Bimsara Pathiraja

I am a fourth-year PhD student in Computer Science at Arizona State University, advised by Chitta Baral and Yezhou Yang. I work on generative models for images, video and speech, mostly on steering them at inference time without retraining.

I have spent two summers as an Applied Scientist Intern at Amazon, working on expressive speech generation and on object detection and tracking. Before ASU I was a research assistant at MBZUAI with Muhammad Haris Khan, and before that I studied Electronic and Telecommunication Engineering at the University of Moratuwa, Sri Lanka.

Email/CV/Google Scholar/GitHub/X

Bimsara Pathiraja

News

Publications

* denotes joint first authors

  1. Student, DFS and teacher outputs for an org chart and a box of six donuts
    Discriminant Flow Steering: Training-Free Capability Transfer in Flow-Matching Models
    Bimsara Pathiraja, Agneet Chatterjee, Chitta Baral, Yezhou Yang
    Under review

    Moves a capability from a black-box teacher into an open flow-matching model using about twenty of the teacher’s images, with no training and no extra cost at inference.

  2. Edits applied to one person or animal among several, chosen by a referring expression
    RefEdit: A Benchmark and Method for Improving Instruction-based Image Editing Model on Referring Expressions
    Bimsara Pathiraja*, Maitreya Patel*, Shivam Singh, Yezhou Yang, Chitta Baral
    ICCV 2025
    Paper / Project / Code / Data
  3. Pipeline: legal questions, LLM reasoning chains, human error analysis and an LLM-based auto-evaluator
    Investigating the Shortcomings of LLMs in Step-by-Step Legal Reasoning
    Venkatesh Mishra*, Bimsara Pathiraja*, Mihir Parmar, Sat Chidananda, Jayanth Srinivasa, Gaowen Liu, Ali Payani, Chitta Baral
    NAACL 2025 Findings
    Paper / Code
  4. A preferred and a less preferred image for the prompt Soon written in smoke
    Dual Caption Preference Optimization for Diffusion Models
    Amir Saeidi, Yiran Luo, Agneet Chatterjee, Shamanthak Hegde, Bimsara Pathiraja, Yezhou Yang, Chitta Baral
    TMLR 2025
    Paper
  5. Bar chart: lower detection calibration error than the baseline
    Multiclass Confidence and Localization Calibration for Object Detection
    Bimsara Pathiraja, Malitha Gunawardhana, Muhammad Haris Khan
    CVPR 2023
    Paper / Code / U.S. patent application
  6. Diagram: a generator writes text carrying a hidden signal, a grader estimates the signal, and mutual information scores the match
    ExpressivityBench: Can LLMs Communicate Implicitly?
    Joshua Tint, Som Sagar, Aditya Taparia, Kelly Raines, Bimsara Pathiraja, Caleb Liu, Ransalu Senanayake
    Findings of EACL 2026 (earlier version at the NeurIPS 2024 Workshop on Behavioral ML)
    Paper
  7. Pedestrian detections in a driving simulator at fog levels 0, 50 and 100, with false positives in heavy fog
    Fairness in Autonomous Driving: Towards Understanding Confounding Factors in Object Detection under Challenging Weather
    Bimsara Pathiraja, Caleb Liu, Ransalu Senanayake
    CVPR 2024 Workshop on Data-Driven Autonomous Driving Simulation
    Paper / Project / Code
  8. Predicted vehicle trajectories on map views
    Class-Aware Attention for Multimodal Trajectory Prediction
    Bimsara Pathiraja, Shehan Munasinghe, Malshan Ranawella, Maleesha De Silva, Ranga Rodrigo, Peshala Jayasekara
    arXiv 2022
    Paper

Outside research

I dance Latin ballroom, cook, work out, and read a lot of non-fiction.

Books I recommend
  • Sapiens Yuval Noah Harari
  • 21 Lessons for the 21st Century Yuval Noah Harari
  • Man’s Search for Meaning Viktor E. Frankl
  • A Briefer History of Time Stephen Hawking
  • Zero to One Peter Thiel
  • Permanent Record Edward Snowden
  • Becoming Michelle Obama
  • 12 Rules for Life Jordan Peterson
  • The Art of War Sun Tzu
  • The Monk Who Sold His Ferrari Robin Sharma
  • The Subtle Art of Not Giving a F*ck Mark Manson
  • Hyperbole and a Half Allie Brosh
  • The Hitchhiker’s Guide to the Galaxy Douglas Adams
  • Crime and Punishment Fyodor Dostoevsky
  • The Archer Paulo Coelho