Computer Science PhD Candidate · Northeastern University

Gerard L. Donahue

I work on temporal machine learning, representation learning, and video understanding, with a focus on learning robust and transferable representations from complex temporal data.

Email Google Scholar GitHub LinkedIn CV
Gerard L. Donahue

Personal Summary

I am a Computer Science PhD Candidate at Northeastern University advised by Prof. Ehsan Elhamifar. My research develops methods for learning from video and other temporal data, including progress prediction, video alignment, action-centric video generation, and real-time reasoning.

I am interested in building adaptive, context-aware AI systems that can understand how actions unfold over time. Before Northeastern, I completed my B.S. in Computer Science at the University of New Hampshire, where I worked on inverse reinforcement learning.

Computer Vision Video Understanding Procedural Learning Video Generation Diffusion Models Vision-Language Modeling Multimodal Representation Learning

Recent News

  1. Continuing PhD research on action-centric video understanding and generation.

    2026

    Current projects focus on fine-grained progress control for action-centric video generation and action-aligned video representations.

  2. Memory-efficient temporal action segmentation project submitted and patent filed.

    2025

    At Honda Research Institute, I led work on adaptive capacity allocation and compressed video replay for memory-constrained procedural video streams.

  3. Achieved PhD candidacy and completed M.S. in Computer Science.

    2024

    I achieved PhD candidacy at Northeastern and completed my M.S. in Computer Science.

  4. Published video alignment work at CVPR 2024.

    2024

    Our paper on self-supervised video alignment for activity progress prediction appeared at CVPR 2024 and was also accepted to the CVPR LPVL workshop.

Publications

  1. Learning to Predict Activity Progress by Self-Supervised Video Alignment

    CVPR 2024

    Gerard Donahue and Ehsan Elhamifar

    Introduces a self-supervised approach for aligning in-the-wild videos and predicting activity progress.

    Paper LPVL Workshop Code

  2. Unbiased Efficient Feature Counts for Inverse RL

    NeurIPS Workshop 2021

    Gerard Donahue, Brendan Crowe, Marek Petrik, Daniel S. Brown, and collaborators

    Develops efficient feature-count estimation methods for inverse reinforcement learning with limited long demonstrations.

    PDF

Experience

  1. Northeastern University

    Aug 2022 – Present

    PhD Research Assistant · Boston, MA

    • Advised by Prof. Ehsan Elhamifar on temporal machine learning and video understanding.
    • Developing methods for action-centric video generation, progress localization, and action-aligned video representations.
    • Published fine-grained video alignment research at CVPR 2024.

  2. Honda Research Institute

    May 2022 – Jan 2023

    PhD Research Intern · San Jose, CA

    • Led a project on memory-efficient temporal action segmentation, resulting in a NeurIPS 2025 submission and first-author patent filing.
    • Proposed adaptive capacity allocation based on frame-level representational complexity.
    • Extended the approach to compressed video replay for continual learning on procedural video streams.

  3. Ultra Intelligence and Communications

    May 2022 – Jan 2023

    Research Scientist Intern · Remote

    • Proposed reinforcement and imitation learning methods for safe autonomous defense agents.
    • Collaborated with Texas A&M researchers on predictive Bayesian modeling for target localization.

  4. University of New Hampshire

    Jan 2021 – May 2022

    Undergraduate Research Assistant · Durham, NH

    • Worked with Prof. Marek Petrik on inverse reinforcement learning.
    • Proposed efficient feature-counting techniques that led to a NeurIPS 2021 workshop paper.

  5. Project Happy, 501c3

    Jul 2020 – Nov 2024

    Co-founder & CTO · Boston, MA

    • Built and scaled a full-stack iOS/Android app and backend platform.
    • Led a software team and helped deliver a community service platform used in partnership with the New Hampshire Department of Education.

  6. Intel Corporation

    May 2020 – Jan 2021

    Software Engineering Intern · Remote

    • Contributed to validation of next-generation Intel Optane SSDs under Agile development practices.

  7. UNH Interoperability Laboratory

    Jun 2018 – May 2020

    Lead Software Developer · Durham, NH

    • Led the NVMe testing group and modernized legacy Bash frameworks into object-oriented Python.
    • Delivered conformance testing tools adopted by industry vendors.