Big data and artificial intelligence are revolutionising the world of FinTech. Italian company Axyon AI has developed a proprietary platform that uses machine learning techniques to deliver high-performing investment strategies, and a recent SHAPE project has helped them develop this even further by harnessing the power of HPC clusters.
With the advent of big data, the increasing use of AI algorithms in finance and the consequently fast- growing need for computational power, it has become necessary for FinTech companies to use large scalable HPC systems able to handle and explore a wide array of machine learning models (e.g. deep neural networks) and to remain up-to-date with state-of-the-art methods.
The application of deep learning models to time-series forecasting and modelling has been studied for decades, though only recently have they been successfully exploited for their predictive and explanatory power in many scientific and business sectors. One of the main reasons for this late development is that extremely high computational power (e.g. an HPC cluster) is required in order to train deep learning models. The other crucial ingredient in this success has been the ability to access huge amounts of data, which has now become cost-effective due to technological advancements in data collection and storage.
Rather than providing minor incremental improvement, this new possibility may be game-changing for many FinTech startups working on areas in the financial industry where even a marginal improvement in model accuracy may translate into enormous value for their customers, being them individuals, companies or financial institutions. Axyon AI is a company that partners with asset managers and hedge funds to deliver consistently high-performing AI-powered investment strategies. This is done by leveraging the technology they have developed over the years that applies machine learning techniques to financial time series.
The founders of Axyon AI took their first steps into the FinTech space in 2014 after striking up a collaboration with a hedge fund, applying technologies such as genetic algorithms to the foreign exchange market. Later on, they started a partnership with the University of Modena, a centre of excellence for artificial intelligence. “This was the driver that pushed us towards using deep learning as our main technology,” says Jacopo Credi, co-founder of Axyon AI.
In late 2016, Axyon AI was accepted on to an acceleration programme run by ING Bank in Amsterdam, where it began growing and improving its platform. Following this, they strengthened their presence in the asset management space by partnering with a number of asset managers. More recently, they embarked on a PRACE project as a way of continuing their collaborations within the world of academia. Specifically, they were eager to partner with an HPC consortium like CINECA in order to improve the scalability of their platform and their resource management.
Lead AI engineer Riccardo Folloni explains further: “Our platform runs deep learning jobs which require a very high computational workload. In the past, we had little knowledge of HPC systems and used to run our jobs on our own systems. This project with PRACE has given us the opportunity to scale up our system to HPC platforms and infrastructure. We were able to work with an expert to modify our platform so that it would automatically scale to use all of the computational power of an HPC infrastructure.”
Specifically, Axyon AI wanted to maximise the efficiency of accessing the different types of remote computational resources available to their proprietary machine learning platform, without losing the flexibility provided by their in-house compute power. Nowadays, this is a mandatory requirement for FinTech companies that are working with proprietary data that cannot be uploaded to cloud systems.
“This wasn’t just an experiment for us to try out powerful computers. I’m proud to say that we have managed to implement everything we learned into our workflow.”
Jacopo Credi, co-founder of Axyon AI
Many companies running projects under PRACE’s SHAPE programme have simulation software that they are looking to launch on a larger scale. The project run by Axyon AI differed from many of these projects as they already had an advanced understanding and techniques for managing workload, but just needed to take their platform to the next level to be able to exploit HPC resources.
The help of an expert from CINECA had a profound effect on the company. “An HPC specialist came and worked with us in our office and used his deep knowledge of the CINECA infrastructure and our problems to really bring the project to the next level,” says Fabio Franzoso, Head of Engineering at Axyon AI. “We were then able to try out our new system on the actual clusters. So in terms of expertise and infrastructure, we were provided with everything we need to develop the project.”
Credi and the rest of the team at Axyon AI are pleased with the project’s impact. “This wasn’t just an experiment for us to try out powerful computers. I’m proud to say that in the months following the conclusion of the project, we have managed to implement everything we learned into our workflow,” says Credi. Folloni adds that they are continuing to develop the new infrastructure so that they are able to use several HPC systems.
Credi and his colleagues are positive about the idea of another PRACE project if a suitable call arises. “Overall, this has been a great experience for the company,” he says. “The computing hours and the expertise made available to us has been invaluable.