COVID diagnostics by the sound of coughing, computer vision trickery with stickers and other news from the world of AI

We strive to inform readers not only about developments in the bitcoin industry, but also to talk about what is happening in related technological areas – cybersecurity, and now in the world of artificial intelligence (AI).

Hodlmonks has collected the most important news from the AI ​​world over the past week.

Key

  • Researchers tricked OpenAI’s advanced computer vision algorithm with a handwritten note.
  • The algorithm generated a video clip from music album covers.
  • Activists from California have filed a lawsuit demanding to ban the use of Clearview AI products in the state.
  • British scientists have developed an algorithm for screening COVID-19 that can diagnose the virus by analyzing the sound of a cough.

OpenAI computer vision tricked with handwritten notes

Researchers at the OpenAI Machine Learning Lab have found that their modern computer vision system can be tricked with tools no more sophisticated than a pen and a notebook.

To deceive the program and force it to incorrectly identify what it sees, it is enough to write down the name of the object and stick it on another.

An example of cheating an algorithm using a sticker. Data: OpenAI.

“We call these attacks typographic. By leveraging the model’s ability to reliably read text, we found that handwritten photographs can trick the model, ”wrote the OpenAI researchers in a blog post.

They noted that such attacks are similar to “Adversarial images”that can fool commercial machine vision systems, but are much easier to create.

Adversarial imagery is a real threat to systems that rely on machine vision. For example, researchers have shown that they can trick Tesla’s self-driving car software to change lanes without warning by simply sticking certain stickers on the road. Such attacks pose a serious threat to a variety of artificial intelligence applications, from medical to military.

The OpenAI developers noted that this is an experimental model and is not being used commercially. In the future, researchers will work to eliminate such vulnerabilities.

British scientists create AI tool to diagnose COVID-19 by the sound of cough

Scientists at the University of Essex have created a COVID-19 screening tool that can diagnose the virus by analyzing the sound of a cough. The app, called DeepCough3D, uses AI to analyze audio recordings of coughs at frequencies that humans cannot hear, writes The Next Web , citing a full study in the journal IEEE Transactions on Service Computing .

Researchers tested the algorithm on more than 8,000 samples of audio recordings of patients coughing in Spanish and Mexican hospitals. Of these, about 2,000 people were sick with coronavirus infection, and the rest did not.

According to the developers, DeepCough3D determines with 98% accuracy whether cough samples belong to people with COVID-19.
Lead researcher Dr. Javier Andreu-Perez believes this tool has the potential to make a difference in the fight against the pandemic.

“It [the tool] is much less invasive than most other COVID-19 tests and also offers quick results, paving the way for pre-screening test solutions,” said Andreu-Perez.

The University of Essex team says their study is different from others as it has proven to be highly accurate in detecting infection in thousands of clinically confirmed samples that have been tested in certified laboratories.

Researchers have also used this tool to classify coughs into three levels of severity, which can help healthcare professionals allocate resources such as ventilators.

In the future, scientists plan to conduct interventional research with the technology and work to certify the instrument so that it can be used in hospitals to diagnose coronavirus infection.

Activists demanded to ban Clearview AI in California through court

A group of activists has filed a lawsuit in a California court to terminate Clearview AI’s operations in the state. Human rights activists demand to prohibit companies from collecting any biometric data, as well as to delete all available personal information of Californians.

Today @ConMijente @NorCalResist and 4 activists filed a lawsuit in CA against Clearview AI.

These activists have criticized police and ICE policies and have openly supported Black Lives Matter. JFL and @braunhagey are proud to represent them. 1 / https://t.co/ZQLPABiOQ7

– Just Futures Law (@JustFuturesLaw) March 9, 2021


The plaintiffs allege that Clearview AI violates the privacy rights of California residents by collecting their photos from the web and extracting biometrics from them without permission.

The lawsuit also alleges that the company’s facial recognition technology is helping the government control protesters, immigrants and people of color.

Today, more than 2,400 US police departments use Clearview AI software. Plaintiffs argue that California law enforcement is still using it to find offenders, despite the fact that several cities in the state have banned the use of facial recognition technology.

AI created a psychedelic clip from track covers

Spanish artist Bruno Lopez created a music video by training the algorithm on thousands of Spotify album covers.

He first created a Python script that dumped album art for every track featured on Spotify’s official editorial playlists. The result was a dataset of approximately 50,000 images.

This data was used as training data for the Nvidia StyleGAN2 architecture. After a few days of training, the model learned how to create her own album covers.

Lopez then created another Python script to organize the results according to the colors and faces featured on the covers. He then animated the images to the beat of the Sound Stabs song “Interference.”

Ericsson uses AI to help hospitals treat COVID-19 patients

Ericsson Corporation announced a partnership with Telia and Salgrenska University Hospital to create planning models and predict hospital needs so that it can better prepare for admission.

The research project uses advanced analysis and artificial intelligence algorithms, combined with medical information, to help a hospital decide what resources it will need. This should help them better manage the demand for their services from coronavirus patients.

Ericsson said the data could be used to improve estimates of how many people with COVID-19 will be hospitalized. This prognosis can then be used to verify that there are sufficient equipment and hospital beds available to treat them.

Optibus launches AI platform for planning public transport routes in cities

Israeli startup Optibus has unveiled an AI-powered SaaS platform for public transport scheduling.

The service can take into account demographic data showing how rerouting could affect residents’ access to public transport, handle workloads such as scheduling and route planning, which the developers say reduce costs and save time.

Optibus also visualizes geographic information when planning routes for cyclists to get to or from public transport.

The company’s solution has already helped optimize transportation in more than 450 cities around the world, including New York, Los Angeles, London, Melbourne, Brasilia, Hong Kong and Singapore. Now it is available as a cloud application for everyone.

The functions of measuring heart rate and respiratory rate using the smartphone camera became available in Google Fit

Google has begun rolling out camera-based heart rate and respiration tracking features in the Fit app. Both functions work using computer vision algorithms and make it possible to measure heart rate and respiratory rate without wearable sensors.

To use the heart rate tracker, the user needs to place their finger on the back camera of the smartphone. The app will immediately start counting your heart rate, observing the smallest changes in skin color.

To use the respiration tracker, the user needs to make sure the front camera has a good view of the torso and will be stationary for 30 seconds. The function will then calculate the number of breaths per minute by measuring small chest movements.

Google Fit’s new features are only available to Pixel smartphone users, but the company promises to roll out trackers to more devices in the near future.

Also on Hodlmonks:

  • The Chinese government has unveiled a five-year plan to accelerate the development of advanced technologies, including artificial intelligence and quantum computing.
  • Scientists at NVIDIA and Harvard University have developed the AtacWorks deep learning toolkit that accelerates analysis of the human genome from two days to half an hour.
  • A study by analyst firm KMPG showed that the pandemic has accelerated the adoption of artificial intelligence systems in enterprises.

Leave a comment