The Shape of Data: Machine Learning and Topology
summary
Uses the mapper algorithm implemented in Python and R to turn high‑dimensional datasets into graphs, then inspects structure rather than fitting a black‑box model. Applies this to 700 MSCI EAFE stocks (correlation graphs) and a 4,000‑student health survey: a 38‑dimensional point cloud of 2,377 boys sliced into 16 overlapping slices reveals a backbone and a hole separating high, medium and low BMI clusters; dieting frequency and depression align strongly for girls and some clusters persist 15 years.
AI-written summary · about
chapters (16)
- 0:00Start
- 1:13Roadmap
- 2:5130000 foot view of machine learning etc.
- 8:36What's topology?
- 9:32Topology with data
- 11:05Simplices
- 11:32Build simplicial complex on your data
- 13:34Mapper for feature discovery
- 19:25Mapper: finance example
- 22:01The dieting paradox
- 23:49Attempting linear methods....
- 25:30Subgroups by behavioral variables
- 30:21Look at groups of boys
- 31:04Topology found us these subgroups
- 36:42Summary statistics, girls
- 37:40Differences persist 15 years later
related talks
34mTopological Data Analysis
20mAllison Gilmore, Data Scientist, Ayasdi
28mAnthony Bak, Principal Data Scientist at Ayasdi @ MLconf SEA
51mTopological space creation and Clustering at BigData scale
43mTopological data analysis: How to quantify "holes" in your data and why?
06mComplex Data Analysis
43mGraphs: Datastructures to Query
40mTopological Techniques for Unsupervised Learning