(Ελληνικά) Πιτυρίγκας Ευριπίδης (Phd Candidate)

Thesis title: Efficient instance-based classification algorithms

Supervisor: Ougiaroglou Stefanos

Advisory Committee Members:

Bratsas Charalampos, Assistant Professor, Dept. of Information and Electronic Engineering, IHU
Kotsakis Rigas, Assistant Professor, Dept. of Information and Electronic Engineering, IHU

Abstract:

Instance-based classification algorithms, and in particular the k-Nearest Neighbor classifier, constitute a broad scientific field, providing a simple yet powerful approach in the field of Machine Learning and Data Mining. However, as the complexity and size of data increase, traditional algorithms often face significant challenges in ensuring efficiency (optimal allocation of computational resources and execution times), accuracy (how well the outputs of an algorithm correspond to the correct or expected result), and scalability (preventing loss of efficiency and accuracy on complex and large-scale datasets). By the term complex data we mainly refer to multidimensional data, data heterogeneity in terms of their type or sources of origin (e.g. data expressed in non-Euclidean spaces such as texts and/or numerical values and/or waveforms and/or images all together as labels in the same data set), nonlinear data (e.g. data in graphs and trees), extremely large volume data, data with incorrect or missing values (e.g. natural numbers expressed with negative or empty values), dynamic data that changes over time (e.g. real-time sensors), etc. Additionally, snapshot-based categorization algorithms include parameters, the definition of which is crucial for both the effectiveness and efficiency of the algorithms. This thesis proposal aims to investigate the above challenges with the aim of designing and implementing new computational algorithms for categorization based on snapshots as well as modifying existing methods.