InFusionLayer: a CFA-based ensemble tool to generate new classifiers for learning and modeling
Researchers open-source first general-purpose Python library implementing Combinatorial Fusion Analysis for ensemble learning.
A research team led by Eric Roginek, Jingyan Xu, and D. Frank Hsu has introduced InFusionLayer, the first general-purpose Python tool designed to implement Combinatorial Fusion Analysis (CFA) for ensemble machine learning. Published on arXiv and accepted to the 2024 IEEE International Conference on Tools with Artificial Intelligence, this open-source library addresses a significant gap: while CFA—a method for combining multiple scoring systems using rank-score characteristic (RSC) functions and cognitive diversity (CD)—has been established in theory, no accessible tool existed for practitioners. InFusionLayer provides a machine learning architecture that leverages these CFA techniques at the system fusion level, enabling users to generate optimized classifiers from a moderate set of base models.
The tool is engineered for practical integration into modern ML workflows, offering compatibility with major frameworks including PyTorch, TensorFlow, and Scikit-learn. This allows data scientists to easily incorporate sophisticated ensemble methods into their existing pipelines for both unsupervised and supervised learning tasks, particularly multiclassification problems. The researchers validated InFusionLayer's performance across various computer vision datasets, demonstrating the practical advantages of its RSC and CD features. By open-sourcing the code on GitHub, the team aims to spur community development and make advanced CFA techniques more accessible, paving the way for more robust and accurate ensemble applications in real-world AI systems.
- First general-purpose Python tool implementing Combinatorial Fusion Analysis (CFA) for ensemble learning, filling a key library gap.
- Uses rank-score characteristic (RSC) functions and cognitive diversity (CD) to optimally combine multiple base models for classification.
- Open-source and compatible with PyTorch, TensorFlow, and Scikit-learn, validated on computer vision datasets and accepted to IEEE ICTAI 2024.
Why It Matters
Provides ML engineers with a standardized, open-source tool to easily build more accurate and robust ensemble models for complex classification tasks.