GANs beyond nice pictures: real value of data generation, Alex Honchar
2 likes576 views
Generative modeling can be used for problems beyond just generation, such as anomaly detection, determining factors of variation in datasets, domain adaptation between not-aligned datasets, and building better embeddings for supervised learning tasks. Generative models can model the underlying distribution of data to check if a point belongs to that distribution or create new points from the distribution. They can learn low-dimensional manifolds on which real-world high-dimensional data like images lie. This allows generative models to be applied to challenges like filtering, style transfer, and improving embeddings to boost performance on downstream tasks.
![Lead: [1, 2 or 3]
Health: [healthy or after infarction]
Hardware: [with bs wander or without]](https://image.slidesharecdn.com/dsm19-190301103616-190301180126/85/GANs-beyond-nice-pictures-real-value-of-data-generation-Alex-Honchar-20-320.jpg)
GANs beyond nice pictures: real value of data generation, Alex Honchar
- 1. Generative modeling for anything but generation Alexandr Honchar
- 2. Artificial intelligence: Almost MsC in applied mathematics @ UNIVR, AI Consultant @ self-employed, Blogger, speaker, 5+ years in ML Healthcare: AI Solution Architect, partner @ Mawi Solutions Quantitative finance: Researcher @ UNIVR
- 3. main question: Can I use generative models even for my problems doesn’t really need to generate any pictures or other stuff?
- 8. Space of cats and dogs A very deep neural network already created some nice embedding space for us from initial pixel space
- 9. Discriminative modeling: A function f(x, w) telling us where new x belongs to from {cat, dog}
- 10. Generative modeling: A modeled distribution P(x| y), where x - our data point and y belongs to {cat, dog} So, we can or create new cat from P(x|y=“cat”), or check if some x_i belongs to P(x_i|y=“dog”) using well known maths
- 11. Natural manifold hypothesis: real-world high dimensional data (such as images) lie on low-dimensional manifolds embedded in the high- dimensional space Short tails, a little fur Long tails, a lot of fur
- 12. Natural manifold hypothesis: real-world high dimensional data (such as images) lie on low-dimensional manifolds embedded in the high- dimensional space
- 15. P(X): anomaly detection / one-class classification
- 16. too many sources and types of noise for supervised learning
- 19. Me and a dataset of ECGs I got from idk where with idk what properties What if it’s even not ECG? What if it’s some unknown signal with unknown features? I want an interpretable, generative and unsupervised representation! 15 TB
- 20. Lead: [1, 2 or 3] Health: [healthy or after infarction] Hardware: [with bs wander or without]
- 22. Epoch 1, z_i ~ (-3, 3)
- 23. Epoch 50, z_i ~ (-3, 3)
- 24. P(X)->P(Y): filterting / domain adaptation
- 25. X Y P(X) -> P(Y) ?
- 29. what if I only could work with all the leads…
- 30. +3-5% in F1 score depending on the target A Generative Modeling Approach to Limited Channel ECG Classification
- 31. Facebook: @rachnogstyle Medium: @alexrachnog Mail: alex@mawi.band Takeaways: - generative modeling >= discriminative modeling - works for anomaly detection - works for determining factors of variation - works for domain adaptation of not-aligned datasets - can build better embeddings for further supervised learning Alexandr Honchar