3 Presentations

• 
  01:30 PM - 01:55 PM

  Nature-Inspired Optimization Techniques for Feature Selection in Data Clustering

  • Mandana Gholamigazafrudy, presenter, University of Regina
  • Malek Mouhoub, University of Regina
  • Samira Sadaoui, University of Regina

  We propose a feature selection approach based on nature-inspired optimization techniques specifically for data clustering. Our research focuses on improving computational efficiency, cluster quality and interpretability as well as minimizing noise and redundancy, resulting in more robust clusters. The proposed approach encompasses Biogeography-Based Optimization (BBO), Particle Swarm Optimization (PSO), Differential Evolution (DE), and Firefly Algorithm (FA) for feature selection, offering a comprehensive investigation of nature-inspired optimization techniques.

  Based on the within-cluster sum of squares (WCSS) of the data clustering algorithm K-means, the fitness function is employed consistently across all the nature-inspired techniques for a standardized metric. To determine the optimal number of clusters, we utilize the WCSS-based Elbow method, aligning with K-means and nature-inspired techniques, and the Silhouette score to capture data variations and improve cluster interpretation.

  We evaluate various nature-inspired feature selection techniques (BBO, PSO, DE, and FA) across several datasets using several metrics, including WCSS, Davies-Bouldin Index (DBI), and Silhouette score. This comprehensive assessment provides insights into the performance of these techniques in the context of dimensionality reduction for data clustering.

• 
  01:55 PM - 02:20 PM

  Optimization in a latent space for autonomous robotic systems testing

  • Dmytro Humeniuk, presenter, Polytechnique Montréal
  • Foutse Khomh, Polytechnique Montréal

  Evolutionary search-based techniques are commonly used for testing autonomous robotic systems. However, one of the big problems in applying optimization algorithms to test generation is finding a good representation of the problem. Commonly used representations for test scenarios for autonomous CPS are rather complex, including multidimensional arrays of different shapes, combinations of discrete and continuous values, as well as sequential dependence between elements. This complicates the usage of the advanced search algorithms and search operators in the test generation process. To address the problem of obtaining an efficient representation, we developed a tool AmbieGenVAE that first learns a simple one-dimensional representation of the test generation problem and then performs optimization of the candidate solutions in a latent space. The AmbieGenVAE approach includes three key steps: (1) collecting a dataset of test scenarios, (2) training a variational autoencoder (VAE) model, and (3) executing an evolutionary search within the latent space of the autoencoder. We evaluated AmbieGenVAE on two case studies: road topology generation for lane keeping assist system (LKAS) testing and box shaped obstacle positioning for autonomous drone testing. Preliminary results show, that in a given evaluation budget of 200 tests, AmbieGenVAE allows to reveal up to 4 times more failures for the LKAS system and 2 times more failures for the autonomous robot, than the baseline evolutionary search.

• 
  02:20 PM - 02:45 PM

  Heuristic Approach for Dredge Mining Optimization Using Mixed Integer Programming

  • Vincent Raymond, presenter, RioTinto

  This study proposes a heuristic approach for optimizing dredge mining operations employing Mixed Integer Programming (MIP). Dredge mining, crucial for extracting minerals from underwater deposits, poses intricate optimization challenges due to various constraints and uncertainties. The heuristic integrates MIP techniques with problem-specific heuristics to efficiently maximize net present value (NPV) while considering dredge capacity, resource depletion, environmental regulations, and operational costs. Key features include clustering, travelling salesman problem variation, variable neighborhood search, and solution refinement mechanisms.