Statistics

I recently posted a new preprint titled Energy-Tweedie: Score meets Score, Energy meets Energy on ArXiv.

We first extend the classic Tweedie’s formula to a wider range of distributions, then derive a novel identity that links the Stein score under a generalized Gaussian (energy model) noise to the gradient of a matched Energy Score (a scoring rule), connecting two seemingly disparate approaches to generative modeling. Among other things, this enables energy-score-based denoising models to be plugged into standard diffusion pipelines, but with heavy‑tailed noise.

We have a pre-print out based on our work with Project CETI, an initiative to decipher sperm whale communication using machine learning. In the paper, we combine approaches that help discover how known properties of human language are learned by generative models when trained on labeled speech audio data with methodology inspired by causal inference and apply it to the communication system of sperm whales, for which we do not have any such ground truth.

A brief introduction to diffusion models

Given that the semester’s over, I have finally gotten around to updating the site. We have a pre-print out based on the work I did during my internship last summer. It examines the problem of inferring the characteristics of different people involved in the process of translating and reviewing text by using Hierarchical Bayesian Models.

Replication of Michalopoulos: The Origins of Ethnolinguistic Diversity