[ML] Generative Adversarial Networks and Building an Automated Bitcoin Trader

GDG Reading & Thames Valley
Thu, Nov 22, 2018, 6:30 PM (GMT)

About this event

Join us for a special ML focused event where we look at some of the issues and latest research in the area of GANs and then dive under the hood of a bitcoin trader using ML.

This event contains a mixture of ML research and implementation, both going in depth with experienced speakers who are excited to share their knowledge so please join us for a very interesting evening!!


18:30 Food, drinks, networking

19:00 Michael Arbel - GAN’s and their stability issues (training) + the latest advances to make training more stable.

20:00 Bret Colloff - Building an Automated Bitcoin Trader with Machine Learning

Talk 1:

"GAN’s and their stability issues (training) + the latest advances to make training more stable"

There has been an explosion of interest in generative adversarial networks (GANs) over the last few years. These models allow approximate samples from a complex high-dimensional target distribution, using a model distribution, where estimation of likelihoods and exact inference are not tractable. GANs have yielded very impressive empirical results, particularly for image generation, far beyond the quality of samples seen from most earlier generative models. These excellent results, however, have depended on adding a variety of methods of regularization and other tricks to stabilize the notoriously difficult optimization problem of GANs. Still, the reason why these additional tricks are needed is not completely understood. In the present work we shade some light on the reasons such instabilities occur and provide a principled method for regularization when the Maximum Mean Discrepancy is used as a loss. Experimental results show that the proposed regularization leads to stable training and outperforms state-of-the art methods on image generation.

Speaker Bio:
Michael Arbel
Michael is a Ph.D. student at the Gatsby Computational Neuroscience Unit under the the supervision of Arthur Gretton. Prior to that he graduated from Ecole Polytechnique with a major in Applied Mathematics and got a Masters Degree in Machine Learning at Ecole Normal Supérieure. He also worked as a Computer Vision Engineer at Prophesee where he developed tracking algorithm for event based sensors.
His interested in unsupervised learning methods (both parametric and non-parametric) and in the theoretical foundations of deep-learning.

Talk 2:

"Building an Automated Bitcoin Trader with Machine Learning"

This talk explores the process of building an automated, machine learning driven trading platform that is connected to cryptocurrency exchanges from inception to live testing. It will cover; monitoring the system and interacting with the live system with a chat interface through Telegram, examining the required knowledge, decisions, and mistakes that arise in developing the system.

Technologies applied
We'll be touching on Python, Docker, Numpy & Sklearn, Keras, PubSub, Cloud Storage, Kubernetes

Main topics:
• Basic currency trading and indicators, and turning these into features.
• Building a Neural Network in Keras with different layers.
• Designing and implementing a scalable cloud platform to operate from.

Speaker Bio:
Bret is a software developer working at Hu:toma AI on the APIs and core platform. He started as a developer/tester at Microsoft working on games and switched to streaming TV before moving to Ericsson and then Hu:toma. He is also a hobbyist in other aspects of development such as machine learning with an interest in scalable cloud-based systems and functional programming