A Unified Approach to Count-Based Weakly Supervised Learning

Authors: Vinay Shukla, Zhe Zeng, Kareem Ahmed, Guy Van den Broeck

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

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our approach on three common weakly-supervised learning paradigms and observe that our proposed approach achieves state-of-the-art or highly competitive results across all three of the paradigms.
Researcher Affiliation Academia Vinay Shukla Department of Computer Science University of California, Los Angeles vshukla@g.ucla.eduZhe Zeng Department of Computer Science University of California, Los Angeles zhezeng@cs.ucla.eduKareem Ahmed Department of Computer Science University of California, Los Angeles ahmedk@cs.ucla.eduGuy Van den Broeck Department of Computer Science University of California, Los Angeles guyvdb@cs.ucla.edu
Pseudocode Yes Algorithm 1 Count Probability p(Pk i=1 ˆyi = s)
Open Source Code Yes 1Code and experiments are available at https://github.com/UCLA-Star AI/Count Loss
Open Datasets Yes We first experiment on the MNIST dataset [30] and follow the MIL experimental setting in Ilse et al. [24]... We also experiment on the Colon Cancer dataset [41]... We experiment on dataset MNIST and CIFAR10 [29]... Publicly available at archive.ics.uci.edu/ml
Dataset Splits No The paper mentions training and testing, and that it chooses the "single best epoch based on validation for our approaches" in Table 6. However, it does not explicitly provide numerical splits (percentages or counts) for training, validation, and test sets. It implies a validation set was used for hyperparameter tuning but doesn't quantify its size or proportion.
Hardware Specification No The paper does not provide specific details about the hardware used for experiments (e.g., CPU/GPU models, memory).
Software Dependencies No The paper refers to a PyTorch framework (Pylon) in its related work, but does not specify the versions of any software dependencies used for its own experiments.
Experiment Setup No The paper states: "We refer the readers to the appendix for additional experimental details." However, the provided text does not include these details, such as hyperparameters or specific training configurations in the main body.