Reverse-engineering deep ReLU networks

Authors: David Rolnick, Konrad Kording

ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the success of our algorithm on both untrained and trained networks. In keeping with literature on Re LU network initialization (He et al., 2015; Hanin & Rolnick, 2018), networks were initialized using i.i.d. normal weights with variance 2/fan-in and i.i.d. normal biases with unit variance. Networks were trained on either the MNIST dataset (nin = 784, nout = 10) or a memorization task of 1000 datapoints (nin = 10, nout = 2) with coordinates drawn i.i.d. from a unit Gaussian and given arbitrary binary labels. Training was performed using the Adam optimizer (Kingma & Ba, 2014) and a cross-entropy loss applied to the softmax of the final layer, over 20 epochs for MNIST and 1000 epochs for the memorization task. The trained networks (when sufficiently large) were able to attain near-perfect accuracy.
Researcher Affiliation Academia 1University of Pennsylvania, Philadelphia, PA, USA. Correspondence to: David Rolnick <drolnick@seas.upenn.edu>.
Pseudocode Yes Algorithm 1 The first layer
Open Source Code No The paper does not provide any explicit statements about open-source code availability or links to a code repository.
Open Datasets Yes Networks were trained on either the MNIST dataset (nin = 784, nout = 10) or a memorization task of 1000 datapoints (nin = 10, nout = 2) with coordinates drawn i.i.d. from a unit Gaussian and given arbitrary binary labels.
Dataset Splits No The paper mentions training on MNIST and a memorization task, but does not provide specific details on dataset splits (e.g., percentages, counts, or predefined splits) for training, validation, or testing.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments.
Software Dependencies No The paper mentions the 'Adam optimizer' but does not specify any software versions or dependencies with version numbers (e.g., Python, deep learning frameworks).
Experiment Setup Yes In keeping with literature on Re LU network initialization (He et al., 2015; Hanin & Rolnick, 2018), networks were initialized using i.i.d. normal weights with variance 2/fan-in and i.i.d. normal biases with unit variance. [...] Training was performed using the Adam optimizer (Kingma & Ba, 2014) and a cross-entropy loss applied to the softmax of the final layer, over 20 epochs for MNIST and 1000 epochs for the memorization task.