Contextual Memory Recall: A Novel Metric for Class Incremental Learning

Authors

• Balasubramanian S, Sri Sathya Sai Institute of Higher Learning
• Sai Subramaniam M., Sri Sathya Sai Institute of Higher Learning
• Sai Sriram Talasu, Sri Sathya Sai Institute of Higher Learning
• Yedu Krishna P., Sri Sathya Sai Institute of Higher Learning
• Pranav Phanindra Sai M., Sri Sathya Sai Institute of Higher Learning
• Darshan Gera, Sri Sathya Sai Institute of Higher Learning
• Ravi Mukkamala, Old Dominion UniversityFollow

Document Type

Conference Paper

Publication Date

2025

DOI

10.1145/3774521.3774568

Publication Title

ICVGIP '25: Proceedings of the Sixteen Indian Conference on Computer Vision, Graphics and Image Processing

Pages

1-8

Conference Name

Sixteen Indian Conference on Computer Vision, Graphics, and Image Processing, December 17-20, 2025, Mandi Himachal Pradesh, India

Abstract

We propose a novel metric for class incremental learning (CIL) called Contextual Memory Recall (CMR), which evaluates how well a CIL model recalls previously learned classes when given relevant past cues. Inspired by human memory, CMR offers newer insights into continual aspects of a CIL model that were not addressed by previously proposed metrics for CIL. Specifically, the standard metric, average incremental accuracy (AIA), overlooks the quality of evolving feature representations, whereas our proposed CMR accounts for it. As a result, methods using feature distillation perform well under AIA but poorly under CMR, while those without feature distillation excel under CMR. This underscores the need for more comprehensive CIL evaluation metrics, with CMR marking a significant step forward. As an extension to CMR, we propose Cue-Proportional Contextual Memory Recall (CPCMR) that penalizes a model's recall capability in proportion to the amount of past cues provided. We present a detailed comparative analysis of various metrics in CIL, demonstrating that CMR more effectively assesses CIL methods based on their ability to learn continually. The code is available at https://github.com/dlclub2311/Contextual-Memory-Recall/tree/main.

Rights

© 2025 Copyright held by the owner/authors.

The work is licensed under a Creative Commons Attribution 4.0 International (CC BY 4.0) License.

Original Publication Citation

Balasubramanian, S., Sai Subramaniam, M., Talasu, S. S., Yedu Krishna, P., Pranav Phanindra Sai, M., Gera, D., & Mukkamala, R. (2025). Contextual memory recall: A novel metric for class incremental learning. In Ragini Verma, Anand Mishra, & Arnav Bhavsar (Eds.), ICVGIP '25: Proceedings of the Sixteen Indian Conference on Computer Vision, Graphics and Image Processing (pp. 1-8). Association for Computing Machinery. https://doi.org/10.1145/3774521.3774568

Repository Citation

Balasubramanian, S., Sai Subramaniam, M., Talasu, S. S., Yedu Krishna, P., Pranav Phanindra Sai, M., Gera, D., & Mukkamala, R. (2025). Contextual memory recall: A novel metric for class incremental learning. In Ragini Verma, Anand Mishra, & Arnav Bhavsar (Eds.), ICVGIP '25: Proceedings of the Sixteen Indian Conference on Computer Vision, Graphics and Image Processing (pp. 1-8). Association for Computing Machinery. https://doi.org/10.1145/3774521.3774568

ORCID

0000-0001-6323-9789 (Mukkamala)