Random Function Descent

Authors: Felix Benning, Leif Döring

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Section 7 We conduct a case study on the MNIST dataset. In Figure 3, RFD is benchmarked against step size tuned Adam [29] and stochastic gradient descent (SGD).
Researcher Affiliation Academia Felix Benning University of Mannheim felix.benning@uni-mannheim.de Leif Döring University of Mannheim leif.doering@uni-mannheim.de
Pseudocode No The paper provides mathematical definitions and properties of RFD but does not include a formal pseudocode block or algorithm.
Open Source Code Yes Code availability: Our implementation of RFD can be found at https://github.com/ Felix Benning/pyrfd and the package can also be installed from Py PI via pip install pyrfd .
Open Datasets Yes For our case study we use the negative log likelihood loss to train a neural network [3, M7] on the MNIST dataset [34]. We also trained a different model (M5 [3]) on the Fashion MNIST dataset [56].
Dataset Splits No The paper mentions 'Validation loss' and 'test data set' being used for tuning, but does not provide specific train/validation/test split percentages or counts for the datasets.
Hardware Specification No The Experiments in this work were partially carried out on the compute cluster of the state of Baden-Würtemberg (bw HPC).
Software Dependencies No The paper does not specify version numbers for any software dependencies used in the experiments (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup Yes For our case study we use the negative log likelihood loss to train a neural network [3, M7] on the MNIST dataset [34]. Training on the MNIST dataset (batch size 1024). We also trained a different model (M5 [3]) on the Fashion MNIST dataset [56] with batch size 128. RFD is benchmarked against step size tuned Adam [29] and stochastic gradient descent (SGD).