Towards New Optimized Artificial Immune Recognition Systems under the Belief Function Theory

Rihab AbdelkhalekIJCAI 2022 · DOI

TL;DR

This paper tackles the challenge of uncertainty in Artificial Immune Recognition Systems (AIRS), a powerful machine learning technique. The authors propose integrating Belief Function Theory to better handle this uncertainty and optimize AIRS performance using gradient descent and genetic algorithms, leading to improved classification accuracy.

What Problem Does This Paper Address?

Artificial Immune Recognition Systems (AIRS) are inspired by the human immune system and have shown great success in machine learning tasks. However, a significant limitation of existing AIRS approaches is their inability to effectively handle uncertainty that can arise at various stages of the classification process. This uncertainty can severely degrade the system's prediction performance. The authors aim to bridge this gap by incorporating Belief Function Theory, a robust framework for managing uncertainty, into AIRS. They also seek to optimize these enhanced systems to achieve even better classification accuracy.

Methods

The core of this research lies in enhancing AIRS by integrating two key components:

  • Belief Function Theory: This theory provides a mathematical framework for representing and combining uncertain information. The paper utilizes the concept of basic belief assignment (bba) to quantify belief in different hypotheses and Dempster's rule of combination to merge evidence from multiple sources. The pignistic probability transformation (BetP) is used for decision-making.
  • Optimization Techniques: To further improve performance, the authors explore optimization strategies:
  • Gradient Descent: Used to minimize a cost function and reduce the error rate within the belief function framework.
  • Genetic Algorithm: Employed to optimize parameters and reduce randomness in the evidential AIRS approach.

The paper also mentions an improved version of AIRS called AIRS3, which incorporates the number of represented antigens (numRepAg) for memory cells and uses a modified K-Nearest Neighbors (KNN) approach. The proposed methods build upon this AIRS3 by introducing evidential versions (EAIRS3, WE-AIRS3) and then optimizing them.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Uncertainty is a Major Bottleneck: The paper highlights that the inability of traditional AIRS to handle uncertainty is a significant obstacle to achieving accurate and effective classification.
  • Belief Function Theory as a Solution: The authors propose Belief Function Theory as a powerful and flexible tool to model and combine evidence, thereby addressing the uncertainty issue in AIRS.
  • Optimization is Crucial for Performance: Beyond handling uncertainty, optimizing the parameters of these evidential AIRS models is shown to be critical for maximizing prediction efficiency and accuracy.
  • Gradient Descent and Genetic Algorithms Offer Different Strengths: Both gradient descent and genetic algorithms are explored as optimization methods, suggesting that different optimization strategies can be applied to fine-tune the evidential AIRS models.

Key Results

The paper states that their proposed evidential AIRS approaches (EAIRS3, WE-AIRS3, and the optimized versions) have shown "excellent results" compared to traditional AIRS approaches. They claim improvements in classification accuracy and prediction performance. Specific quantitative results are not detailed in the provided text, but the authors refer to their previous publications for experimental comparisons.

Limitations and Future Work

  • Limitations: The paper acknowledges that the initial evidential AIRS (EAIRS3) did not consider the numRepAg component from AIRS3, treating all antigens in the memory cell pool with equal relevance. This was addressed in a subsequent weighted version (WE-AIRS3).
  • Future Work: The authors suggest exploring other bio-inspired algorithms, such as Particle Swarm Optimization (PSO), to further enhance prediction efficiency and develop more optimized evidential AIRS approaches. They also mention ongoing work in adapting new feature selection techniques within an uncertain environment to reduce the number of features and select only the most relevant ones, aiming to improve classification performance.

Summary updated

BibTeX
@inproceedings{abdelkhalek2022towards,
  author = {Rihab Abdelkhalek},
  title = {{Towards New Optimized Artificial Immune Recognition Systems under the Belief Function Theory}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2022},
  doi = {10.24963/ijcai.2022/821},
}
Figure