Presentation: The Move to AI: From HFT to Laplace Demon

Location:

Duration

Duration: 
5:25pm - 6:15pm

Day of week:

Level:

Persona:

Key Takeaways

  • Learn the principles and awareness needed to leverage AI in trading.
  • Hear about the time/data balance for choosing an algorithm for your next AI project
  • Understand the power of being able to keep feeding an algorithm with more and more sources of data.

Abstract

The race for low latency data continues. 10 years ago, Flashboys were helping HFT make money with low-latency infrastructures. Today, hedge funds build AI brains pumping hundreds of sources of data in real-time, seeking ubiquity to build Laplace Demons.

Interview

Question: 
You mention a Laplace Demon in your title. Can you tell us a bit about this talk and how a LaPlace Demon comes in with High Frequency Trading?
Answer: 

This talk is based on my experience working with hedge funds in places like New York, 10 years ago providing infrastructure to HFTs and Exchanges, and now streaming exotic data to them

Hedge funds over the past ten years have evolved from going after the time dimension (HFT) to working in the data dimension (a Laplace Demon). A Laplace Demon can predict anything as it knows everything instantly and understand their causality links. After HFT opportunities diminished, HFT turned into a Formula 1 race (nice to watch, but expensive and exclusive). One of these funds have over the past few years started to consume enormous amounts of real time data and make market predictions and bets out of it.

To get to the point of working with real time data, I discuss in the talk the history of AI with hedge funds starting with Renaissance Technologies. To describe the options to build our Laplace demon, I will cover some fairly recent AI evolutions -GPU, backprop- as applied to trading, and then discuss how I made my specific choice. With Albert, we will demo one of the frameworks. This is a non-commercial open source framework called MOA.

Question: 
In the system you describe, are you able to make decisions in the Microseconds that High Frequency Trading is accustomed too?
Answer: 

No. For hedge funds to consume lots of sources of data simultaneously and then to be able to make correlation and find causation, the time horizon (second/minute) we are talking about is much slower than High Frequency Trading (HFT: micro/millisecond). HFT is a way to be faster than others, AI is a way to be smarter than others.

In HFT, the longest timeframe you have is 10 milliseconds and that’s very long. You typically fight for nanoseconds (you are happy when you can shave off a microsecond). 15 years ago it was estimated that one millisecond for a broker was worth several million dollars. So this talk is about using AI to make better decision, not necessarily faster ones.

Question: 
What about enriching data? Does the approach you discuss talk about how you can add dimensions to the data?
Answer: 

The end game is for you to be able to predict the next tick and make less mistakes than others. So what hedge funds are starting to do is, in addition to consuming market data (which they’ve been doing forever), to consume social data from sentiment APIs like Vetr or Estimize, and social/social trading networks like Twitter, Facebook and Stocktwits.. We are even starting to see them consume IoT data sources. It’s not theoretical anymore that a guy trading meat in Chicago can make an educated decision on when and for how much he should buy based on API calls he makes to IoT devices on cows Argentina. Things that were just not possible until recently are all of a sudden possible. The race to train your own Laplace Demon has just started, a great time to join.

Speaker: Albert Bifet

Associate Professor @Telecom ParisTech

Albert Bifet is Associate Professor at Telecom ParisTech and Honorary Research Associate at the WEKA Machine Learning Group at University of Waikato. Previously he worked at Huawei Noah's Ark Lab in Hong Kong, Yahoo Labs in Barcelona, University of Waikato and UPC BarcelonaTech. He is the author of a book on Adaptive Stream Mining and Pattern Learning and Mining from Evolving Data Streams. He is one of the leaders of MOA and Apache SAMOA software environments for implementing algorithms and running experiments for online learning from evolving data streams. He was serving as Co-Chair of the Industrial track of IEEE MDM 2016, ECML PKDD 2015, and as Co-Chair of BigMine (2016-2012), and ACM SAC Data Streams Track (2017-2012).

Find Albert Bifet at

Speaker: Eric Horesnyi

CEO @streamdata.io

Eric is the CEO of streamdata.io. He was a founding team member at Internet Way (French B2B ISP, sold to UUnet) then Radianz (Global Finance Cloud, sold to BT). He is a High Frequency Trading infrastructure expert, passionate about Fintech, IoT and Cleantech. Eric looks after 3 bozons and has worked in San Francisco, NYC, Mexico and now Paris.

Find Eric Horesnyi at

Similar Talks

VP of High Frequency Engineering @Barclays
Director of Research @FastForwardLabs
Research Engineer @FastForwardLabs, Keras Contributor
CTO who understands the science around helping people do their best

Tracks

Conference for Professional Software Developers