Optimizing Data Labeling: Active Learning, Weak Supervision & LLMs | CS 203 Week 4 скачать в хорошем качестве

Optimizing Data Labeling: Active Learning, Weak Supervision & LLMs | CS 203 Week 4

How do you label 100,000 examples without spending ₹25 lakhs and 8,000 hours? This lecture covers four powerful techniques to optimize your ML data labeling pipeline. 📌 Topics Covered: • Active Learning - Let your model pick which examples to label (2-10x savings) • Weak Supervision with Snorkel - Write labeling functions instead of manual labels • LLM-Based Labeling - Use GPT/Claude as cheap annotators • Noisy Label Detection - Find and fix label errors with Cleanlab 🔢 Includes worked examples: • Uncertainty sampling, margin sampling, entropy • Snorkel label model math (agreement equations, solving for α) • PyTorch gradient descent for LF accuracy estimation 📁 Code & Demos: https://github.com/nipunbatra/stt-ai-... Part of CS 203: Software Tools and Techniques for AI IIT Gandhinagar | Prof. Nipun Batra #MachineLearning #DataLabeling #ActiveLearning #WeakSupervision #Snorkel #LLM #Cleanlab