BDW17 London - Edward Kibardin - Mitie PLC - Learning and Topological Data Analysis for Advanced Data Segmentations and Predictions
2 likes178 views
The document discusses the application of deep learning and topological data analysis (TDA) for the segmentation and prediction of unstructured data across various fields such as medical imaging and customer activities. It outlines challenges and methodologies in analyzing text and sensor signals, and emphasizes the integration of deep generative models with TDA for improved data contextualization. Various research references are provided to support the methodologies presented.
BDW17 London - Edward Kibardin - Mitie PLC - Learning and Topological Data Analysis for Advanced Data Segmentations and Predictions
- 1. Deep Learning and Topological Data Analysis for Advanced Data Segmentation and Prediction Big Data Week, 2017 Edward Kibardin
- 2. The vast majority of data is unstructured and unlabeled Medical and DNA profiling Images TextStock market transactions Customer Activities Sensor signals (IoT) System Logs Sound Unstructured Data
- 3. The vast majority of data is unstructured and unlabeled Medical and DNA profiling Images TextStock market transactions Customer Activities Sensor signals (IoS) System Logs Sound Unstructured Data
- 4. • How many categories in my dataset? • Which categories are the best for the business? • Why some objects are not like the others? • How I can contextualize new objects? • Is there a simpler way to describe my data? Unstructured Data Business question to unstructured and unlabeled data
- 5. Text
- 6. Text Analysis Legacy approach to text analysis
- 7. Text Analysis Understand the language in maintenance job descriptions. Discover new job types, for better reporting and further investigations. • 365,000 maintenance job • Each description from 5 to 1000 words New challenges
- 10. Text Analysis
- 11. a. Compute a combinatorial model approximating the structure of the underlying space b. Then compute topological invariants of this structure c. Represent these topological invariants in 2d space Topological Data Analysis Pipeline a b c Reference: Teng Ma ; Zhuangzhi Wu ; Pei Luo ; Lu Feng. Reeb graph computation through spectral clustering, 2011.
- 12. Sensor signals (Internet of Things)
- 13. Sensor Signals (Internet of Things) Legacy approach to sensor data analysis
- 14. New challenges Human Activity Recognition Using Smartphones Technical Research Centre for Dependency Care and Autonomous Living Universitat Politècnica de Catalunya (BarcelonaTech) Activities: Standing, sitting, laying, walking, walking upstairs, walking downstairs, stand-to-sit, sit-to-lay, lay-to-sit, sit-to-stand, stand-to-lay Sensor Signals (Internet of Things)
- 17. Customer Activity Legacy approach to customer activity analysis
- 18. Deep Generative Nets + TDA 1. Learning of deep generative model 2. Fine-tuning using topological loss
- 20. Links Topology And Data (Gunnar Carlsson): http://www.ams.org/journals/bull/2009-46-02/S0273-0979-09-01249-X/S0273-0979-09- 01249-X.pdf Discrete Morse Theory and Persistent Homology (Kevin P. Knudson): http://www.math.fsu.edu/~hironaka/FSUUF/knudson.pdf Topological Persistence and Simplification (Herbert Edelsbrunner, David Letscher, Afra Zomorodian): http://math.uchicago.edu/~shmuel/AAT-readings/Data%20Analysis%20/PersTop.pdf Extracting and Composing Robust Features with Denoising Autoencoders (Pascal Vincent, Hugo Larochelle, Yoshua Bengio, Pierre-Antoine Manzagol) http://www.iro.umontreal.ca/~vincentp/Publications/denoising_autoencoders_tr1316.pdf
- 21. More examples using TDA and Deep Learning
- 24. 20 News groups text dataset Motorcycles Christian Atheism Religion.misc Politics.guns Politics.misc Politics.mideast Scy.crypt Scy.med Hockey Baseball Autos ForsaleMac.hardware Electronics Scy.space Comp.graphics Windows.x Ms-windows.misc Pc.hardware
Editor's Notes
- #3: Our data storages are full of row unlabeled data
- #4: Our data storages are full of row unlabeled data
- #6: Our data storages are full of row unlabeled data
- #7: Our data storages are full of row unlabeled data
- #12: 0-dimentional simplexes 1-dimentional simplexes 2-dimentional simplexes
- #13: Our data storages are full of row unlabeled data
- #14: Our data storages are full of row unlabeled data
- #15: Our data storages are full of row unlabeled data
- #16: Our data storages are full of row unlabeled data
- #17: Our data storages are full of row unlabeled data
- #18: Our data storages are full of row unlabeled data
- #20: Our data storages are full of row unlabeled data