Conformal Prediction for Partial Label Learning

Authors: Xiuwen Gong, Nitin Bisht, Guandong Xu

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments are conducted on SOTA PLL methods and benchmark datasets to verify the effectiveness of the proposed framework. In this section, we empirically test the validity of the proposed framework CP-PLL in quantifying the uncertainty (i.e., predictive confidence) of partial label learning models by implementing it on top of the state-of-the-art PLL models and various datasets in terms of average set size (the smaller the better).
Researcher Affiliation Academia 1 University of Technology Sydney 2 The Education University of Hong Kong
Pseudocode Yes Algorithm 1: CP-PLL Algorithm Goal: Constructing PLL set predictor Cα P LL(X). Input: PLL calibration dataset {(xi, ˆyi)}n i=1, pre-trained model f, a testing instance xt; Output: the prediction set Cα P LL(xt). 1: Compute the partial label score function SP LL(xi, ˆyi) given Eq. (4); 2: Compute the PLL quantile function QP LL given Eq. (5); 3: Generate the prediction set Cα P LL(xt) given Eq. (6).
Open Source Code Yes Code is publicly available at https://github.com/kalpiree/CP-PLL.
Open Datasets Yes We evaluate CP-PLL on various benchmark datasets, including CIFAR-10 and CIFAR-100 (Krizhevsky and Hinton 2009), and their corresponding long-tailed versions, i.e., CIFAR-10-LT, CIFAR-100-LT.
Dataset Splits Yes We split the held-out training data with 60% as the calibartion data and 40% as the testing data on all datasets.
Hardware Specification No No specific hardware details (e.g., GPU models, CPU types, memory) are mentioned in the paper for running experiments.
Software Dependencies No The paper mentions using '18-layer Res Net as the backbone' and 'SGD with momentum of 0.9 and weight decay of 0.001 as the optimizer', but does not provide specific version numbers for any software dependencies like Python, PyTorch, or TensorFlow.
Experiment Setup Yes We use 18-layer Res Net as the backbone. The mini-batch size is set to 256 and all the methods are trained using SGD with momentum of 0.9 and weight decay of 0.001 as the optimizer. The initial learning rate is set to 0.01. We train the model for 800 epochs.