Zero to Hero: Few Shot Learning + Multi-armed Bandit
summary
Shows how to combine Few-Shot prompting with multi-armed bandit algorithms to pick which example sets to feed a large language model when solving boolean question answering (BoolQ). Uses live coding in a Jupyter Notebook to implement and evaluate this approach with concrete LM prompting and bandit-based selection, and offers practical tips for applying it to NLP tasks.
AI-written summary · about
related talks
26mFew-Shot Learning in Conversational AI
22mOnline Learning and Bandits with Queried Hints
26mLLM Inference with Few-Shot Prompting and Chain of Thought
58mIn-Context Principle Learning from Mistakes
38mPractical Few-Shot Learning
41mKeynote: Personalization, Bandits, and Causal inference
21mLeveraging Multi-Armed Bandit Algorithms for Dynamic Decision Making
32mMulti-Armed Bandits in Recommendations