MatLogica is a B2B Deeptech/Fintech, that has invented a breakthrough (patent pending) software solution that unlocks the full potential of modern CPUs for computationally intensive tasks, bringing speed-ups of up to 1000x (scroll down to the ‘Geeky part’ to see how we achieve that).
Our main focus to date has been on investment banking, where competition and new regulations require complex calculations at scale. MatLogica’s newest breakthrough in Machine Learning opens vast opportunities across other industries.
Our Progress To-Date
The solution is presently being licensed to Banks and Asset Managers to supercharge their financial analytics. A recent case study demonstrated how a major European Bank benefited with huge leaps in speed and accuracy in the calculation of prices and risk for their complex derivatives business, which enabled them to slash grid costs by half, increase software development turnaround by 3-4x and open the door for new revenue streams.
In 2021 MatLogica, has been recognised as a ‘Category Leader’ in ‘Chartis RiskTech Quadrant® for xVA Solutions: Analytical Components’. The company has joint benchmarks with Intel, with one demonstrating a 1770x speed up for xVA pricing. MatLogica has also received an innovation grant from Innovate UK and completed several funding rounds.
MatLogica Entering Machine Learning Space
In 2022 MatLogica has achieved another disruptive breakthrough! Our technology enabled our partner, prof. Roland Olsson, to design state-of-the-art neural network architectures for time series analysis (scientific paper available here). It is up-to 3x more accurate than the available cutting-edge methods and the training time is several times lower due to MatLogica’s technology.
Professor Olsson commented, “during my 23 years as a Machine Learning scientist, I haven’t encountered a technology better suited for neural net research than the one from MatLogica.“
Time Series Analysis is an incredibly powerful predictive tool, leverageable across numerous industries and is at the cusp of how businesses are making decisions in the new world order! By also focusing on the distribution of this additional solution, MatLogica will have access to a much wider range of revenue streams across a spectrum of business, industry, and scientific applications.
MatLogica’s objective is to become the go-to solution when it comes to high volume repetitive calculations and machine learning, enabling developers to easily unlock the full capacity of modern CPUs.
With the objective in financial markets clearly defined, entering the Machine Learning sector will deliver the scope to drive MatLogica’s revenues exponentially.
MatLogica will be undergoing a Series A round in September 2022. Register your interest and join the computational revolution!
The Geeky Part (explained in simple terms)
Due to modern CPU features as multithreading (utilising multiple cores), and vectorisation (each core processing 32 operations in one go), a CPU can perform about 9 trillion elementary operations per second (9 TFLOPS)* when using full capacity.
Popular object-oriented programming languages (such as C++ or Python) are not natively designed to utilise these capabilities. Taking advantage of multi-threading and vectorisation is a tedious, error-prone, and expensive task, requiring specialist low-level programming knowledge and is rarely used in practice. It can also lead to undefined behaviour that is difficult to troubleshoot, making it a non-viable option for financial institutions, where stability is of critical importance.
So, when it comes to intensive calculations or machine learning, the actual performance of a powerful machine is at 2-10% of its capacity.
This is what MatLogica revolutionises. Developers can continue using their favourite programming language and by plugging in our toolkit into their existing software, MatLogica will automatically generate efficient machine code that takes full advantage of vectorization and multithreading, enabling maximum performance from a modern CPU. It also generates highly efficient code to compute risk sensitivities (AAD), a prerequisite for machine learning, bringing unequalled capabilities for developing neural networks mixed with existing quant analytics.
* Using 56-core Intel® Xeon® Platinum 9282 Processor at 2.6 GHz clock frequency and AVX-512 vectorisation, float precision
The story begins in 2019 when the founder, Dmitri Goloubentsev, was developing a regulatory-driven computationally intensive xVA model for a major bank. With Dmitri’s geeky background, in-depth knowledge of the core mechanics of a CPU, and extensive experience developing financial models, he knew that popular software development tools were unable to leverage the full potential of modern CPUs. He had a Eureka moment when he mapped the mechanics of calculation-intensive algorithms with the layout of the CPU. In tandem with co-founder Evgeny Lakshtanov, they developed a toolkit that perfectly fits the market. The solution creates the most powerful calculation engine yet for automatic multi-thread safe parallelisation and automatic differentiation (AAD, Automatic Adjoint Differentiation), the most accurate methodology to compute sensitivities, widely used in finance and Machine Learning.