Learning Orthogonal Multi-Index Models: A Fine-Grained Information Exponent Analysis
This work addresses a theoretical bottleneck in machine learning for researchers, providing a more efficient method for learning multi-index models, though it is incremental as it builds on existing information exponent analysis.
The paper tackles the problem of learning orthogonal multi-index models, showing that focusing only on the lowest-degree terms in the information exponent leads to suboptimal sample complexity rates. By incorporating both second- and higher-order terms, they achieve a sample complexity of ˜O(dP^{L-1}) for online SGD, improving over the previous ˜O(Pd^{L-1}) rate.
The information exponent ([BAGJ21]) and its extensions -- which are equivalent to the lowest degree in the Hermite expansion of the link function (after a potential label transform) for Gaussian single-index models -- have played an important role in predicting the sample complexity of online stochastic gradient descent (SGD) in various learning tasks. In this work, we demonstrate that, for multi-index models, focusing solely on the lowest degree can miss key structural details of the model and result in suboptimal rates. Specifically, we consider the task of learning target functions of form $f_*(\mathbf{x}) = \sum_{k=1}^{P} φ(\mathbf{v}_k^* \cdot \mathbf{x})$, where $P \ll d$, the ground-truth directions $\{ \mathbf{v}_k^* \}_{k=1}^P$ are orthonormal, and the information exponent of $φ$ is $L$. Based on the theory of information exponent, when $L = 2$, only the relevant subspace (not the exact directions) can be recovered due to the rotational invariance of the second-order terms, and when $L > 2$, recovering the directions using online SGD require $\tilde{O}(P d^{L-1})$ samples. In this work, we show that by considering both second- and higher-order terms, we can first learn the relevant space using the second-order terms, and then the exact directions using the higher-order terms, and the overall sample and complexity of online SGD is $\tilde{O}( d P^{L-1} )$.