Yoshimasa Kubo
Ph.D., Computer Science, Dalhousie University, Halifax, Nova Scotia, Canada, 2019
Dissertation: Learning Stochastic Weight Masking to Resist Adversarial Attacks
Supervisors: Thomas Trappenberg and Sageev OoreM.Sc., System Information, Future University-Hakodate, Hokkaido, Japan, 2014
Master’s thesis: Quick Learning Algorithm of Class-Proximity SOM Using Index Layer for Cluster Visualization
Supervisor: Hitoshi MatsubaraB.Sc., Computer Science – Applied (Minor: Mathematics), University of Central Oklahoma, Oklahoma, USA, 2006
Postdoctoral Fellow
University of California San Diego, San Diego, USA — May 2024–August 2025
Supervisor: Maxim Bazhenov
Researched biologically inspired neural networks and neural mechanisms of sleep replay consolidation.
Developed recurrent neural network models (including insect-inspired RNNs) for complex time-series analysis.
Prepared and authored manuscripts for leading journals and conferences.
Postdoctoral Fellow
University of Lethbridge, Lethbridge, Canada — January 2020–September 2023
Supervisor: Artur Luczak
Developed biologically inspired neural networks for image classification and reinforcement learning.
Supervised and mentored students implementing and optimizing machine learning models.
Prepared documentation and user guides for Compute Canada / cluster systems.
Research Intern
Okinawa Institute of Science and Technology Graduate University, Okinawa, Japan — April 2013–September 2013
Supervisor: Kenji Doya
Researched self-organizing maps with reinforcement learning.
Developed a self-organizing map in an RL setting and reported research progress regularly.
Software Engineer
Mitsubishi Research Institute DCS Co., Ltd., Tokyo, Japan — April 2008–February 2012
Designed, developed, and tested physical distribution systems.
Evaluated and integrated new technologies into system proposals and designs.
Developed PL/SQL stored procedures for system analysis.
Programmer
Pacific Software Publishing, Washington, USA — January 2007–April 2007
Programmed and tested web-based systems.
Biologically Inspired Neural Networks
Development of neural network models inspired by brain mechanisms, including dendritic computation, recurrent dynamics, and sleep-related learning processes. Focus on designing learning algorithms that are both effective and biologically plausible.
Equilibrium Propagation (EP) and Alternatives to Backpropagation
Advancing energy-based learning frameworks such as Equilibrium Propagation and holomorphic EP. Research includes recurrent EP models, stability improvements, and integrating EP into supervised, sequential, and reinforcement learning settings.
Continual Learning and Sleep Replay Consolidation (SRC)
Designing sleep-inspired mechanisms—such as replay consolidation and awake rehearsal—to mitigate catastrophic forgetting. Creating biologically grounded continual learning algorithms that remain stable over long task sequences.
Recurrent Neural Networks and Dynamical Models
Developing biologically plausible recurrent neural architectures (MRNN-EP, Wave-RNN, dendritic RNNs) for sequential data, time series, and sensory processing tasks. Investigating recurrent dynamics and activity patterns that resemble cortical computation.
Reinforcement Learning with Biologically Plausible Updates
Integrating EP-based learning into actor–critic frameworks and reinforcement learning environments, bridging the gap between biological plausibility and practical control tasks.
Prospective Master’s Students
Due to the high volume of supervision inquiries, I no longer respond to prospective student emails directly.
If you are interested in Master’s supervision, please complete the following form:
Please note that I will only review submissions that meet the criteria described on this page. Generic inquiries, requests for PhD or postdoctoral positions, incomplete submissions, or requests outside my research area will not receive a response.
Please do not contact me by email regarding graduate supervision. Supervision requests sent by email instead of through the form will not receive a response.