NOTE! This site uses cookies and similar technologies.

If you not change browser settings, you agree to it. Learn more

I understand

Learn more about cookies at :

May 28, 2020 | Towards neural networks that can withstand attacks

© monsitj -

At a time when artificial intelligence is making inroads into our everyday lives, List, a CEA Tech institute, is driving advances in cybersecurity that could result in more robust neural networks. CES 2020 provided an opportunity to showcase two demonstrator systems.

From autonomous vehicles to video surveillance, the potential uses for AI in our everyday lives are vast. Hackers, however, are rapidly coming up with attacks on these new applications for AI. Most attacks take advantage of the vulnerability of deep learning systems to disrupt the signal (image, sound) and "trick" or, in some cases influence the AI's decisions. List, a member of the Carnot Network, develops trustworthy AI. The institute recently came up with some effective ways to fend off attacks.

Specifically, they intentionally introduced random modifications of the neural activations from the earliest stages of the neural network design process. The goal is to scramble the network during the learning phase as well as during operation. The researchers' approach enables the machine to remember only the relevant parts of incoming information and to not be fooled by an attack. An alternative for existing machines that need additional protection is to introduce the noise directly into the incoming signal. These modifications are made using an "overlayer" that mitigates or neutralizes the effects of an attack.

The loss of performance that occurs when the defect is introduced is offset by the fact that the system is more robust and can better withstand attacks. A demonstrator presented at CES 2020 was well received. An article* was also published very recently in Neural Information Processing Systems, a major scientific journal in the field of AI.

*Pinot, R., Meunier, L., Araujo, A., Kashima, H., Yger, F., Gouy-Pailler, C., and Atif, J. (2019). Theoretical evidence for adversarial robustness through randomization. In Advances in Neural Information Processing Systems 32, pp. 11838–11848.

Read article at



May 19, 2020 | Non-destructive testing: CIVA now even better!

CIVA List 250
psdesign1 -

CIVA*, developed by List, a CEA Tech institute and member of the Carnot Network, is the software of choice for non-destructive testing. With a host of new features in the 2020 version, CIVA's prospects for remaining the world's leading NDT software suite are excellent.

The 2020 version of CIVA has been released. This non-destructive testing (NDT) software suite is sold in more than 40 countries, mainly in Europe, Asia, and North America. It can handle all of the main inspection techniques used in the manufacturing and maintenance of industrial components. This new major release incorporates the latest advances in simulation and analysis for NDT.

The models that power CIVA have been updated and optimized, and the software can now handle increasingly complex defects. Its hybrid models offer the precision of finite-element numerical models and the performance of semi-analytical models. In addition, new industry-specific plug-ins have been developed to do things like calculate the dispersion of guided waves in underground pipes. There is even an environment designed specifically for the optimization of hot-tapping inspections.**

Last, but not least, the 2020 release brings new capabilities to all NDT techniques currently available in CIVA. Now CIVA's ultrasound, Eddy current, guided wave, X-ray, and other tools can benefit from additional simulation-driven data processing and database generation features for statistical studies, performance assessments (sensitivity analysis and detection probability), and statistical (machine) learning.

Splash CIVA NDE 2020

*CIVA is sold exclusively by EXTENDE -

**A type of pipe connection frequently encountered in the energy sector (nuclear, Oil & Gas) that is difficult to inspect.

Read article at



May 15, 2020 | A portable, versatile radiological source detector

Gadosphere List 250
Oneinchpunch – Adobe Stock

​In the fight against nuclear terrorism, not all weapons are created equal. To help level the playing field, researchers recently developed a portable detector that is sensitive to both gamma and neutron radiation. Now a single device can be used to detect both types of radiological sources, even in crowds!

The researchers started with an existing neutron radiation detector, which they adapted to form a system made up of a gadolinium plate capable of capturing neutrons that have been slowed by a plastic shell around the plate. The plastic, which is sensitive to gamma radiation, slows the neutrons down. The neutrons are then captured by the gadolinium, emitting gamma rays in the process. The challenge was to distinguish between the gamma rays coming from outside sources (whether environmental or artificial, including potential threats) and those generated by the neutrons captured by the gadolinium plate.

Researchers at List, a CEA Tech institute, tackled the problem with signal analysis methods capable of identifying the unique characteristics of the gamma rays emitted by the gadolinium plate. Here's how: First, gamma rays' energy varies depending on their origin. Second, the fact that the gadolinium simultaneously emits rays in several directions when it reacts (coincidence). And, last, the time that elapses between when the nuclei are pushed back in the plastic by the neutrons, and when they are captured by the gadolinium.

The different possible scenarios were simulated to identify the best detection windows, for example. An initial prototype of the physical detector was built, and its capacity to discriminate between the gamma rays emitted by the neutrons captured by the gadolinium was confirmed in lab tests.

Read article at



May 11, 2020 | Participate in Gazelle Accelerator, the Call of Interest to SMEs and Start-ups

Participate in the following Call of Interest to SMEs and Start-ups, as part of the Gazelle Accelerator programme, business acceleration activity of EIT Manufacturing.

This Call of Interest is looking for promising SMEs & Start-ups who will answer to challenges that manufacturing companies have expressed in the following industry 4.0 technologies:

  • Analytics & Artificial Intelligence
  • Internet of Things
  • Simulation & Virtual Reality
  • Sustainable Manufacturing

Gazelle Accelerator will encourage, as well, proposals that are focused on providing solutions in these and other technologies, to help industries to face economic impact, and accommodation of new emerging-tech-enabled business models, due to coronavirus outbreak.

Gazelle Accelerator aims at supporting existing technology-based companies, SMEs and Start-ups, by accelerating their international business and innovation capacities.

Selected SMEs and Start-ups will benefit from a programme that includes activities that goes from business coaching with experts, access to finance, access to market; to facilitating business development in other European countries.

Call for interests will be opened until May 25th.

Participate in Call of Interest and Pitch your solution to corporates at end of June 2020.

Do not miss this opportunity!

More info



May 5, 2020 | New version of LIMA shifts to deep learning

neurones 250
Adobe Stock

The LIMA language analysis engine now has deep learning modules that can analyze 60 languages with state-of-the-art performance.

When a user enters a search term into a search engine, the text is thoroughly analyzed: every word, phrase, and sentence is detected and processed by special software. This analysis must be done before a search can be performed. However, it is also a prerequisite for other applications like automatic summarization and translation. List's language analysis engine, LIMA, is already widely used. Integrating deep learning modules into the software has made it even more powerful.

The latest advances in neural networks and a corpus of annotated texts in different languages provided by the Universal Dependencies cooperative were used to make the software more efficient, expand the number of languages supported, and add three learning modules. The first module segments text into words and sentences; the second performs a morphological, lexical, and syntactic analysis, and the third annotates the named entities identified.[1]

The previous version of LIMA could analyze six languages (English, French, German, Spanish, Portuguese, Chinese, and Arabic). The new version, called Deep LIMA, can analyze more than 60 languages with performance at the state of the art.

[1]An international cooperative project to create treebanks of the world's languages (

Read article at