Scholar
Andres Potapczynski
Google Scholar ID: Gl1cs3AAAAAJ
New York University
Citations & Impact
Citations
291
H-index
8
i10-index
8
Publications
17
Co-authors
11
list available
Contact
Publications
Resume (English only)
Academic Achievements
- Paper 'Training Flexible Models of Genetic Variant Effects from Functional Annotations using Accelerated Linear Algebra' accepted at ICML 2025. Proposes DeepWAS, a neural model using fast linear algebra to predict variant effects, outperforming LD score regression.
- Paper 'Customizing the Inductive Biases of Softmax Attention using Structured Matrices' accepted at ICML 2025. Introduces new attention scoring functions based on high-rank structured matrices (e.g., BTT, MLR).
- Paper 'Effectively Leveraging Exogenous Information across Neural Forecasters' accepted at NeurIPS TSALM 2024. Develops a decoder method applicable to NBEATS, NHITS, PatchTST, and S4, significantly improving forecasting performance.
- Paper 'Searching for Efficient Linear Layers over a Continuous Space of Structured Matrices' accepted at NeurIPS 2024. Introduces a continuous parameterization over structured matrices expressible as Einsums and derives key insights on compute-optimal scaling.
- Conducted systematic studies on replacing dense layers with sub-quadratic structured alternatives for improved efficiency.
