Embeddings... Embeddings Everywhere
summary
The talk surveys the use of embeddings across data science, explaining how vector embeddings represent items (words, images, users) and enable tasks like similarity search, clustering, and downstream machine learning. It discusses practical tools and libraries from the Python ecosystem for creating and using embeddings, and shows workflows for applying embeddings to analytics and visualization problems.
AI-written summary · about
chapters (21)
- 0:00Embeddings dominate today
- 1:10Human modeling challenges
- 3:59Three embedding needs
- 5:46Modeling students and content
- 7:11Student-content distance
- 9:20Data vs domain knowledge
- 11:27RNNs for sequential data
- 12:49Dense-layer content embeddings
- 14:37Distance in architecture
- 15:40BERT for content
- 16:44Pretext tasks for embeddings
- 17:48Student knowledge tracking
- 19:12Interpretability vs flexibility
- 20:14Course dropout prediction
- 21:41Industry model evaluation
- 23:29Result metrics and distributions
- 25:17Dropout prediction results
- 26:44Neural network failures
- 27:48Embeddings with little data
- 28:52Visualizing embeddings
- 30:17Lessons and recommendations
related talks
01h 43mWorkshop: Amazing Embeddings — How Vector Representations Are Applied
27mData with Purpose: The Power of Embeddings
28m"Embeddings: How Computers Learned to Read"
39mUsing Embeddings to Understand the Variance and Evolution of Data Science...
30mThe Unlimited Potential of Neural Search to Unlock the New Way of Data Comprehension
27mMachine Learning for Advanced AI Search
28mEmbeddings
50mWhat Comes After ChatGPT? Vector Databases - the Simple & Powerful Future of ML?