High-dimensional Asymptotics of Denoising Autoencoders

Authors: Hugo Cui, Lenka Zdeborová

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We further show that our results accurately capture the learning curves on a range of real data sets. (...) We show that these formulae also describe quantitatively rather well the denoising MSE for real data sets, including MNIST [15] and Fashion MNIST [16].
Researcher Affiliation Academia Hugo Cui Statistical Physics of Computation Lab Department of Physics EPFL, Lausanne, Switzerland hugo.cui@epfl.ch Lenka Zdeborová Statistical Physics of Computation Lab Department of Physics EPFL, Lausanne, Switzerland
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code Yes The code used in the present manuscript can be found in the following repository.
Open Datasets Yes We show that these formulae also describe quantitatively rather well the denoising MSE for real data sets, including MNIST [15] and Fashion MNIST [16]. (...) For each data set, samples sharing the same label were considered to belong to the same cluster. (...) For each cluster, the corresponding mean µ and covariance Σ were numerically evaluated from the empirical mean and covariance over the 6000 boots (shoes) in the Fashion MNIST training set, and the 6265 1s (7s) in the MNIST training set.
Dataset Splits No The paper mentions training and test sets but does not explicitly specify a separate validation set or its split.
Hardware Specification No The paper mentions using "Pytorch implementation" for numerical simulations but does not specify any particular hardware details such as GPU/CPU models, memory, or cloud computing instances.
Software Dependencies No The paper mentions using "Pytorch implementation of full-batch Adam [43]" but does not specify version numbers for Pytorch or Python, which are necessary for reproducible software dependencies.
Experiment Setup Yes Dots represent numerical simulations for d = 700, training the DAE using the Pytorch implementation of full-batch Adam, with learning rate η = 0.05 over 2000 epochs, averaged over N = 10 instances. Error bars represent one standard deviation. (...) training a DAE (p = 1, σ = tanh) trained with n = 784 training points, using the Pytorch implementation of full-batch Adam, with learning rate η = 0.05 and weight decay λ = 0.1 over 2000 epochs, averaged over N = 10 instances.