About Me

I’m a Ph.D. student in Computer Science at UC San Diego, advised by Marc Niethammer in the Biomedical Image Analysis Group. My research focuses on computer vision for medical imaging, particularly multimodal registration and segmentation for disease progression prediction and atlas construction. I earned my M.S. in Computer Science from UNC Chapel Hill in 2025 and my B.S. in Computer Engineering from Istanbul Technical University in 2022.

During my summer internships at Meta, I worked on alignment and safety for generative models: GRPO-based methods for LLM counterfactual fairness and output distribution matching at Meta Superintelligence Labs (2026), and training-free concept erasure in text-to-image diffusion models at Meta GenAI (2025). Previously, at United Imaging Intelligence (2024), I developed a self-supervised method for medical image denoising using conditional diffusion models.

I’m looking for full-time research positions starting in summer 2027. Please contact me via bdemir@ucsd.edu if you know of a good fit.

For more details, see my CV.

News

  • [Jun 2026] Returned to Meta as a research intern at Superintelligence Labs.
  • [Jun 2026] Attended CVPR in Denver , where our disease progression paper received the Best Paper Award at CV4Clinic.
  • [Dec 2025] Attended NeurIPS in San Diego .
  • [Aug 2025] Transferred from UNC Chapel Hill to UC San Diego for my Ph.D.
  • [May 2025] Joined Meta GenAI as a research intern, working on concept erasure in diffusion models.
  • [Oct 2024] Attended MICCAI in Marrakesh, Morocco , where our multimodal registration work received the Best Oral Presentation Award at WBIR.
  • [Jul 2024] Attended MIDL in Paris, France .
  • [May 2024] Joined United Imaging Intelligence as a research intern, working on self-supervised medical image denoising.
  • [Aug 2022] Started my Ph.D. in Computer Science at UNC Chapel Hill.
  • [Jun 2022] Graduated from Istanbul Technical University, ranked second in Computer Engineering.

Publications

Registration & Applications

  1. Disease progression framework using longitudinal MRI scans and foundation-model features
    How Useful Are Vision Foundation Model Features for Out-of-the-Box Disease Progression Prediction?
    Başar Demir, Soumitri Chattopadhyay, Thomas Hastings Greer, Boqi Chen, Marc Niethammer, for the Alzheimer’s Disease Neuroimaging Initiative.
    CV4Clinic @ CVPR 2026.
    PaperBest Paper
  2. Global-to-local CT/MR–ultrasound registration framework
    Global to Local Registration: Transferring Pretrained Registration Models to CT/MR–Ultrasound via Anatomical Supervision
    Gabriella d’Albenzio, Mengting Liu, Shixing Ma, Chunna Yang, Yuhao Wei, Shrisharanyan Vasu, Aniketh Vijesh, Xihan Ma, Zachary P. Taylor, Tanmoy Sarkar Pias, Başar Demir, Miklós Gyöngy, Yasin Ceran, Marc Niethammer, Tina Kapur, Stephen Aylward, Gabor Fichtinger, Zhe Min, Mirabela Rusu.
    ASMUS @ MICCAI 2026.
    Paper
  3. Pre- and post-ablation lung CT scans with registered overlays comparing B-spline and uniGradICON registration
    Comparing B-spline and uniGradICON-AI Foundational Deformable Image Registration Methods: Evaluating the Tradeoff Between Accuracy and Speed
    Anirudha Shastri, Başar Demir, Robert Salkin, Arvind Dev, Marc Niethammer, Etay Ziv, Krishna Nand Keshava Murthy.
    SPIE Medical Imaging 2026.
    Paper
  4. Registration guided by intermediate features from a pretrained diffusion model
    Guiding Registration with Emergent Similarity from Pre-Trained Diffusion Models
    Nurislam Tursynbek, Hastings Greer, Başar Demir, Marc Niethammer.
    MICCAI 2025.
    PaperCodeOral
  5. Comparison of uniGradICON and multiGradICON on T1-weighted MRI and mean diffusivity registration
    multiGradICON: A Foundation Model for Multimodal Medical Image Registration
    Başar Demir, Lin Tian, Thomas Hastings Greer, Roland Kwitt, François-Xavier Vialard, Raúl San José Estépar, Sylvain Bouix, Richard Jarrett Rushmore, Ebrahim Ebrahim, Marc Niethammer.
    WBIR @ MICCAI 2024.
    PaperCodeBest Oral
  6. Domain adaptation framework for segmentation-guided multimodal registration
    Multimodal Image Registration Guided by Few Segmentations from One Modality
    Başar Demir, Marc Niethammer.
    MIDL 2024.
    PaperCode

LLM Alignment & Reinforcement Learning

  1. MaD-RL distribution matching compared with correctness-only reinforcement learning and an entropy bonus
    MaD-RL: Matching Distributions for Calibrating LLMs with Reinforcement Learning
    Sourabh Kulkarni*, Ksheeraj Sai Vepuri*, Başar Demir*, Jason Bohrer, Emily Shen, Jianfa Chen, Nan Jiang, Ankit Jain, Harihar Subramanyam, Mannat Singh, Chirag Nagpal*.
    * Equal contribution
    2026.
    PaperMeta AI
  2. Counterfactual invariance method: matched prompts, reward distributions, and symmetrized reward evaluation
    Aligning Language Models Towards Counterfactual Invariance
    Başar Demir*, Emily Shen*, Ksheeraj Sai Vepuri, Sourabh Kulkarni, Harihar Subramanyam, Ankit Jain, Nan Jiang, Chirag Nagpal.
    * Equal contribution
    2026.
    Paper

Segmentation

  1. UniMedMICS segmentation across different anatomical regions and imaging modalities
    UniMedMICS: Generalizing Training-free Universal Multimodal In-Context 3D Medical Image Segmentation
    Soumitri Chattopadhyay, Yinzhu Jin, Başar Demir, Marc Niethammer.
    WACV 2027.
  2. Spatial and semantic contextual prompts fused for multimodal medical segmentation
    Uncertainty-Aware Spatio-Semantic Contextual Prompts for Multimodal Medical Segmentation
    Soumitri Chattopadhyay, Başar Demir, Marc Niethammer.
    MICCAI 2026.
    PaperCodeEarly Acceptance, Top 9%
  3. nnInteractive segmentation results for dilated, eroded and cavity-perturbed visual prompts
    On the Robustness of Foundational 3D Medical Image Segmentation Models Against Imprecise Visual Prompts
    Soumitri Chattopadhyay, Başar Demir, Marc Niethammer.
    ISBI 2026.
    PaperCodeOral

Image Denoising

  1. DiffDenoise framework: diffusion model training, stabilized reverse sampling, and knowledge distillation
    DiffDenoise: Self-Supervised Medical Image Denoising with Conditional Diffusion Models
    Başar Demir*, Yikang Liu*, Xiao Chen, Eric Z. Chen, Lin Zhao, Boris Mailhe, Terrence Chen, Shanhui Sun.
    * Equal contribution
    2025.
    Paper

Software

Lung, knee, and brain registration examples from uniGradICON
uniGradICON / multiGradICON

I wrote the training and fine-tuning code and maintain the repository and Python package.

30K downloads and 200+ GitHub stars. PyPI

Medical image registration results displayed in 3D Slicer
SlicerUniGradICON

I built this 3D Slicer extension for uniGradICON and multiGradICON. You can install it through the Slicer Extension Manager.

Academic Service

Earlier Experience