Lost in Gans
What does a Data Scientist do when there's not enough data? One can adjust some dataflow pipelines, scrape raw data, ask the management to buy some... But that's too conventional. Here we generate the data! Some time ago I realized that I need more ultrasonic images for my project. That was the moment I started looking into different Deep Learning approaches, more specifically Generative techniques.
In this talk we will discuss some of them:
- Style Transfer
- Pix2Pix
We will also cover an image-filtering technique Deep Dream which produces hallucination-like visuals and play with images of pandas and axolotls. How to pick the right generative model for a particular use-case, what are their pros and cons and how easy to get started.