Manufacturing Bits: March 16

Tripping up neural networks
For years, Russia has been an active area in R&D.

In one example, Russia’s Skolkovo Institute of Science and Technology (Skoltech) has demonstrated how certain patterns can cause neural networks to make mistakes in recognizing images. Leveraging the theory behind this research, Skoltech can design defenses for pattern recognition systems that are vulnerable to attacks.

A subset of AI, machine learning is a technology that make use of a neural network in a system. In this system, the neural network crunches data and identify patterns. It then matches certain patterns and learns which of those attributes are important.

Many companies, governments and universities utilize machine learning in one form or another. It’s used for face recognition, search engines and many other applications.

There are several issues with machine learning, though. In machine learning, a system requires large data sets. If the data sets are insufficient, a system can generate questionable results.

Certain patterns can also cause these systems to make mistakes, according to researchers from Skoltech. These are akin to Turing patterns, which are found in nature. Machine learning is also vulnerable to adversarial perturbations. “(These are) small but peculiar details in an image that cause errors in neural network output. Some of them are universal: that is, they interfere with the neural network when placed on any input,” according to researchers from Skoltech.

These perturbations also present a security risk. For example, in 2018, Cornell University demonstrated a way to attack autonomous vehicles using computer vision. It deceived traffic sign recognition with malicious ads and logos.

Researchers from Skoltech have explored the theory that connects these so-called universal adversarial perturbations (UAPs) and classical Turing patterns. Apparently, these patterns were first described by Alan Turing, a mathematician and computer scientist who was influential in the development of theoretical computer science.

The nature adversarial perturbations are still mysterious for researchers. “This intriguing property has a long history of cat-and-mouse games between attacks and defenses. One of the reasons why adversarial attacks are hard to defend against is lack of theory. Our work makes a step towards explaining the fascinating properties of UAPs by Turing patterns, which have solid theory behind them. This will help construct a theory of adversarial examples in the future,” said Ivan Oseledets, a professor who leads the Skoltech Computational Intelligence Lab at the Center for Computational and Data-Intensive Science and Engineering (CDISE).

Skoltech is a cooperation of nine Russian universities and organizations. In 2011, the Skolkovo Foundation and the Massachusetts Institute of Technology (MIT) formed a partnership in several areas.

New superconductors
Skoltech, the National University of Science and Technology MISiS and the Russian Academy of Sciences (RAS) have synthesized a new and promising high-temperature superconductor–yttrium hydride (YH6).

Superconductors are devices that have zero electrical resistance, making them attractive for a range of applications. But superconductors must be cooled down to temperatures at or near absolute zero on the Kelvin scale to work. This, in turn, limits their applications.

Yttrium hydride ranks among the three highest-temperature superconductors known to date, according to researchers from Skoltech, MISiS and RAS. “Pressure‐stabilized hydrides are a new rapidly growing class of high‐temperature superconductors, which is believed to be described within the conventional phonon‐mediated mechanism of coupling. Here is the synthesis of one of the best‐known high‐TC superconductors—yttrium hexahydride. YH6 is reported, which displays a superconducting transition at ≈224 K at 166 GPa,” said Ivan Troyan from the Institute of Crystallography of RAS, in Advanced Materials. Others contributed to the work.

Yttrium hydride, a compound of hydrogen and yttrium, is part of the class of so-called rare-earth metal hydrides. The technology is used for various applications, including components used in nuclear microreactors.

Microreactors are a new class of advanced reactors. More than 20 U.S. companies are working on designs that are smaller than traditional nuclear reactors.

Oak Ridge National Laboratory (ORNL) and others are working on microreactors using these materials. A single microreactor generates 1 to 10 megawatts of electric power, according to the U.S. Department of Energy (DOE).

A single megawatt of electricity can power approximately 1,000 homes. “That means these systems could provide up to 100,000 homes with clean power—24 hours a day, 7 days a week—for 10 years without stopping,” according to the DOE.

Anti-counterfeit labels
ITMO University and St. Petersburg Academic University have developed an invisible anti-counterfeit label technology based on rare earths. These labels will help protect goods from being counterfeited.

Companies are looking for ways to protect their goods from being counterfeited. But many methods are temporary or expensive solutions.

Russian scientists have devised a new label technology, which is difficult to decipher. They are made from semiconductor materials with the help of lasers.

The label is based on the down conversion photoluminescence from erbium‐doped silicon. “For fabrication of these labels, a femtosecond laser is applied to selectively irradiate a double‐layered Er/Si thin film, which is accomplished by Er incorporation into a silicon matrix and silicon‐layer crystallization,” according to researchers in Advanced Materials, a technology journal.

This process in turn creates anti‐counterfeiting labels. “With a laser, we add ions of a rare-earth metal called erbium that create a unique image on a sticker made of a silicone nanofilm. To do that, we first make a lattice of holes on the film that are invisible to the naked eye. Some of these holes contain erbium ions, others don’t. When subjected to laser radiation, the holes with erbium change color – and thus they allow us to correctly read the image,” said Dmitriy Zuev, head of the project and assistant professor at ITMO’s Department of Physics and Engineering.

“Our labels are based on erbium ion luminescence, which is characterized by several parameters: intensity, wavelength, and radiative lifetime. A combination of these parameters allows us to create additional layers of protection. That’s why when you get the hidden image with an infrared sensor, you will also be able to read the information about the luminescence parameters. This provides an additional degree of protection,” added Artem Larin, a PhD student at ITMO’s Department of Physics and Engineering.