Loss-Based Attention for Deep Multiple Instance Learning

Authors: Xiaoshuang Shi, Fuyong Xing, Yuanpu Xie, Zizhao Zhang, Lei Cui, Lin Yang5742-5749

AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on multiple types of benchmark databases demonstrate that the proposed attention mechanism is a general, effective and efficient framework, which can achieve superior bag and image classification performance over other state-of-the-art MIL methods, with obtaining higher instance precision and recall than previous attention mechanisms.
Researcher Affiliation Academia 1University of Florida, Gainesville, FL, USA 2University of Colorado Denver, Denver, CO, USA 3Northwestern University, Xi an, China {xsshi2015, shampool, zizhaozhang}@ufl.edu, fuyong.xing@ucdenver.edu, cuilei1989@163.com, lin.yang@bme.ufl.edu
Pseudocode No The paper describes the method using mathematical equations and diagrams, but does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks.
Open Source Code Yes Source codes are available on https://github.com/xsshi2015/Loss-Attention.
Open Datasets Yes We evaluate the proposed method, referred to as Loss Attention, on multiple benchmark MIL datasets (MUSK1, MUSK2, FOX, TIGER and ELEPHANT), MNIST-based and CIFAR-10-based MIL datasets, CIFAR-10 and Tiny Image Net image databases.
Dataset Splits Yes Following (Ilse, Tomczak, and Welling 2018) (Wang et al. 2018), we adopt 10-fold-crossvalidation and repeat five times per experiment for MIL and histopathology datasets. For experiments on MNISTbags, we utilize a fixed division into training and test sets. We build training sets with 50, 100, 150 and 200 bags, respectively, and a test set containing 1,000 bags.
Hardware Specification No The paper does not provide specific details about the hardware used for experiments, such as GPU/CPU models or memory specifications.
Software Dependencies No The paper mentions models like LeNet5 and ResNet18, but does not provide specific software dependencies or library versions (e.g., Python, PyTorch/TensorFlow versions) that would be needed for reproducibility.
Experiment Setup No The details of architectures, the parameter λ, ramp-up function ω(m), optimizer and hyperparameters are shown in the supplemental material (Tables A2, A3 and A4). This information is not provided in the main text of the paper.