Baihan Lin

Understanding the representations of artificial or biological neural networks is crucial in discovering the neural information processing mechanisms of the brain. Representational Similarity Analysis (RSA), is an analytical framework in computational and cognitive neuroscience, comparing models and brains in terms of their representational geometries. Representational Similarity Analysis: Unlocking the Neural Representations of Brains and Machines is the first book on representational similarity analysis, surveying the advances in computational neuroscience. This book is organized into five distinct sections. The first, introduces the reader to representation patterns and relation to neuroscience and psychology. The second section explores how to understand the data including data modalities in both modern neuroscience and AI research. The third section, reviews Representational similarity analysis (RSA) in depth, covering all aspects from metrics, interpretation and modeling. Next, section offers tutorials of RSA computations including setup, case studies and practical considerations. The last section summaries the possible future frontiers of representational studies.

This book offers a comprehensive guide to reinforcement learning (RL) and bandits methods, specifically tailored for advancements in speech and language technology. Starting with a foundational overview of RL and bandit methods, the book dives into their practical applications across a wide array of speech and language tasks. Readers will gain insights into how these methods shape solutions in automatic speech recognition (ASR), speaker recognition, diarization, spoken and natural language understanding (SLU/NLU), text-to-speech (TTS) synthesis, natural language generation (NLG), and conversational recommendation systems (CRS). Further, the book delves into cutting-edge developments in large language models (LLMs) and discusses the latest strategies in RL, highlighting the emerging fields of multi-agent systems and transfer learning. Emphasizing real-world applications, the book provides clear, step-by-step guidance on employing RL and bandit methods to address challenges in speech and language technology. It includes case studies and practical tips that equip readers to apply these methods to their own projects. As a timely and crucial resource, this book is ideal for speech and language researchers, engineers, students, and practitioners eager to enhance the performance of speech and language systems and to innovate with new interactive learning paradigms from an interface design perspective.