Machine Learning Startup L2F Raises CHF 3m to Launch its Opensource Project Giotto

What is the L2F mission and core innovation?

In many ways artificial intelligence works like a glorified linear regression: in regression you fit a line to explain data, in A.I. you fit a more intricate shape to it. But the principle is the same, you use shapes to extract meaning from data. 

As mathematicians, we asked ourselves whether the science of shapes itself, called topology, could be of any assistance to A.I. And it was key. Finding meaning in data is so much easier when you have the right perspective, and shapes allow humans and computers to speak a common language. When we realized that there was no easy to use software to interface machine learning with topology, we started creating it, and Giotto is our open-source project that makes these methods available to every machine learner. 

Giotto is a bit like having an operating system on a computer, only it is for operating machine learning. When you have the right interface to use the hardware, you can do so much more. With shapes as interface, A.I. becomes immediately more relatable and more powerful, both for the data-scientist to build better models, but also for the decision maker to make a conscious decision. 

How will the 3-million-investment from 4FO Ventures help achieve your vision?

The work will climax October 21st, when Giotto will be released. The investment made this project possible, and will continue to benefit the improvement of our technology. 

One part of our job at L2F has been to develop the open source machine learning Giotto, and the other has been to develop L2F’s commercial software. L2F now offers software modules, which allow you to build your own custom A.I. software based on Giotto. The combination of modules adapts to a huge variety of problems and serves an equally diverse clientele.  Our goal is to push these on a global scale. 

 

South Africa has an incredibly dynamic and exciting start-up ecosystem. In my view it is of capital importance for tech entrepreneurs to understand the potential as well as the needs of emerging markets. In South Africa, I was able to measure the maturity of their technology and the current state of the competition. 

 

Being in the state of discovery is extremely motivating for us, and working with machine learning will definitely add to that feeling. When we started L2F, we felt that machine learning could be the solution to some really important problems. It’s an amazing time to be working with Artificial Intelligence, you can feel the excitement to push the limits of both the technology, and science itself.  

In our early days, when we were working on the Kaggle New-York taxi competition, we realized how hard it was both to understand and to focus the learning process of these A.I. algorithms. With our mathematical background, we wanted to use some tools to improve our intuition, and it meant discovering shapes that the dataset was implicitly drawing. “BIG” data lives in high-dimensional spaces, where it is impossible to represent curves like we are used to, and so we had to invent our own ways to imagine what was happening there. Finding the shapes in data is a great way to explain reality, it is a bit like finding behaviors in your sub-conscious, you don’t see the patterns at first, but with a new paradigm you can explain reality with more precision.

We believe that A.I. can do great things, but there will be a trust issue if we cannot explain its results. If you ask a data-scientists to justify or challenge these patterns, they will probably struggle to give a convincing answer. And this is the problem we are trying to address at L2F, we believe every data-scientist needs to have back and forth conversations with A.I. algorithms to be able to understand, challenge and explain the results. Giotto provides a new experience of machine learning, by immersing the user into high dimensional data spaces with an intuitive experience of topology. 

We were extremely lucky to find that a professor of ours at the EPFL, Prof. Kathryn Hess was actually pioneering techniques of topological data analysis to explain the functioning of the brain. Since that moment, we have been working together on the tools and methods to exhibit shapes in data.