2019-06-25, 16:00–16:30, Room 1
With the computational resources becoming more powerful over time, tremendous advancements are being made in the field of Deep Learning. Generative Adversarial Networks (GANs) are one amongst such advancements. Interested in knowing how to "generate" content (images, music, speech, prose, and much more) instead of "classifying" one into categories? Let's dive into the granularities of Generative Adversarial Networks (GANs): One of the BIGGEST Breakthroughs in the Deep Learning Revolution.
The advancements in the field of Deep Learning are approaching a breakneck speed. Recent years have witnessed enormous research activities in Deep Learning, and Generative Adversarial Networks (GANs) is one of them. GANs are one of the most intriguing Deep Nets that have ever been built. GANs belong to a class of algorithms called the Generative Algorithms which help in predicting features given a certain label. This has led to the generation of artificial content (like images, music, speech, prose, and much more). Generative Adversarial Networks have a wide array of applications in the real world (including Super-resolution imaging). This talk aims at discussing the working of GANs, their applications to the real world (including Geo-Imagery), and demonstrating a quick hands-on code implementation using Python.
The flow of the talk will be as follows:
- Self Introduction
- A Succinct Prelude to Deep Learning
- Understanding Discriminative and Generative Algorithms
- A Brief Introduction to Adversarial Networks
- Working and Architecture of Generative Adversarial Networks (GANs)
- Quick Hands-on and Code Walkthrough (using Python)
- Tips to Train GANs Better
- Getting to know one of the Strongest Counterparts of GANs: Variational Autoencoders
- Discussing the Applications of GANs
- Roadmap to Further Study About GANs
- End of Talk
- Questions and Answers Session