Reinforcement Learning Methods in Speech and Language Technology

Baihan Lin is an AI researcher and neuroscientist at Columbia University, specializing in speech and natural language processing (NLP). With a PhD in computational biology from Columbia University and an MS in applied mathematics from the University of Washington, Baihan has dedicated his research to developing intelligent speech and text-based systems that can augment human-AI and human-human interactions in healthcare, and held research positions at IBM, Google, Microsoft, Amazon and BGI Genomics. He has created and deployed various pioneering machine learning solutions in the speech and language domains, such as the first-ever online and reinforcement learning (RL)-based speaker diarization system and RL-based interactive spoken language understanding (SLU) systems for children with speech and communication disorders. Baihan's research focuses on deep learning, RL and NLP has led to deployed real-world applications, such as AI companions for therapists and surrounding-aware virtual realities. He has authored 50+ peer-reviewed publications and patents, with an H-index of 13, and served program committees or reviewers for over 15 conferences, including INTERSPEECH and NeurIPS, as well as over 20 journals. Baihan was the chair of the conference tutorials at INTERSPEECH-22 and WACV-22 on RL and bandits for speech, NLP, computer vision and multi-fidelity signal processing, and the chair of the IJCAI-23 workshop on knowledge-based compositional generalization. His research has also contributed to the development of RSAToolbox, an open-sourced software that performs statistical inference to understand neural systems and the theory of neural networks.

Part I. A New Learning Paradigm in Speech and Language Technology.- Chapter 1. RL+SLT: Emerging RL-Powered Speech and Language Technologies.- Chapter 2. Why is RL+SLT Important, Now and How?.- Part II.  Bandits and Reinforcement Learning: A Gentle Introduction.- Chapter 3. Introduction to the Bandit Problems.- Chapter 4. Reinforcement Learning: Preliminaries and Terminologies.- Chapter 5. The RL Toolkit: A Spectrum of Algorithms.- Chapter 6. Inverse Reinforcement Learning Problem.- Chapter 7. Behavioral Cloning and Imitation Learning.- Part III. Reinforcement Learning in SLT Applications.- Chapter 8. Reinforcement Learning Formulations for Speech and Language Applications.- Chapter 9. Reinforcement Learning in Automatic Speech Recognition (ASR): The Voice-First Revolution.- Chapter 10. Reinforcement Learning in Speaker Recognition and Diarization: Decoding the Voices in the Crowd.- Chapter 11. Reinforcement Learning in Natural Language Understanding (NLU): Teaching Machines to Comprehend.- Chapter 12. Reinforcement Learning in Spoken Language Understanding (SLU): Giving Machines an Ear for Understanding.- Chapter 13. Reinforcement Learning in Text-to-Speech (TTS) Synthesis: Giving Machines a Voice.- Chapter 14. Reinforcement Learning in Natural Language Generation (NLG): Machines as Wordsmiths.- Chapter 15. Reinforcement Learning in Large Language Models (LLM): The Rise of AI Language Giants.- Chapter 16. Reinforcement Learning in Conversational Recommendation Systems (CRS): AI's Personal Touch.- Part IV.  Advanced Topics and Future Avenues.- Chapter 17. Emerging Strategies in Reinforcement Learning Methods.- Chapter 18. Navigating the Frontiers: Key Challenges and Opportunities in RL-Powered Speech and Language Technology.- Chapter 19. Reflections, Resources, and Future Horizons in RL+SLT.