Axyon AI offers an AI platform specifically designed for asset management, with several interesting strengths for those approaching machine/deep learning applications
Autore: By InnovationOpenLab
For those who make financial investments, the difference between a good return and a disappointing one comes from knowing how to ‘read’ the market over time, much more than from some isolated intuition. It is therefore logical that Finance is one of the fields in which machine and deep learning technologies have found an immediate home: there is no shortage of financial data to feed to AI algorithms, and the returns are immediate and concrete. Of course, the right algorithms are needed, and developing them requires skills and experience. Although a relatively young company - it was founded in 2016 - Axyon AI, based in Modena, has acquired both the former and the latter.
"We decided early on to invest in research and development at the intersection of the two topics that interested us most, namely AI and finance," explains Daniele Grassi, CEO and co-founder of Axyon AI. And the revival of machine learning that characterised the mid-2010s provided the perfect opportunity to put this intention into practice. "Machine and deep learning," Grassi recounts, "made a decisive leap in those years from a computational point of view, but Finance maintained its own specificity compared to other fields. For example, in machine vision, an algorithm that ‘learns’ to recognise certain objects remains effective for a long time, perhaps even forever. In Finance this is not the case: what works today may not work tomorrow".
The specificity of Finance has become an opportunity for Axyon AI, which has developed an AI platform tailored to the problems and complexities of financial markets. Complexities that are also technical, from the point of view of machine learning: "In analysing financial data," explains Grassi, "it is essential to have a codified and rigorous process that allows you to train, evolve and select machine learning models according to the data you have and the results you want to obtain. Which in our case are rankings of financial assets with respect to their performance over various time horizons".
The one developed by Axyon AI is in fact a machine learning process that can safely be described as ‘industrialised’. The Axyon AI Platform is not based on a few ad-hoc trained models but on a sort of ‘assembly line’ of machine learning by time series, in which various supervised ML models are applied to the input information to select the most suitable ones. These, in an immediately subsequent step, are combined to produce the final meta-model that will then yield the desired predictions.
"Our basic idea is that machine learning in finance can be standardised if you codify it in a rigorous process," Grassi summarises. Axyon AI's platform does not exclude customisation, but always in an assembly-line logic: by assembling and optimising machine learning components that are already in place and not by starting from scratch every time. "We are neither consultants nor craftsmen of AI - Grassi summarises: "These two roles have extreme value in some specific cases, but we operate on the part of machine learning that can be made scalable".
A well-characterised Axyon AI also has a well-defined market at the moment: primarily medium and small asset managers, private banks, hedge funds, family offices. "All those," comments Grassi, ‘"for whom using AI today is not just about having an extra edge but a matter of survival, not only from the point of view of performance and risk management but also from that of the perception of end customers".
Axyon AI can count on three technological ‘pluses’ that also have an important appeal to the market. The first is a long-standing relationship with Cineca for supercomputing environments applied to AI, through a number of targeted projects that have enabled the Modena-based company to gain experience that has proved very useful in the design of a platform that is partly on-premises and partly dynamically distributed, depending on computational convenience, on various cloud providers.
Another interesting element is that the ML models of the Axyon AI platform are ‘explainable’ and therefore not ‘black boxes’ whose prediction logic is not explicable. In times of the AI Act Europe, this is not a detail to be underestimated. Also, for obvious privacy issues Axyon AI does not work with personal or private data provided by a user company. "It has only happened, but rarely, that some customers passed us public data provided by their commercial providers and that for various reasons could only be processed on-premise," comments Grassi.
If we combine a technology that appears decidedly solid with an equally robust financial endowment - in January Axyon AI closed a new funding round of 3.9 million euros - and the support of names such as CDP Venture Capital, ING Ventures and UniCredit, we can see that the Modena company has all it needs to grow well and organically. Especially now that AI is a ‘must’ and no longer something to be explained first and then, hopefully, sold.