joe-Wong},
      year={2026},
      eprint={2601.22318},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2601.22318},
}

NeurIPS

Reviving Stale Updates: Data-Free Knowledge Distillation for Asynchronous Federated Learning

B. Askin, H. R. Roth, Z. Sun, C. Joe-Wong, G. Joshi, and Z. Xu

In Conference on Neural Information Processing Systems (NeurIPS), 2026

arXiv Bib Code

@misc{askin2026revivingstaleupdatesdatafree,
      title={Reviving Stale Updates: Data-Free Knowledge Distillation for Asynchronous Federated Learning},
      author={Baris Askin and Holger R. Roth and Zhenyu Sun and Carlee Joe-Wong and Gauri Joshi and Ziyue Xu},
      year={2026},
      eprint={2511.00655},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2511.00655},
}

ICML RLxF ICML DEMO

PubSwap: Public-Data Off-Policy Coordination for Federated RLVR

A. Nayak^*, B. Askin^*, M. Ustaomeroglu, C. Joe-Wong, and G. Joshi

2026, Initial version at ICML 2026 Workshop on RL from World Feedback and ICML 2026 Decision-Making from Offline Datasets to Online Adaptation: Black-Box Optimization to Reinforcement Learning

arXiv Bib

@misc{nayak2026pubswappublicdataoffpolicycoordination,
      title={PubSwap: Public-Data Off-Policy Coordination for Federated RLVR},
      author={Anupam Nayak and Baris Askin and Muhammed Ustaomeroglu and Carlee Joe-Wong and Gauri Joshi},
      year={2026},
      eprint={2604.12160},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2604.12160},
}

NeurIPS

Ravan: Multi-Head Low-Rank Adaptation for Federated Fine-Tuning

A. Raje, B. Askin, D. Jhunjhunwala, and G. Joshi

In Conference on Neural Information Processing Systems (NeurIPS), 2025

arXiv Bib

@article{raje2026ravan,
  title={Ravan: Multi-head low-rank adaptation for federated fine-tuning},
  author={Raje, Arian and Askin, Baris and Jhunjhunwala, Divyansh and Joshi, Gauri},
  journal={Advances in Neural Information Processing Systems},
  volume={38},
  pages={48564--48592},
  year={2026}
}

AISTATS

Federated Communication-Efficient Multi-Objective Optimization

B. Askin, P. Sharma, G. Joshi, and C. Joe-Wong

In Proceedings of the 28th International Conference on Artificial Intelligence and Statistics (AISTATS), May 2025

arXiv Bib Code Poster

@inproceedings{askin2025fedcmoo,
  title = {Federated Communication-Efficient Multi-Objective Optimization},
  author = {Askin, B. and Sharma, P. and Joshi, G. and Joe-Wong, C.},
  booktitle = {Proceedings of the 28th International Conference on Artificial Intelligence and Statistics (AISTATS)},
  pages = {4627--4635},
  year = {2025},
  volume = {258},
  series = {Proceedings of Machine Learning Research},
  month = may,
  publisher = {PMLR},
  url = {https://proceedings.mlr.press/v258/askin25a.html},
}

UAI

FedAST: Federated Asynchronous Simultaneous Training

B. Askin, P. Sharma, C. Joe-Wong, and G. Joshi

In Proceedings of the 40th Conference on Uncertainty in Artificial Intelligence (UAI), Jul 2024

arXiv Bib Code Poster

@inproceedings{askin2024fedast,
  title = {{FedAST}: Federated Asynchronous Simultaneous Training},
  author = {Askin, B. and Sharma, P. and Joe-Wong, C. and Joshi, G.},
  booktitle = {Proceedings of the 40th Conference on Uncertainty in Artificial Intelligence (UAI)},
  pages = {138--172},
  year = {2024},
  volume = {244},
  series = {Proceedings of Machine Learning Research},
  month = jul,
  publisher = {PMLR},
  url = {https://proceedings.mlr.press/v244/askin24a.html},
}

Preprint

Aligned Data Can Induce Misalignment via Context Confusion

Y. Bakman, D. N. Yaldiz, B. Askin, S. Roy, M. Ziyadi, S. Avestimehr, and S. P. Karimireddy

2026

arXiv Bib

@misc{bakman2026context,
  title = {Aligned Data Can Induce Misalignment via Context Confusion},
  author = {Bakman, Y. and Yaldiz, D. N. and Askin, B. and Roy, S. and Ziyadi, M. and Avestimehr, S. and Karimireddy, S. P.},
  year = {2026},
  url = {https://arxiv.org/abs/2609.38379},
}

ICML Position

Position: Federated Learning is a Lens towards a Democratized Future for the Scaling Law Era

H. H. Jiang, B. Askin, G. Joshi, and C. Joe-Wong

In International Conference on Machine Learning (ICML) Position Track, 2026

Link Bib

@inproceedings{
jiang2026position,
title={Position: Federated Learning is a Lens towards a Democratized Future for the Scaling Law Era},
author={Harry H. Jiang and Baris Askin and Gauri Joshi and Carlee Joe-Wong},
booktitle={Forty-third International Conference on Machine Learning Position Paper Track},
year={2026},
url={https://openreview.net/forum?id=td1fQFeIiI}
}