S Divakar Bhat

Home Research Publications Projects Writing CV / Resume Contact

CV / Resume

S Divakar Bhat portrait

S Divakar Bhat

AI Researcher, Honda R&D Japan

PhD Student, The University of Tokyo

sdivakarbhat@gmail.com | +81 80-4919-2378

Website LinkedIn GitHub Scholar

Research Areas

Reliable and adaptive computer vision, test-time adaptation, continual learning, long-tailed learning, CLIP/VLM adaptation, and robust AD/ADAS perception.

Education

  • PhD, Department of Information Science and Technology, The University of Tokyo, Tokyo, Japan (Oct 2025 - Oct 2028)
    Advisor: Prof. Toshihiko Yamasaki
  • M.Tech, Electrical Engineering (Control and Computing), IIT Bombay, Mumbai, India (Jul 2018 - Aug 2021)
    Thesis: You Never Stop Learning: Exploring Continual Learning in Visual Recognition
    Advisor: Prof. Subhasis Chaudhuri | CGPA: 9.42/10
  • B.Tech, Electrical Engineering, Govt. Model Engineering College, Cochin, India (Aug 2013 - May 2017)
    CGPA: 8.76/10

Work Experience

  • AI Researcher, Honda Innovation Lab, Honda R&D Co., Ltd., Tokyo, Japan (Oct 2021 - Present)
    • Research and development of road/free-space segmentation models for unstructured environments.
    • Quantization and deployment of deep learning models on edge platforms.
    • Camera-based risk minimization algorithms for safer driving scenarios.
    • On-vehicle testing of perception algorithms under varied target conditions.
  • Research Assistant, India-Trento Program for Advanced Research Phase-IV (Jul 2018 - Jun 2021)
    • Remote sensing analysis with optical and radar imagery for dynamic earth-process monitoring.
    • Curriculum learning methods to improve incremental classification accuracy and convergence.
    • Few-shot incremental classification with continual learning constraints.

Selected Publications

  • AdaPrior: Bayesian-Inspired Adaptive Prior Correction for Long-Tailed Continual Learning
    S Divakar Bhat, Amit Popat More, Mudit Soni, Bhuvan Aggarwal
    CVPR 2026 Main Conference (Highlight)
    Paper

  • Consistent Yet Wrong: Evidence Insensitivity in Spatial Vision-Language Models
    S Divakar Bhat, Toshihiko Yamasaki
    CVPR 2026 Workshop on Multimodal Learning and Applications (MULA), Oral + Poster
    Paper

  • Seeing What’s Not There: Negation Understanding Needs More Than Training
    B. Aggarwal, A. More, M. Soni, S. D. Bhat
    ICLR 2026
    Search

Show more publications - **Prior2Posterior: Model Prior Correction for Long-Tailed Learning** S. D. Bhat, A. More, M. Soni, S. Agrawal WACV 2025 (Oral) [Paper](https://doi.org/10.1109/WACV61041.2025.00133) - **PC-GZSL: Prior Correction for Generalized Zero Shot Learning** S. D. Bhat, A. More, M. Soni, B. Aggarwal WACV 2025 [Paper](https://doi.org/10.1109/WACV61041.2025.00697) - **Model Ensemble to Fuse Geometric and Learning Solutions for Camera Rotation Estimation** B. Aggarwal, A. More, S. D. Bhat, M. Soni ICCVW 2025 [Paper](https://doi.org/10.1109/ICCVW69036.2025.00023) - **Robust Loss Function for Class Imbalanced Semantic Segmentation and Image Classification** S. D. Bhat, A. More, M. Soni, Y. Yasui IFAC World Congress 2023 [Paper](https://doi.org/10.1016/j.ifacol.2023.10.320) - **CILEA-NET: A Curriculum-driven Incremental Learning Network for Remote Sensing Image Classification** IEEE JSTARS, 2021 [Paper](https://ieeexplore.ieee.org/abstract/document/9442875) - **SemGIF: A Semantics Guided Incremental Few-shot Learning Framework with Generative Replay** BMVC, 2021 [Paper](https://www.bmvc2021-virtualconference.com/conference/papers/paper_0673.html) - **Directed Variational Cross-encoder Network for Few-shot Multi-image Co-segmentation** ICPR, 2021 [Paper](https://ieeexplore.ieee.org/abstract/document/9412967)

Selected Talks and Presentations

  • CVPR 2026 Main Conference Highlight Poster: AdaPrior (Video)
  • CVPRW 2026 MULA Oral and Poster: Consistent Yet Wrong

Awards and Achievements

  • Best M.Tech Thesis Award, Department of Electrical Engineering, IIT Bombay (2021)
  • Finalist, INAE Innovative Student Projects Award (2021)
  • GATE 2018: 99.3 percentile (out of 121,383 candidates)

Patents

  • S. D. Bhat and A. More, “Learning device, learning method, and storage medium”, US Patent App. US20240265678A1 (2024)
  • 8 additional patents filed worldwide

Skills

  • PyTorch, Python, OpenCV (Python), TensorFlow Lite, C/C++, C#, Android Studio, Bash, Linux system administration

Languages

  • English (Fluent)
  • Japanese (Intermediate, N2)
  • Hindi (Fluent)
  • Konkani (Mother tongue)
  • Malayalam (Fluent)