With Giotto, Artificial Intelligence Gets a Third Dimension


Reading time ( words)

The Giotto project, launched by EPFL startup Learn to Forecast, intends to revolutionize the way we use artificial intelligence. Drawing on the science of shapes, Giotto pushes AI forward by making it more reliable and intuitive in areas such as materials science, neuroscience and biology. Giotto is open-source and available free of charge on GitHub, and it’s already being used by some EPFL scientists.

Researchers use artificial intelligence to solve complex problems, but it’s not a transparent science: AI’s computational capabilities often exceed our understanding and raise issues of reliability and trust among users. “Algorithms are becoming increasingly complex,” says Matteo Caorsi, the lead scientist at Learn to Forecast (L2F). “It’s very hard to understand how they work and thus to trust the solutions they provide or predict when they might get things wrong.”

Shapes are hidden within data

To address this problem, L2F followed an intuitive approach based on the science of shapes. The result is Giotto, a free and open-source library that aims to revolutionize the way we use machine learning. “Humans understand shapes and colors better than numbers and equations,” says Aldo Podestà, the CEO of L2F, “which is why we think that we can use topology – the science of shapes – to build a new language between AI and users.”

Giotto offers a toolkit that uses algorithms inspired by topology to address some of the shortcomings of machine learning. Users don’t need to be fluent in advanced mathematics since Giotto is a turnkey method of revealing structures previously hidden within a dataset. “This new form of AI is based on graphs and their multidimensional versions, in other words, geometrical objects that can reveal essential structures within the data,” says Thomas Boys, a co-founder at L2F.

Until now, machine learning algorithms sought performance, even if that meant depriving users of a fuller understanding of the nature of the results. “Giotto helps identify the framework underlying all relationships among the data, and this allows users to understand the data better and extract meaning from them with greater accuracy,” adds Boys. The project is named for Giotto di Bondone, the 13th-century artist who first introduced perspective into painting. L2F hopes to usher in a similar paradigm change in data science by combining machine learning with topology.

New horizons

To develop Giotto, its creators worked with EPFL researchers who use topology every day. This includes Professor Kathryn Hess Bellwald, the head of the Laboratory for Topology and Neuroscience. “One of Giotto’s main advantages is that, because of its user-friendliness, it will be possible for scientists from all kinds of fields to use these tools as a regular part of their data science toolkit,” says Prof. Hess Bellwald. “This should lead to new insights in many different areas that one could not attain without Giotto.”

Learn to Forecast (L2F) was founded at EPFL in 2017. It aims to use artificial intelligence to address a wide variety of issues. The company raised three million francs via 4FO Ventures to develop the Giotto library, and it now has 25 employees.

 

Share

Print


Suggested Items

CES Press Kickoff Presentation

01/07/2020 | Nolan Johnson, I-Connect007
On January 5, Editor Nolan Johnson attended the CES press kickoff presentation “2020 Trends to Watch,” which was hosted by CES Vice President of Research Steve Koenig and CES Director of Research Lesley Rohrbaugh. Koenig and Rohrbaugh set the stage for the week with their presentation, answering the question, “What’s happening in the industry?”

NASA Sounding Rocket Technology Could Enable Simultaneous, Multi-Point Measurements — First-Ever Capability

10/21/2019 | NASA
NASA engineers plan to test a new avionics technology — distributed payload communications — that would give scientists a never-before-offered capability in sounding rocket-based research.

Worldwide Semiconductor Equipment Billings at $13.3 Billion in 2Q19; Down 20%

09/12/2019 | SEMI
Worldwide semiconductor manufacturing equipment billings reached $13.3 billion in the second quarter of 2019, down 20% from the same quarter of 2018 and 3% from than the previous quarter.



Copyright © 2020 I-Connect007. All rights reserved.