Reinforcement Learning, Kernels, Reasoning, Quantization and Agents - Full Workshop

Explore the fundamentals and current applications of reinforcement learning in this comprehensive workshop that examines why RL has become ubiquitous in AI development and whether it represents the breakthrough needed for advancing large language models beyond their current capabilities. Learn about the core principles of reinforcement learning, discover what constitutes an effective reward function, and understand how RL enables the creation of intelligent agents. Delve into kernel methods to determine their continued relevance in modern machine learning and identify key areas of focus for practitioners. Investigate advanced quantization techniques that allow models like DeepSeek-R1 to be compressed to 1.58-bits while maintaining performance, and master methods for preserving accuracy during the quantization process. Gain insights from Daniel Han, founder of Unsloth, an open-source startup focused on making AI more accessible, who brings extensive experience from collaborating with Google, Meta, and Hugging Face teams on debugging open-source models including Llama, Phi, and Gemma, as well as his previous work at NVIDIA optimizing TSNE algorithms for 2000x performance improvements.