Combining Case-Based Reasoning with Deep Learning: Context and Ongoing Case Feature Learning Research

David Leake, Zachary Wilkerson, David J. Crandall

Keywords: case-based reasoning, deep learning, feature learning, neuro-symbolic AI

TL;DR: This paper presents a brief survey of integrations of deep learning with case-based reasoning (CBR), focusing especially on extracting features for CBR retrieval from neural networks

Abstract: Case-based reasoning (CBR) and neural network models have complimentary strengths. Neuro-Symbolic hybrids of the two are promising for leveraging the inherently interpretable CBR process with capabilities of neural systems, facilitating processing of nonsymbolic inputs such as images and increasing system accuracy. Such hybrids also can help alleviate challenges for pure neural systems by facilitating explanation of system behavior and integrating domain knowledge. This paper briefly surveys prior directions in combining CBR with deep neural networks and summarizes ongoing research on combining CBR and deep learning to generate indices for case retrieval. This work focuses especially on selecting network architectures and feature extraction locations within a network to improve generation of CBR-useful features.

Submission Number: 17