We experience how the world works and learn internal world models to solve new problems quickly.
How does the brain learn these internal representations and use them to predict the next state, best action and utility?
It has also been shown that these world models are distorted in those suffering from neurological disorders.
If we can rectify these world models, can we solve neurological disorders?
To explore these questions, I develop artificially intelligent systems, grounded to theory and experiments,
to understand learning computations for intelligence, and how it might fail.
I hope to develop technologies to improve learning outcomes and alleviate learning disabilities.
Currently, I am a postdoctoral fellow at the Harvard Machine Learning Foundations Group,
working on Neuroscience inspired Reinforcement Learning. I am advised by
Cengiz Pehlevan,
Demba Ba,
Lucas Janson and
Boaz Barak.
Previously, I was a research scientist at the Centre for Frontier AI Research (CFAR), A*STAR developing
vision-language reasoning datasets and models. I was advised by
Cheston Tan.
In 2022, I completed my Ph.D. in Computational Neuroscience at the National University of Singapore (NUS)
under the Integrative Sciences and Engineering Programme (ISEP).
My doctoral thesis was to develop a biologically plausible reinforcement learning agent that learned
new paired associations after a single example.
I was co-advised by Andrew Tan and
Shih-Cheng Yen and collaborated with
Camilo Libedinsky.
I completed my B.Sc in Life Sciences in 2017 at the National University of Singapore where
I worked on Brain-Computer Interfaces to control wheelchair using either human EEG or macaque intracortical spike data.
Besides research, I co-founded Nugen.ai that hopes to characterize
affective states during learning to aid parents and teachers in personalizing education.
I also love to tour either on my motorcycle or backpacking,
and be involved in the arts scene.
2023 - 2024: Postdoctoral Fellow (Theory of representational learning), Harvard University School of Engineering and Applied Sciences
2022 - 2023: Research Scientist (Vision-Language reasoning), Centre for Frontier AI Research, A*STAR
2018 - 2022: Ph.D. Candidature (Biologically plausible One-shot learning), National Unviersity of Singapore
2019: Center for Brains, Minds and Machines (CBMM) Summer School 2019, Massachusetts Institute of Technology
2017 - 2018: Research Engineer (Social roles & relationships recognition), A*STAR Artificial Intellignce Initiative (A*AI)
2013 - 2017: B.Sc. Life Sciences + University Scholars Programme (USP) + Special Programme in Science (SPS)
Lin, Z.*, Azaman, H.*, Kumar, M. G., Tan, C. (2023). Composing Word Groups using Visually Grounded Reinforcement Learning. (Under review.)
Guertler, L., Kumar, M. G., Tan, C. (2023). NoiseOut: Learning to Gate Improves Robustness in Deep Neural Networks. (Under review.)
Lee, C.*, Kumar, M. G.*, Tan, C. (2023). DetermiNet: A Large-Scale Diagnostic Dataset for Complex Visually-Grounded Referencing using Determiners. International Conference on Computer Vision, ICCV 2023
Kumar, M. G., Tan, C., Libedinsky, C., Yen, S. C., & Tan, A. Y. Y. (2023). One-shot learning of paired association navigation with biologically plausible schemas. arXiv preprint arXiv:2106.03580. https://arxiv.org/abs/2106.03580 [GitHub]
Kumar M. G. (2023). Biologically plausible computations underlying one-shot learning of paired associations . Scholarbank@NUS. https://scholarbank.nus.edu.sg/handle/10635/238485
Kumar, M. G., Tan, C., Libedinsky, C., Yen, S. C., & Tan, A. Y. Y. (2022). A nonlinear hidden layer enables actor-critic agents to learn multiple paired association navigation. Cerebral Cortex. https://doi.org/10.1093/cercor/bhab456 [GitHub]
Kumar M. G., Kai Keng Ang, Rosa Q. So. (2017). Reject Option to reduce False Detection Rates for EEG-Motor Imagery based BCI. In Engineering in Medicine and Biology Society, EMBC 2017. 39th Annual International Conference of the IEEE. https://doi.org/10.1109/EMBC.2017.8037479
Look forward to connecting with you!