Abstract: Graph neural networks (GNNs) are effective at learning representations from network data but face challenges with large graphs, which lack the Euclidean structure of time-series and image data. However, graphs have limits, such as the graphon—a bounded symmetric kernel that serves as both a random graph model and a …
Abstract: In this talk I will describe recent progress in characterizing structure that emerges during the training of deep neural networks for classification. Neural Collapse is a phenomenon that emerges during the training of deep classifiers in which the top-layer feature embeddings of samples from the same class tend to …
Abstract: Although incorporating causal concepts into deep learning shows promise for increasing explainability, fairness, and robustness, existing methods require unrealistic assumptions or aim to recover the full latent causal model. This talk proposes an alternative: domain counterfactuals. Domain counterfactuals ask a more concrete question: “What would a sample look like …
Abstract: Mirror descent, introduced by Nemirovsky and Yudin in the 1970s, is a primal-dual convex optimization method that can be tailored to the geometry of the optimization problem at hand through the choice of a strongly convex distance-generating potential function. It arises as a basic primitive in a variety of …
Abstract: Tensor decompositions generalize matrix decompositions from matrix data (i.e., 2-D arrays) to tensor data (i.e., N-D arrays) and are a fundamental technique for uncovering low-dimensional structure in high-dimensional datasets, with applications across all of science and engineering. Conventional tensor decompositions seek low-rank tensors that best fit the data with …
Abstract: Analyzing changes in network evolution is central to statistical network inference, as underscored by recent challenges of predicting and distinguishing pandemic-induced transformations in organizational and communication networks. We consider a joint network model in which each node has an associated time-varying low-dimensional latent vector of feature data, and connection …
Abstract: Large pre-trained models trained on internet-scale data are often not ready for safe deployment out of the box. They are heavily fine-tuned and aligned using large quantities of human preference data. When we want to align an AI/ML model to human preference or values, it is worthwhile to ask …