دانلود کتاب تفسیرپذیری در یادگیری عمیق

This book is motivated by the large gap between the black-box nature of deep learning architectures and the human interpretability of the knowledge models they encode. It is increasingly important that the artificial intelligence models are accurate and understandable such that artificial and human intelligence can co-exist and collaborate. In certain life-threatening applications, interpretability is vital for root cause analysis and human decision-making. This book focuses on a comprehensive curation, exposition, and illustrative discussion of recent research tools for interpretability of deep learning models, with a focus on neural network architectures. A significant part of the work complements existing textbooks on deep learning and neural networks, and builds upon works of the past decade where the focus has been visualization and interpretability of the knowledge encoded in the networks. These works are sourced from the leading conferences and journals in computer vision, pattern recognition, and deep learning. Furthermore, we include several case studies from applicationoriented articles in different fields, including computer vision, optics, and natural language processing. In current graduate courses related to deep learning, machine learning, and neural networks, there is an absence of teaching/learning material which dealswith the topic of interpretability/explainability. This ismainly attributed to the fact that the previous focus of the machine learning community was precision, whereas the question of interpretability is an emergent topic. However, it is gaining traction as increasingly relevant subject, with books [81], [428], lecture notes [532], new courses as well as perspectives [520] being published. Nonetheless, the focus on general machine learning in these works implies that the question of interpretability in deep learning, which is now ubiquitously used across a large variety of machine learning applications, remains currently unaddressed to sufficient depth. This textbook will be therefore one of the pioneer textbooks dedicated to this topic. It will potentially lead to the creation of specialized graduate courses on this topic, since a need for such courses is perceived but unavailability of organized material on the topic is a prominent obstacle.