When Sir Francis Bacon in 1597 put to paper the age-old aphorism that “knowledge is power”, he could not have anticipated quite how powerful intelligence and data would become, and how the nature of knowledge itself has changed so dramatically.
Artificial Intelligence has heralded a new kind of knowledge and created unique opportunities for how humans can use it. Paper and ink have been replaced by machines and algorithms transforming data into a resource that can be exploited like coal and oil.
The relationship between finance and the quantitative skills needed in the industry are growing ever greater. A legion of quants, making use of advances in machine learning, now lead the field.
Bjarne Stroustrup for example, the father of C++, joined Morgan Stanley’s Technology Division in 2014.
It is also the direction of travel when it comes to the Big Tech companies that create the cloud computing infrastructure which is the backbone of the banking industry and its data services.
Few last year expected to read the news that Guido van Rossum, the creator of the Python programming language, had left retirement to join Microsoft’s Developer Division, particularly considering Microsoft’s attitude to open-source in the past.
Staying on top of the advances, both in theory and on the infrastructure side, is a considerable challenge when it comes to machine learning.
The Machine Learning Institute Certificate in Finance (MLI) – a comprehensive six-month course for individuals who work in or aspire to a career in machine learning in finance – attempts to keep on top of these trends.
We spoke to Dr. Paul Bilokon, head of faculty at the MLI, about how the course provides the rigorous skills needed for those working in a new era of quantitative finance.
Paul tells us that one of the challenges in designing the course is to keep on top of trends in the industry so that the MLI can provide the most relevant content and skills to its students.
Microsoft’s hiring of the founder of Python is an interesting indicator of the dynamics of this market, and part of the MLI course covers the cloud infrastructure aspect of machine learning.
Paul tells us that the course “offers a number of primers on the core technologies that are affecting the machine learning landscape” including Python and areas around deployment of these technologies and the structures that service the system.
But before one can begin to handle these deployment technologies, it is necessary to understand the theoretical mathematics behind machine learning.
Paul explains that the MLI “looks at the theory of the mathematics of how machine learning is done. How do we actually compute something, how do the neural networks work …?”
In terms of how this relates to the cloud providers, Paul explains that the course allows students to look at both “small examples from smaller data sets that can be run on laptops” as well as big machine learning problems that require access to more advanced equipment, although Paul emphasises that access to such equipment is not a prerequisite of the course – there is no requirement for students to have access to specific computational resources.
We covered a story last year about Google Anthos, and how the new cloud software platform has recorded substantial successes with customer wins including HSBC, PayPal and Lloyds. Although Google arguably provides the most user options, Microsoft still retains the largest share of the cloud market and is unlikely to give this up easily.
Paul comments that “the hiring of Guido van Rossum is a big move for Microsoft showing that they are investing in the Python infrastructure and playing a much more active role in the Python world.”
With Microsoft’s adoption of Python, it is competitively placed to offer one of the more advanced dashboard input interfaces for banks and financial institutions to use for machine learning experiments.
Nevertheless, Paul tells us that there are a number of other “interesting and independent products when it comes to running machine learning experiments” which demonstrates the diversity and dynamism of the industry.
One example is Weights & Biases which allows the user to track and visualize machine learning experiments. It is also framework and environment agnostic, meaning it can run on AWS, Azure, Google Cloud and so on.
Paul tells us that the MLI programme also tries to be framework agnostic: “we try not to bias towards a specific framework. Instead, we give students the kind of skills that they can use in any framework to achieve success in their machine learning projects.”
Towards the end of the course, students engage in projects of their own. Paul gives some examples:
“Anything that involves crunching large amounts of data and coming up with non-trivial conclusions based on that data. For example, it could be generation of alpha in a trading environment for a trading desk, or the question of how to best hedge a particular derivative. And in asset management, how do you best manage the portfolio.”
Although the MLI’s focus is financial, student projects are not necessarily limited to this:
“The major recent advances have been in protein folding and drug discovery, where machine learning has been critical – the protein folding problem was to a large extent solved last year with AlphaFold.
“Students also do projects where they come up with optimal decisions using reinforcement learning, and again, last year, there was a major advance from DeepMind (MuZero).”
Developments in the industry
These advancements in infrastructure and on the theoretical and algorithmic side (developments like Muzero, AlphaFold, and Transformers) demonstrate how quickly things are moving in the field.
Paul tells us that “we try to keep on top of these various theoretical advances and incorporate the ideas into the course and keep the curriculum constantly updated.”
Paul also directs us to developments within the academic financial literature:
“There have been recent developments by the likes of Blanka Horvath [Honarary Lecturer of Mathematics at Imperial College, London and member of the MLI faculty], who has done work on deep learning volatility and how to use neural networks to estimate fairly modern and sophisticated volatility models.
“There has also been a lot of work coming from JPMorgan by Hans Buehler in deep hedging, as well as papers published by veterans of the industry, such as Jay Cao, Jacky Chen, John Hull, and Zissis Poulos, as well as Petter N. Kolm and Gordon Ritter, on using reinforcement learning to hedge derivatives.
“And another advance in finance is the use of market generators – the use of neural networks for generating synthetic data which is especially useful for situations where the data is sparse.
“Our challenge as designers of this course is to stay on top of all of these advances. And to make sure that our students understand the latest advances, as well as the theoretical foundations of all this work.”
Radovan Vojtko from Quantpedia, a quant research company that specialises in the analysis of research papers related to algorithmic and quantitative trading, agrees that the usage of machine learning is a very strong trend in academic literature, he shared with #Disruptionbanking:
“The number of papers that utilize machine learning is steadily growing. According to Quantpedia, the most common and potentially promising ML application is in the analysis of alternative data sets (satellite data, sentiment data derived from social media, web traffic data, etc.).
“Alternative data sets are usually very large and hard to process. Machine learning helps find and extract a signal which would be otherwise hidden.
“Moreover, in Quantpedia, we see a continuous rise of machine learning algorithms used in the traditional and well-known anomalies such as momentum or numerous stock-picking strategies.
“The research papers that used classic linear regression are being revisited, and the comparison between easily overfitted linear models and ML often results in ML’s outperformance.” Radovan concludes.
Looking beyond theoretical developments, we also asked Paul to comment on the industry more broadly and to identify and account for the machine learning hotspots that we see across the globe.
From Singapore to Israel, and then closer to home in Europe we see hotspots where centres for maths, computer science and technology have grown up. In Europe, for example, there are banking hotspots in institutional cities like Frankfurt, Budapest, Krakow, Warsaw and Dublin where machine learning and mathematical ecosystems have grown to support the industry. One of the hotspots can be seen in Hungary, which Paul tells us has a strong mathematical community and a tradition of teaching non-trivial maths problems.
Another one of these dynamic ecosystems can be found in Israel, which Paul tells us has “an outstanding base for deep learning and a lot of very exciting startups, not just in finance but also, for example, in the medical applications of deep learning.”
This dynamism is evidenced by JP Morgan’s Technology Centre that was established in Herzliya, Israel back in 2017 – as we covered here – by Yoav Intrator, which works in areas like cybersecurity, blockchain, machine learning and cloud computing.
This location, Paul tells us, comes down to “the machine learning data science talent there and also the outstanding universities, like The Technion for example, giving access to leading research.”
Closer to home
When it comes to the UK, we asked Paul whether there is recognition of ML’s importance and the sort of investment in human capital that is seen in Israel.
Paul draws our attention to the All-Party Parliamentary Group on Artificial Intelligence : “which holds regular meetings to look at how ML and AI can be promoted and how it’s possible to develop them further.
“There are also many academic institutions in the UK fostering collaborations with Google DeepMind, for instance at UCL.
“And at Imperial College, where I teach in Antoine Jacquier’s MSc in mathematics and finance, machine learning courses have been introduced into the curriculum, not just in finance, but across the board. So people are doing more machine learning, which is now integrated into the curricula.”
Paul therefore thinks that “there is certainly the drive, and Boris Johnson has mentioned AI in several of his speeches more recently. So there is at least the understanding that this is a priority.”
With AI’s value demonstrated during the pandemic, we expect the Prime Minister might have moved on from the sentiments he expressed at the UN back in 2019 when he asked of AI: “What will it mean? Helpful robots washing and caring for an ageing population? Or pink-eyed terminators sent back from the future to cull the human race?”
Looking across the industry, great progress is clearly being made, and we are yet to see pink-eyed terminators! DeepMind (acquired by Google in 2014) has made headlines with its advances in protein folding – “a key scientific problem in drug discovery” through AlphaFold which can to a large extent predict the shapes of proteins.
Another breakthrough developed by DeepMind is MuZero which Paul tells us “is a fairly universal reinforcement learning system, which can help play humans in chess, shogi, go, and Atari Games: it’s a big advance in what they previously had with AlphaGo and AlphaZero.”
And MIT continues, as Paul tells us, to “occupy a key role in the USA in teaching the machine learning and general technological talent.”
Both DeepMind and MIT also upload lectures online and contribute to the learning ecosystem that exists there.
Why the MLI?
The competitive advantage of the MLI, however, is to keep up with these advances and provide a systematic and structured learning course utilising a first-rate faculty with some of the most experienced teachers in the field. As Paul explains, the MLI faculty “processes a lot of this information and has come up with good ways of teaching it, using the expertise and teaching experience we have.”
Paul explains that with all of the advances we are seeing, “the challenge remains to actually take someone who’s at a basic level in terms of their data science skills and to take them to the next level – that is not a trivial challenge.
“It ultimately involves a lot of strategic pedagogical thinking. And that is what we do on the MLI: we track all the recent advances but try to come up with ways of teaching people so that they get real skills – the real maths and technological expertise – to become competitive in the field. I think that is the biggest challenge for us.
Together with Matthew Dixon and Igor Halperin, Paul Bilokon has co-authored ‘Machine Learning in Finance: From Theory to Practice’, which has become the first graduate-level textbook on this exciting new subject. This textbook is now one of the recommended texts on the MLI reading list.
Upgrading your quant skills
While some amateurs might wish to attempt to learn a programming language like Python online for free, this will not necessarily take you far in the new era of quantitative finance.
Today’s industry is focusing more sharply on machine leaning and data science, which goes way beyond knowledge of a programming language. Paul explains that “there is a huge conceptual gap between just knowing Python and being able to do data science in a creative manner. For people to acquire the necessary skills to understand machine learning, the underlying maths, and to be creative with it, that would probably take a decade of self-study.
“We have decades of experience in this and the knowledge that allows you to skip this decade of self-study.”
While it might be possible for someone to quite easily learn rudimentary Python to become a “Pythonista”, Paul compares this to simply learning the basic moves of chess:
“You can learn chess, maybe in a day. You can learn the moves but that doesn’t mean that you’re going to be creative in playing against a grandmaster. Because you can learn the moves fairly quickly but it’s as good as useless when what’s ultimately important is what you can do with it. And that depends on strategy and tactics and the kind of stuff that people will learn over the years by playing experts.
“That’s the only way to really learn chess. And people confuse learning the simple moves or rules of Python with learning the complex strategy and tactics of machine learning and data science.”
This is what makes the MLI special – six months of exposure to the grandmasters of machine learning in a pedagogical and structured course that will take students to a competitive level in quantitative finance.
So, if you want to compete successfully in the financial industry, we recommend you check out The Machine Learning Institute Certificate in Finance (MLI) today.
Author: Curran Snell
#Pythonista #Cloud #MachineLearning #ArtificialIntelligence #Israel #Quantitative #Python #C++ #Microsoft #Google #MLI