Learning from Corrupted Binary Labels via Class-Probability Estimation

Authors: Aditya Menon, Brendan Van Rooyen, Cheng Soon Ong, Bob Williamson

ICML 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on label noise tasks corroborate our analysis.We now present experiments that aim to validate our analysis4 via three questions.
Researcher Affiliation Collaboration Aditya Krishna Menon ADITYA.MENON@NICTA.COM.AU Brendan van Rooyen BRENDAN.VANROOYEN@NICTA.COM.AU Cheng Soon Ong CHENGSOON.ONG@NICTA.COM.AU Robert C. Williamson BOB.WILLIAMSON@NICTA.COM.AU National ICT Australia and The Australian National University, Canberra The Australian National University and National ICT Australia, Canberra
Pseudocode No No structured pseudocode or algorithm blocks were found in the paper.
Open Source Code Yes Sample scripts are available at http://users.cecs.anu.edu. au/ akmenon/papers/corrupted-labels/index.html.
Open Datasets Yes We report results on a range of UCI datasets.
Dataset Splits Yes For each dataset, we construct a random 80% 20% train-test split.The regularisation parameter for the model was tuned by cross-validation (on the corrupted data) based on squared error.
Hardware Specification No No specific details about the hardware (e.g., GPU/CPU models, memory, or cloud computing resources) used for running the experiments were mentioned in the paper.
Software Dependencies No The paper mentions training a neural network with ℓ2 regularization, but does not provide specific version numbers for any software dependencies or libraries used (e.g., Python, TensorFlow, PyTorch, scikit-learn versions).
Experiment Setup Yes We focus on CCN learning with label flip probabilities ρ+, ρ {0, 0.1, 0.2, 0.3, 0.4, 0.49};... we use as our base model a neural network with a sigmoidal hidden layer, trained to minimise squared error5 with ℓ2 regularisation. The regularisation parameter for the model was tuned by cross-validation (on the corrupted data) based on squared error.