I focus on representation learning, and how we can find ideal represenetations of data automatically and explicitly.
Deep learning has shown impressive power in learning compact representations of otherwise complex data, such as images or natural language. While geometric methods like contrastive learning have proven to be beneficial for learning good representations, these methods are often impractical at scale and are traded off for hand-engineered hyperparameter tuning, data augmentation etc.
I aim to modernize Riemannian geometry for a computational world, so that geometric-based representation learning methods like contrastive learning can scale in the way modern computing demands. While I currently focus on computer vision and robotics, I believe the "geometric viewpoint" of representation learning can provide both interpretability and power to any area that a neural network may be useful.
Publications & Workshop Presentations
Michael Psenka, Druv Pai, Vishal Raman, Shankar Sastry, Yi Ma. Representation Learning through Manifold Flattening and Reconstruction submitted to SLowDNN. Link to paper.
Nov 2022
Druv Pai, Michael Psenka, Chih-Yuan Chiu, Manxi Wu, Edgar Dobriban, Yi Ma. Pursuit of a discriminative representation for multiple subspaces via sequential games submitted to Journal of the Franklin Institute. Link to paper.
Sept 2022
Xili Dai, Shengbang Tong, Mingyang Li, Ziyang Wu, Michael Psenka, Kwan Ho Ryan Chan, Pengyuan Zhai, Yaodong Yu, Xiaojun Yuan, Heung Yeung Shum, Yi Ma. CTRL: Closed-Loop Transcription to an LDR via Minimaxing Rate Reduction Published in Entropy. Link to paper.
Nov 2021
Michael Psenka and Nicolas Boumal. Second-order optimization for tensors with fixed tensor-train rank. Poster presentation at OPT2020. Link to paper. Link to poster.
Dec 2020
Michael Psenka, Tolga Birdal, Leonidas Guibas. Reconstruction Without Registration. Video presentation at IROS2020 geometric methods workshop. Link to paper. Link to presentation.
Oct 2020
Ryan Arbon, Mohammed Mannan, Michael Psenka, Seyoon Ragavan. A Proof of The Triangular Ashbaugh-Benguria-Payne-Pólya-Weinberger Inequality. To appear in Journal of Spectral Theory. Link to paper.
Sept 2020
Please see my CV for more research projects and further details.