Learning in Intelligent Models under Explainability and Sample Constraints

Escaping Low-Data Regimes via Explainability for Medical and Fault Diagnosis Systems
Erscheinungsjahr 2026
ISBN: 978-3-032-40813-6
Verlag: Springer

Buch, Englisch, 130 Seiten, Format (B × H): 155 mm x 235 mm

Reihe: SpringerBriefs in Computer Science

This book provides a unified framework for learning in intelligent systems under conditions of limited data and strict explainability requirements. It addresses a critical gap in modern machine learning, where high-performance models often rely on large datasets and operate as black boxes, limiting their applicability in high-stakes domains.The book introduces the LIMESC framework, a novel approach that integrates explainability, learning, and domain knowledge into a single methodological structure. It systematically explores how models can remain robust, interpretable, and adaptable when data are scarce, noisy, or evolving.Core topics include neural computing, deep learning under small-data regimes, regularization and optimization strategies, class-incremental learning without memory, and dataset knowledge transfer. The book further examines post-hoc explainability methods and transitions toward intrinsic interpretability through causal and neuro-symbolic approaches. Additional perspectives such as topological data analysis and reinforcement learning in constrained environments are also presented.The framework is grounded in real-world applications, particularly medical diagnostics and fault detection systems, where explainability and reliability are essential. Through a combination of theoretical insights and practical methodologies, the book offers a structured pathway toward designing adaptive and interpretable machine learning systems.This book is intended for researchers, advanced graduate students, and practitioners in machine learning, artificial intelligence, and biomedical engineering.

Zielgruppe

Research

Autoren/Hrsg.

Onchis, Darian M.

Fachgebiete

• Mathematik | Informatik EDV | Informatik Angewandte Informatik
• Mathematik | Informatik Mathematik Numerik und Wissenschaftliches Rechnen Angewandte Mathematik, Mathematische Modelle
• Mathematik | Informatik EDV | Informatik Informatik Künstliche Intelligenz

Inhaltsverzeichnis

.- Small Datasets and supervising Knowledge.
.- Neural computing and supervised learning.
.- Post-hoc Explainability and feature-attribution learning.
.- Intrinsic Explainability, causual and counterfactual learning.
.- Neuro- symbolic approach for learning optimization.
.- Topological data analysis for machine learning.
.- Unsupervised and reinforcement learning in limited scenerios.
.- Applied systems for trustworthy vidion and multi-agent explainability.

Über Autor(innen)

Darian M. Onchis is a researcher in machine learning, explainable artificial intelligence (XAI), and neuro-symbolic systems, with a particular focus on learning under data constraints. His work bridges statistical learning, deep learning, and structured knowledge integration, with applications in medical diagnostics and fault detection systems.