Learning Orthogonal Multi-Index Models: A Fine-Grained Information Exponent Analysis
complexity analysisexact directionsgaussian single-index modelshermite expansionhigher-order termsinformation exponentlearning tasksmulti-index modelsonline stochastic gradient descentorthonormal directionsrelevant subspacerotational invariancesample complexitysecond-order termstarget functions
The information exponent (Ben Arous et al. [2021]) and its extensions