Expand-and-Cluster: Parameter Recovery of Neural Networks

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Martinelli, Flavio, Simsek, Berfin, Gerstner, Wulfram, Brea, Johanni
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911934212734976
author Martinelli, Flavio
Simsek, Berfin
Gerstner, Wulfram
Brea, Johanni
author_facet Martinelli, Flavio
Simsek, Berfin
Gerstner, Wulfram
Brea, Johanni
contents Can we identify the weights of a neural network by probing its input-output mapping? At first glance, this problem seems to have many solutions because of permutation, overparameterisation and activation function symmetries. Yet, we show that the incoming weight vector of each neuron is identifiable up to sign or scaling, depending on the activation function. Our novel method 'Expand-and-Cluster' can identify layer sizes and weights of a target network for all commonly used activation functions. Expand-and-Cluster consists of two phases: (i) to relax the non-convex optimisation problem, we train multiple overparameterised student networks to best imitate the target function; (ii) to reverse engineer the target network's weights, we employ an ad-hoc clustering procedure that reveals the learnt weight vectors shared between students -- these correspond to the target weight vectors. We demonstrate successful weights and size recovery of trained shallow and deep networks with less than 10\% overhead in the layer size and describe an `ease-of-identifiability' axis by analysing 150 synthetic problems of variable difficulty.
format Preprint
id arxiv_https___arxiv_org_abs_2304_12794
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Expand-and-Cluster: Parameter Recovery of Neural Networks
Martinelli, Flavio
Simsek, Berfin
Gerstner, Wulfram
Brea, Johanni
Neural and Evolutionary Computing
Can we identify the weights of a neural network by probing its input-output mapping? At first glance, this problem seems to have many solutions because of permutation, overparameterisation and activation function symmetries. Yet, we show that the incoming weight vector of each neuron is identifiable up to sign or scaling, depending on the activation function. Our novel method 'Expand-and-Cluster' can identify layer sizes and weights of a target network for all commonly used activation functions. Expand-and-Cluster consists of two phases: (i) to relax the non-convex optimisation problem, we train multiple overparameterised student networks to best imitate the target function; (ii) to reverse engineer the target network's weights, we employ an ad-hoc clustering procedure that reveals the learnt weight vectors shared between students -- these correspond to the target weight vectors. We demonstrate successful weights and size recovery of trained shallow and deep networks with less than 10\% overhead in the layer size and describe an `ease-of-identifiability' axis by analysing 150 synthetic problems of variable difficulty.
title Expand-and-Cluster: Parameter Recovery of Neural Networks
topic Neural and Evolutionary Computing
url https://arxiv.org/abs/2304.12794