MIML library: a Modular and Flexible Library for Multi-instance Multi-label Learning

TL;DR

The paper addresses the lack of an integrated, extensible library for multi-instance multi-label (MIML) learning by introducing the MIML library, a modular Java framework built on top of Weka and Mulan. It provides 43 algorithms and a comprehensive, XML-driven workflow for data management, partitioning, transformations between MIML, MI, and ML representations, distance measures, evaluation, and reporting. Key contributions include a package-based architecture, flexible data formats, configurable experiments, and extensive documentation to facilitate development and comparison of MIML methods. The work aims to accelerate experimentation and advance MIML research by enabling reproducible, configurable studies across diverse algorithms and datasets.

Abstract

MIML library is a Java software tool to develop, test, and compare classification algorithms for multi-instance multi-label (MIML) learning. The library includes 43 algorithms and provides a specific format and facilities for data managing and partitioning, holdout and cross-validation methods, standard metrics for performance evaluation, and generation of reports. In addition, algorithms can be executed through $xml$ configuration files without needing to program. It is platform-independent, extensible, free, open-source, and available on GitHub under the GNU General Public License.

Figures (1)

• Figure 1: Architecture of MIML library.