Journal Club
A cumulative seminar archive for in-semester paper discussions on uncertainty quantification, scientific machine learning, and related statistical methodology.
From EHVI to Diffusion Models: Pareto Set Learning for Expensive Multi-Objective Bayesian Optimization
Open SessionMulti-objective Bayesian optimization (MOBO) is a framework for expensive black-box problems with multiple competing objectives, where each evaluation may require costly simulations or experiments.
FourCastNet: a weather forcasting model using Neural Operators
Open SessionFourCastNet 1 casts global weather forecasting as a learned, resolution-invariant neural operator: an Adaptive Fourier Neural Operator (AFNO) with a ViT backbone, trained on ERA5 at and rolled out autoregressively.
An Introduction to PINNs and Their Convergence
Open SessionPhysics-Informed Neural Networks (PINNs) are a framework for approximating partial differential equation (PDE) solutions by embedding governing physical laws directly into the neural network’s loss function via automatic differentiation.
Tabular Foundation Models - From Tree-based Methods to TabPFN
Open SessionFrom GBDTs to TabPFN: how deep learning finally started to competing on tabular data through in-context-learning foundation models, and where this new paradigm still breaks.
Marginal Tail-Adaptive Normalizing Flows
Open SessionTBA
Uncertainty Quantification in Vision Transformer
Open SessionLikelihood-guided Regularization in Attention Based Models.
Introduction to Flow Matching
Open SessionFlow matching is a framework for generative modeling that learns a vector field transporting a simple reference distribution to a complex target via an ODE.