An elevator mistook a corgi for a scooter, an AI algorithm was granted patent‑filing rights, and other AI news

We aim to inform readers not only about developments in the bitcoin industry, but also about what is happening in adjacent technology spheres—cybersecurity, and now the world of artificial intelligence (AI).

ForkLog has gathered the week’s most important AI news.

The Russian Ministry of Economic Development defines what counts as AI projects

The Russian Ministry of Economic Development defined criteria for AI projects.

According to the document, AI includes solutions aimed at the development, creation and deployment of technologies in the following areas:

• computer vision;
• natural language processing;
• speech recognition and synthesis;
• intelligent decision-support systems;
• advanced AI methods.

Such projects may qualify for grants from the Innovation Support Foundation and the Skolkovo Foundation, with a total amount of 12 billion rubles.

Representatives of organisations have already launched a competition for the relevant programs.

Face-recognition fare payment rolled out on the Moscow Metro

The Moscow Metro launched testing of fare payment using facial recognition Face Pay.

The capital’s transport department explained how the service works. First, the system scans the passenger’s face with a camera installed at the turnstile. It then compares the image with biometric data the person has on file with their bank and deducts the fare from their account.

In the first phase, the experiment was rolled out on the Filyovskaya Line for 1,000 volunteers.

“We will gradually connect all lines to the test and expand the focus group, and by the end of the year the service will be available on all stations, as we promised,” the mayor’s office said.

Media: Chinese regulator delays review of Nvidia-ARM deal

The Chinese antitrust regulator delayed review of Nvidia’s and ARM’s deal for an indefinite period.

Regulators have been in talks with the companies since early 2021, but an official review has yet to begin. This is likely related to the ongoing US-China trade war.

Earlier, the European Commission suspended review of the merger of the companies, as the regulator went on vacation. British authorities also intervened in the multi-billion deal on national-security grounds.

Microsoft developed its own language model for the Bing search engine

Microsoft developed the language model Make Every Feature Binary (MEB) with 135 billion parameters and integrated it into the Bing search engine to improve results.

It analyzes queries and helps identify the most relevant pages on the web. MEB performs this task alongside other ML algorithms embedded in the search system.

According to company representatives, the language model handles 100% of Bing queries across all regions and languages.

“This is the largest universal model created at Microsoft, and it demonstrates an excellent ability to remember facts with robust continual learning on vast amounts of data,” the blog says.

For comparison: Google’s language model BERT, used in its search engine, contains 340 million parameters, and OpenAI’s GPT-3 has 175 billion parameters.

Yandex’s autonomous-driving business valued at $7 billion

Yandex’s autonomous-vehicle division valued at $7 billion and named among the leaders in the global market.

According to a Morgan Stanley report, sustaining this status requires increasing investments in this direction.

Researchers use AI to create a 3D map of a macaque brain

Researchers from the University of Science and Technology of China created a 3D map of a monkey brain with 1 µm resolution using artificial intelligence.

Scientists introduced a genetically modified virus into the macaque, which causes neurons to express a glowing protein. After the animal’s death, they produced detailed images from 250 brain slices.

Together with developers from MIT and the University of California, Los Angeles, they added AI that stitched the photos into a single image and compressed it.

According to the researchers, this method could be applied to other organs and body parts.

Australian court rules AI can patent inventions

Artificial intelligence can be an inventor under Australian patent law. The precedent was set by a local federal judge.

Shortly before, British researchers filed patent applications for two inventions attributed to the neural network Dabus in several countries. Regulators, however, rejected the applications, stating that the inventor must be a human, not a machine.

Nevertheless, the Australian court overturned the patent office’s decision and sent the application back for reconsideration.

In China, an object-recognition system in an elevator mistook a dog for a scooter

In Shandong province a lift identified a corgi dog as an electric scooter. It told the owner that entry with a mobile vehicle was prohibited.

The object-recognition algorithm proved insufficiently trained.

Researchers create a service to predict future technological breakthroughs

MIT researchers have developed an online system powered by AI to predict major technological breakthroughs.

They analysed 97% of the US patent system. This data underpinned an algorithm that found software to be the fastest-growing domain.

The developers claim their system will give entrepreneurs, researchers, investors and policymakers hints about future opportunities in technology.

Enthusiast builds a smaller version of OpenAI’s DALL-E that generates images from text

The developer known as Boris Dayma created the DALL-E mini model for generating images from any text description and published an interactive demo.

The project is based on OpenAI’s original DALL-E, but its architecture is 27 times smaller. According to the author, it used far fewer computing resources and took three days to create.

The source code is available on GitHub.

Read on ForkLog:

• The Pentagon tested a decision-making system for competing with Russia and China.
• Google developed a mobile processor with tensor cores for machine learning.
• Amazon launched a system for tracking delivery drivers with AI cameras.
• DeepMind developed a universal architecture for creating artificial intelligence.
• The neural network learned to fool identification systems using a “universal” face.
• Apple deployed AI to detect illegal content on iPhone users.

Subscribe to ForkLog news on Telegram: ForkLog AI — all AI news!