The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Adaptivity in Machine Learning with Samory Kpotufe - #512
Aug 23, 2021
In this engaging conversation, Samory Kpotufe, an associate professor at Columbia University, delves into the complexities of adaptive algorithms in machine learning. He highlights the importance of self-tuning algorithms that can adjust to varying data. The discussion covers transfer learning, emphasizing practical applications and challenges. Samory also touches on innovative methods in unsupervised learning and anomaly detection, especially within resource-constrained devices. His insights into the intersection of fractals and high-dimensional data add a fascinating layer to the conversation.
AI Snips
Chapters
Transcript
Episode notes
Clustering Challenges
- Defining "data grouping well" depends on distance or density notions.
- Choosing the right notion depends on the downstream task, leading to theoretical questions.
Intrinsic Low Dimensionality
- High-dimensional data often has an intrinsic low-dimensional structure.
- Manifold learning aims to discover this structure for better algorithm performance.
Algorithm Performance and Data Structure
- Algorithms like nearest neighbor perform better with inherently low-dimensional data.
- This inherent structure improves performance, regardless of data dimensionality.