projects
Explore some of my research projects below, each with a brief overview and a link to a detailed description.
Score-based Pullback Riemannian Geometry
In this work, we introduce a score-based pullback Riemannian metric that provides closed-form geodesics and interpretable autoencoding, capturing the intrinsic dimensionality and geometry of data.
In this work, we introduce a score-based pullback Riemannian metric that provides closed-form geodesics and interpretable autoencoding, capturing the intrinsic dimensionality and geometry of data.
ScoreVAE: Variational Diffusion Auto-encoder
In this work, we introduce a variation of the VAE framework that overcomes limitations of conventional VAEs and allows for latent space extraction from pretrained diffusion models.
In this work, we introduce a variation of the VAE framework that overcomes limitations of conventional VAEs and allows for latent space extraction from pretrained diffusion models.
Diffusion Models Encode the Intrinsic Dimension of Data Manifolds
In this work, we prove that diffusion models encode the intrinsic dimension of the data and propose a novel method to estimate the intrinsic dimension from a trained diffusion model.
In this work, we prove that diffusion models encode the intrinsic dimension of the data and propose a novel method to estimate the intrinsic dimension from a trained diffusion model.