Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Ancestral Mamba: Enhancing Selective Discriminant Space Model with Online Visual Prototype Learning for Efficient and Robust Discriminant Approach

Jiahao Qin, Feng Liu, Lu Zong

TL;DR

Ancestral Mamba addresses online continual learning for dynamic visual data by integrating online prototype learning into a selective discriminant space model. The approach combines Ancestral Prototype Adaptation (APA), which maintains and refines class prototypes, with Mamba Feedback (MF), a targeted mechanism that emphasizes challenging distinctions to sharpen decision boundaries. Empirical results on CIFAR-10 and CIFAR-100 show that Ancestral Mamba achieves higher accuracy and reduced forgetting than competitive baselines, supported by ablations that confirm the value of both APA and MF. This work offers a robust, memory-efficient framework for continual visual learning with clear applicability to graphics and other evolving visual tasks.

Abstract

In the realm of computer graphics, the ability to learn continuously from non-stationary data streams while adapting to new visual patterns and mitigating catastrophic forgetting is of paramount importance. Existing approaches often struggle to capture and represent the essential characteristics of evolving visual concepts, hindering their applicability to dynamic graphics tasks. In this paper, we propose Ancestral Mamba, a novel approach that integrates online prototype learning into a selective discriminant space model for efficient and robust online continual learning. The key components of our approach include Ancestral Prototype Adaptation (APA), which continuously refines and builds upon learned visual prototypes, and Mamba Feedback (MF), which provides targeted feedback to adapt to challenging visual patterns. APA enables the model to continuously adapt its prototypes, building upon ancestral knowledge to tackle new challenges, while MF acts as a targeted feedback mechanism, focusing on challenging classes and refining their representations. Extensive experiments on graphics-oriented datasets, such as CIFAR-10 and CIFAR-100, demonstrate the superior performance of Ancestral Mamba compared to state-of-the-art baselines, achieving significant improvements in accuracy and forgetting mitigation.

Ancestral Mamba: Enhancing Selective Discriminant Space Model with Online Visual Prototype Learning for Efficient and Robust Discriminant Approach

TL;DR

Ancestral Mamba addresses online continual learning for dynamic visual data by integrating online prototype learning into a selective discriminant space model. The approach combines Ancestral Prototype Adaptation (APA), which maintains and refines class prototypes, with Mamba Feedback (MF), a targeted mechanism that emphasizes challenging distinctions to sharpen decision boundaries. Empirical results on CIFAR-10 and CIFAR-100 show that Ancestral Mamba achieves higher accuracy and reduced forgetting than competitive baselines, supported by ablations that confirm the value of both APA and MF. This work offers a robust, memory-efficient framework for continual visual learning with clear applicability to graphics and other evolving visual tasks.

Abstract

In the realm of computer graphics, the ability to learn continuously from non-stationary data streams while adapting to new visual patterns and mitigating catastrophic forgetting is of paramount importance. Existing approaches often struggle to capture and represent the essential characteristics of evolving visual concepts, hindering their applicability to dynamic graphics tasks. In this paper, we propose Ancestral Mamba, a novel approach that integrates online prototype learning into a selective discriminant space model for efficient and robust online continual learning. The key components of our approach include Ancestral Prototype Adaptation (APA), which continuously refines and builds upon learned visual prototypes, and Mamba Feedback (MF), which provides targeted feedback to adapt to challenging visual patterns. APA enables the model to continuously adapt its prototypes, building upon ancestral knowledge to tackle new challenges, while MF acts as a targeted feedback mechanism, focusing on challenging classes and refining their representations. Extensive experiments on graphics-oriented datasets, such as CIFAR-10 and CIFAR-100, demonstrate the superior performance of Ancestral Mamba compared to state-of-the-art baselines, achieving significant improvements in accuracy and forgetting mitigation.

Paper Structure

This paper contains 22 sections, 10 equations, 3 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Continual Learning Architectures
  4. Efficient Continual Learning Techniques
  5. Prototype-based Learning and Adaptive Feedback
  6. Methodology
  7. Overview of the Ancestral Mamba Architecture
  8. Ancestral Prototype Adaptation (APA)
  9. Mamba Feedback (MF)
  10. Training and Optimization
  11. Experiments and Results
  12. Experimental Setup
  13. Datasets and Benchmarks
  14. Baselines and Comparison Methods
  15. Evaluation Metrics
  16. ...and 7 more sections

Figures (3)

  • Figure 1: The APA module can reduce the distance between the new sample and the most similar prototype and increase the distance between the other prototypes during the learning process.
  • Figure 2: Overview of the proposed Ancestral Mamba. The input data is processed by the selective discriminant space model (SDSM) to extract initial features and prototype mappings. These features are then passed through the Ancestral Prototype Adaptation (APA) module and the Mamba Feedback (MF) mechanism to learn discriminative features and adapt to evolving data distributions.
  • Figure 3: Average incremental accuracy on (a) CIFAR-10(memory size=0.1k) and (b) CIFAR-100(memory size=0.5k).