Explore geometric deep learning applied to inverse problems on graphs in this 33-minute conference talk from the 2018 Physics Next workshop. Delve into topics like non-Euclidean geometries, physics-based geometric learning, and graph neural networks. Learn how these techniques are applied to particle physics, neutrino detection, and community detection. Discover the latest developments in computational-to-statistical gaps and supervised community detection, and gain insights into current challenges and open problems in the field.
Overview
Syllabus
Intro
DEEP LEARNING ON REGULAR GRIDS
TOWARDS NON-EUCLIDEAN GEOMETRIES
PHYSICS-BASED GEOMETRIC LEARNING
GEOMETRIC STABILITY IN EUCLIDEAN DOMAINS
CONVOLUTIONAL NEURAL NETWORKS
NON-EUCLIDEAN GEOMETRIC STABILITY
DEFORMATIONS AND METRICS
DIFFUSION AND METRIC STABILITY
LINEAR STABLE GENERATORS
LAPLACIAN INTERPRETATION
SURFACE REPRESENTATIONS
PARTICLE PHYSICS WITH GRAPH NEURAL NETWORKS
ICECUBE NEUTRINO DETECTION
INVERSE PROBLEMS ON GRAPHS
DABL-DRIVEN COMMUNITY DETECTION
REACHING DETECTION THRESHOLD ON SEM
GRAPH NEURAL NETWORKS ON GRAPH HIERARCHES
COMPUTATIONAL-TO-STATISTICAL GAPS
SUPERVISED COMMUNITY DETECTION
CURRENT AND OPEN PROBLEMS
Taught by
APS Physics
Related Courses
-
Machine Learning with Graphs - Fall 2019
-
A Sheaf-based Approach to Graph Neural Networks
-
Physics-Inspired Learning on Graph - Michael Bronstein, PhD
-
Equivariant Machine Learning Structured Like Classical Physics
-
From Shallow to Deep Learning for Inverse Imaging Problems - Carola-Bibiane Schönlieb, Cambridge
