Ensuring a consistent perception of the world by humans and machines is not only foundational for a Video Search company like muse.ai, it is also key to evolve artificial intelligence so that it can empathize and collaborate with humans to devise win/win outcomes.
In this context, I gave the following presentation (at SecondHome) and described:
how the concept of Artificial Intelligence is as old as human imagination;
the sequence of key developments that lead to today’s “AI hype”;
how many jobs have already been automated away (gratefully); and
how machines see color, objects, hear sounds, recognize speech, and relate concepts.
The running theme throughout the presentation was that context, and inherent biases, have a large impact on how humans perceive color, sounds, relate concepts, and if machines are to understand the world like we humans do, they also need to be loaded with the same preconceptions.
This talk was given at Thomson Reuters in London for an audience of Quants and Data Scientists, and comprised of 3 sections:
the first section described how data has many different properties, some of which are mutually exclusive, and how these aspects define how an optimal system must be designed. This section also included a data description from a high-level C-suit perspective (4 V’s) all the way down to a systems engineer that has to worry about how to efficiently transmit and store data;
the second part focused on a number of Machine Learning algorithms and how these always boil down to an optimization problem, and how some algorithms map better to sequential processors (e.g. CPUs) while others ideally map into fine-grained parallel architectures (e.g. FPGAs);
finally, the previous two sections are tied together by exemplifying a number of large scale systems that combine huge-data sets with rapid SDLC.
After 5 years of working alongside some extraordinary individuals1 on some elite systems2, the time has come to step-up to the CTO role at an ambitious start-up.
Having applied FPGAs in space3, and in finance to implement the World’s Fastest Matching and Crossing Engine (i.e. Exchange), I believe that the next really exciting challenge is to employ this amazing technology to AI/ML.
This field also brings to life some core-aspects of my Ph.D. that was focused on using FPGAs to accelerate the heart of most AI algorithms, i.e. finding the vector (x) that minimises the error of systems of linear equations (Ax = b) subject to a certain constraints (x ≤ c).
Checkout Carlos Faham’s page for little more insight into the kinds of problems we are solving.
As part of a Morgan Stanley’s recruitment effort, I was invited to give a talk about Field Programmable Gate Arrays in Finance at my alma mater – Imperial College London.
Since this was mainly targeted at students, this presentation focused on a myriad of applications of FPGAs in this industry. These applications ranged from simple use-cases like kill-switches to all the way to complete Trading-Systems-on-a-Chip.
This talk also explained why, everything being equal, traders prefer to go to the fastest exchange.
Last week, I was in New York for the first conference exclusively dedicated to Python for Quants.
This was a phenomenal event that brought together some of the best quants and technologists in the field, including the masterminds behind Athena (JP Morgan) and the Quartz (BAML) platforms.
My talk lasted for a little less than an hour and was focused on giving a broad overview on Bitcoin. It started by describing common concerns, misconceptions, and media confusion. Then it went into detailing the key technologies that were combined to enable this remarkable innovation. This technical part was followed by a section describing a number key analytics and statistics. Finally, the focus was set on the wonderful new services that this crypto-currency is enabling.
Special thank you to the organisers, James Powell and Dr. Yves Hilpisch, for hosting this extraordinary event and for the invitation to give this talk.
Today is a historical day for Europe! This is due to the successful launch of the GIOVE/A satellite. This satellite is the first to provide validation signals for the ambitious European Galileo Project. This global navigation system will eventually deploy around 30 satellites providing a complementary positioning system to the current American (GPS) and Russian (GLONASS) systems. Aboard this satellite there is both hardware and software developed by me.
Life has changed quite drastically since joining J.P. Morgan Chase almost a year ago. Since then, I have been on a psychedelic trip involving all sorts of wonderful and weird things. These include Vanilla flavors; wicked Japanese Uridashis; traders clamoring across the room; going to work before dawn and returning after sunset; getting to wear cufflinks and not getting weird looks; knowing that it is Friday because one particular guy is wearing a flowery shirt; etc; and getting to hang out with some really extraordinary individuals.
I have integrated the Analytics Strategies Group which is responsible for the Core functionalities of Athena:
“Athena is J.P. Morgan’s cross-market risk management and trading system. It is currently used in our foreign exchange and commodities businesses, and is being rolled out more broadly across our fixed income businesses. Athena includes a globally replicated object-oriented database, a powerful dependency graph and a fully integrated stack across pricing, risk and trading tools. The code is a combination of Python, C++, and Java: C++ and Java for speed, and Python for flexibility and rapid but controlled releases. Athena is designed to pull developers close to the business to help increase revenues while improving operational processes and controls to reduce costs.” – J.P. Morgan Chase.
I have just submitted my Ph.D. thesis entitled “Accelerating Iterative Methods for Solving Systems of Linear Equations using FPGAs“
This work is the culmination of almost four years of focused research where a number of high-performance aspects were explored in the field of accelerating a basic and recurrent problem in Scientific Computing. This exploration was possible through the usage of Field Programmable Gate Arrays. FPGAs are fascinating devices that can be configured to be anything from a typical microprocessor to a highly specialized neural network. One of the great advantages of these devices is that data/computations can be accessed/processed in parallel whilst in typical high-end microprocessors there is only a very limited opportunity for parallelism (even when considering multiple multi-core CPUs). This FPGA advantage can translate into massive improvements in performance, especially in applications that can be computed simultaneously. Some examples of such applications include Monte Carlo simulations, string matching (i.e. in routing, virus detection, etc), high-performance data storage and retrieval, amongst many other applications.
I’m a big fan of Aronofsky’s especially of his “Requiem for a Dream” and his π movie.
“Restate my assumptions: One, Mathematics is the language of nature. Two: everything around us can be represented and understood through numbers. Three: if you graph the numbers of any system, patterns emerge. Therefore, there are patterns everywhere in nature. Evidence: the cycling of disease epidemics; the wax and wane of caribou populations; sun spot cycles; the rise and fall of the Nile. So, what about the stock market? The universe of numbers that represents the global economy. Millions of hands at work, billions of minds. A vast network, screaming with life. An organism. A natural organism. My hypothesis: within the stock market, there is a pattern as well… Right in front of me… hiding behind the numbers. Always has been” – Max Cohen