Marc Finzi
You can contact me at maf820@nyu.edu
I'm a Postdoc at CMU working with Zico Kolter . Previously I completed my PhD in Computer Science at NYU, working with Professor Andrew Wilson. I study deep learning and in my research I have been developing new methods for incorporating inductive biases into deep learning models, understanding the impacts these biases have on learning and generalization, and developing ML models for scientific applications. I am also very interested in leveraging large pretrained models for tasks other than those that they were intended for, making use of the more generic inductive biases that can be induced by sufficient data, as well as overlapping structure.
I did my BS in Physics at Harvey Mudd College and and recieved a masters in ORIE at Cornell in 2017-2019.
Publications
-
Non-vacuous generalization bounds for large language models
Arxiv, 2024 [ArXiv] -
CoLA: Exploiting Compositional Structure for Automatic and Efficient Numerical Linear Algebra
NeurIPS, 2023 [ArXiv, Code, Docs, ] -
Large language models are zero-shot time series forecasters
NeurIPS, 2023 [ArXiv, Code] -
User-defined event sampling and uncertainty quantification in diffusion models for physical dynamical systems
ICML, 2023 [ArXiv, Code] -
A Stable and Scalable Method for Solving Initial Value PDEs with Neural
Networks
ICLR, 2023 [ArXiv, Code] -
The Lie Derivative for Measuring Learned Equivariance
ICLR (Oral), 2023 [ArXiv, Code] -
PAC-Bayes Compression Bounds So Tight That They Can Explain Generalization
NeurIPS, 2022 [ArXiv, Code] -
Deconstructing the Inductive Biases of Hamiltonian Neural Networks
ICLR (Spotlight), 2022 [ArXiv, Code] -
Residual Pathway Priors for Soft Equivariance Constraints
NeurIPS, 2021 [ArXiv, Code] -
A Practical Method for Constructing Equivariant Multilayer Perceptrons for Arbitrary
Matrix Groups
ICML (Long Oral), 2021 [ArXiv, Code, Documentation] -
SKIing on Simplices: Kernel Interpolation on the Permutohedral Lattice for Scalable
Gaussian Processes
ICML (Long Oral), 2021 [ArXiv, Code] -
Probabilistic Numeric Convolutional Neural Networks
ICLR, 2021 [ArXiv, Code] -
Simplifying Hamiltonian and Lagrangian Neural Networks via Explicit Constraints
NeurIPS (Spotlight), 2020 [ArXiv, Code] -
Learning Invariances in Neural Networks from Training Data
NeurIPS, 2020 [ArXiv, Code] -
Generalizing Convolutional Neural Networks for Equivariance to Lie Groups on Arbitrary
Continuous Data
ICML, 2020 [ArXiv, Code] -
Semi-Supervised Learning with Normalizing Flows
ICML, 2020 [ArXiv, Code] -
There Are Many Consistent Explanations of Unlabeled Data: Why You Should
Average
ICLR, 2019 [ArXiv, Code]