Volatility Based Kernels and Moving Average Means for Accurate Forecasting with Gaussian Processes
ICML, 2022
We adapt stochastic volatility models for use in Gaussian process forecasting with finance and climatology applications.
Download here
ICML, 2022
We adapt stochastic volatility models for use in Gaussian process forecasting with finance and climatology applications.
Download here
ICML, 2022
A careful look at the role of the marginal likelihood and its role in generalization and model selection in modern machine learning.
Download here
ICML, 2021
We introduce a way to discover volumes of low loss in the parameter space of neural networks, as well as a method to form ensembles of models taken from these low loss volumes.
Download here
Neurips, 2020
We introduce a method to learn levels of invariance to augmentation in neural networks using only training data.
Download here
Pre-print, 2020
We show that many of the properties of generalization become understandable when viewed through the lens of effective dimensionality, which measures the dimensionality of the parameter space determined by the data. We relate effective dimensionality to posterior contraction in Bayesian deep learning, model selection, width-depth tradeoffs, double descent, and functional diversity in loss surfaces, leading to a richer understanding of the interplay between parameters and functions in deep models.
Download here
Neurips, 2019
We propose a model for determining distributions over covariance functions for use in conjunction with Gaussian processes.
Download here
Applied Mathematics Graduate Theses & Dissertations., 2018
Master’s Thesis outlining the development of a robust and accurate statistical model for estimating precipitation distributions at arbitrary locations. Advisor: Will Kleiber
Download here
Published:
Oral presentation of “Learning Invariances in Neural Networks” at Neurips 2020.
Published:
Oral presentation of “Loss Surface Simplexes for Mode Connecting Volumes and Fast Ensembles” at ICML 2021.
Published:
Oral presentation of “Function Space Distributions over Kernels” (Best Paper winner) for the time series workshop at ICML 2019. Slides here.