Information-Theoretic Lower Bounds for Compressive Sensing with Generative Models

Abstract

It has recently been shown that for compressive sensing, significantly fewer measurements may be required if the sparsity assumption is replaced by the assumption the unknown vector lies near the range of a suitably-chosen generative model. In particular, in (Bora {\em et al.}, 2017) it was shown roughly $O(k\log L)$ random Gaussian measurements suffice for accurate recovery when the generative model is an $L$-Lipschitz function with bounded $k$-dimensional inputs, and $O(kd\log w)$ measurements suffice when the generative model is a $k$-input ReLU network with depth $d$ and width $w$. In this paper, we establish corresponding algorithm-independent lower bounds on the sample complexity using tools from minimax statistical analysis. In accordance with the above upper bounds, our results are summarized as follows: (i) We construct an $L$-Lipschitz generative model capable of generating group-sparse signals, and show that the resulting necessary number of measurements is $\Omega (k\log L)$; (ii) Using similar ideas, we construct ReLU networks with high depth and/or high depth for which the necessary number of measurements scales as $\Omega \left(kd\frac{\log w}{\log n}\right)$ (with output dimension $n$), and in some cases $\Omega (kd\log w)$. As a result, we establish that the scaling laws derived in (Bora {\em et al.}, 2017) are optimal or near-optimal in the absence of further assumptions.

arXiv commentTo appear in IEEE Journal on Selected Areas in Information Theory. In this version, Theorem 7 was updated to allow for general depth/width scalings

References in corpus (10)

• Compressed Sensing using Generative Models
• A Sharp Condition for Exact Support Recovery of with Orthogonal Matching Pursuit
• Deep Decoder: Concise Image Representations from Untrained Non-convolutional Networks
• Compressed Sensing with Deep Image Prior and Learned Regularization
• Representation Benefits of Deep Feedforward Networks
• Modeling Sparse Deviations for Compressed Sensing using Generative Models
• Distance-based and continuum Fano inequalities with applications to statistical estimation
• The universal approximation power of finite-width deep ReLU networks
• Lower Bounds for Compressed Sensing with Generative Models
• Solving inverse problems via auto-encoders