Sequencing the Neurome: Towards Scalable Exact Parameter Reconstruction of Black-Box Neural Networks

TL;DR

This paper introduces a scalable method for exactly reconstructing neural network parameters from query access, leveraging the network's initialization bias and a novel informative query algorithm, demonstrated on large models.

Contribution

It presents a new approach combining inductive bias and an innovative query generation technique to efficiently reconstruct large neural networks exactly.

Findings

Reconstructed networks with over 1.5 million parameters.

Achieved less than 0.0001 max parameter difference.

Demonstrated robustness across various architectures and datasets.

Abstract

Inferring the exact parameters of a neural network with only query access is an NP-Hard problem, with few practical existing algorithms. Solutions would have major implications for security, verification, interpretability, and understanding biological networks. The key challenges are the massive parameter space, and complex non-linear relationships between neurons. We resolve these challenges using two insights. First, we observe that almost all networks used in practice are produced by random initialization and first order optimization, an inductive bias that drastically reduces the practical parameter space. Second, we present a novel query generation algorithm that produces maximally informative samples, letting us untangle the non-linear relationships efficiently. We demonstrate reconstruction of a hidden network containing over 1.5 million parameters, and of one 7 layers deep, the largest and deepest reconstructions to date, with max parameter difference less than 0.0001, and illustrate robustness and scalability across a variety of architectures, datasets, and training procedures.