Tuning Frequency Bias in Neural Network Training with Nonuniform Data

Authors: Annan Yu, Yunan Yang, Alex Townsend

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

Reproducibility Variable Result LLM Response
Research Type Experimental This section presents three experiments with synthetic and real-world datasets to investigate the frequency bias of NN training using squared L2 loss and squared Hs loss.
Researcher Affiliation Academia Annan Yu Cornell University ay262@cornell.edu Yunan Yang ETH Z urich yyn0410@gmail.com Alex Townsend Cornell University townsend@cornell.edu
Pseudocode No The paper describes methods, such as gradient descent, but does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks.
Open Source Code No The paper does not provide an explicit statement about releasing source code for the described methodology or a link to a code repository.
Open Datasets Yes MNIST handwritten digit database. ATT Labs [Online]. Available: http://yann.lecun.com/exdb/mnist, 2, 2010. and We utilize a data set {xi}2500 i=1 in (Wright & Michaels, 2015), which comes with carefully designed positive quadrature weights {ci}2500 i=1 .
Dataset Splits No The paper uses datasets like MNIST but does not provide specific details on training, validation, and test splits (e.g., percentages, sample counts, or citations to predefined splits).
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory, or cloud instance types) used for running the experiments.
Software Dependencies No The paper mentions 'keras' in a citation but does not list any specific software dependencies with version numbers required for reproducibility.
Experiment Setup Yes We set up two 2-layer Re LU-activated NNs with 5 104 hidden neurons in each layer and train them using the same training data and gradient descent procedure, except with different loss functions Φ and eΦ. and We train the autoencoder using mini-batch gradient descent with batch size equal to 256.