Saturday, November 26, 2022

World No 1 Tech News Website

Perceptron: Face-tracking “Earables,” analog A.I. chips, and accelerators for accelerating particles

Must read

fatima khan
fatima khanhttps://myelectricsparks.com/fatima-khan/
A brand new writer in the fields, Fatima has been taken under my electric spark's RGB- rich and ensures she doesn't engage in excessive snark on the website. It's unclear what command and Conquer are; however, she can talk for hours about the odd rhythm games, hardware, product reviews, and MMOs that were popular in the 2000s. Fatima has been creating various announcements, previews, and other content while here, but particularly enjoys writing regarding Products' latest news in the market she's currently addicted to. She is likely talking to an additional blogger with her current obsession right now.

The field of research in machine learning, as well as A.I., which is now a significant technology for almost every business and industry, is much too extensive for anyone to comprehend the entire. This column, Perceptron, is designed to gather some of the most critical research and papers of the past -especially regarding artificial intelligence- and then explain the significance of these findings.

An ” earable” which uses sonar to detect facial expressions was just one of some of the ideas that caught our attention during the past couple of weeks. It also featured ProcTHOR, a framework developed by the Allen Institute for A.I. (AI2) that creates procedurally generated scenarios that can be applied to teach robots in real-world environments. Other highlights include that Meta developed an artificial intelligence system that can predict the protein’s structure based on an amino acid sequence. Researchers from MIT have developed innovative hardware that claims to provide speedier computation for A.I. and requires less energy.

“Earable ” created by a team from Cornell, is a bit like an expensive pair of headphones. The speakers transmit acoustic signals to the sides of the wearer’s face. Meanwhile, a microphone picks up the barely detectable reflections caused by the lips, nose, eyes, lips, and other facial characteristics. The “echo profiles” enable the eye to detect movements such as eyes moving upwards and eyebrows raising and squinting, which are then processed by an A.I. algorithm that transforms into full facial expressions.

Image Credits: Cornell

The earpiece has some limitations. It’s only able to last for 3 hours with battery power. It must transfer the processing to smartphones, and the echo-translating A.I. algorithm has to train with 32 hours of face data before it can begin to recognize facial expressions. The researchers claim that it’s a sleeker experience than the conventional recorders used for animations in movies or T.V. games. For instance, for the game of mystery L.A. Noire, Rockstar Games constructed a set-up with 32 cameras trained to look at each actor’s face.

Maybe one day, Cornell’s earable might be used to make animated humanoid robots. However, those robots have to master the art of maneuvering the room first. Fortunately, AI2’s ProcTHOR is a step (no joke intended) into this area by creating hundreds of custom scenes, including libraries, classrooms, and offices where the robots are simulated to complete tasks like lifting objects or moving furniture.

The scene’s concept, including realistic lighting and an unspecified subset of a vast assortment of surface materials (e.g., tiles, wood, etc.) as well as household items and objects, is to expose simulation robots to as wide a variety as they can. It’s a well-established concept of A.I. that virtual environments’ performance can boost real-world systems’ performance. Independent car companies like Alphabet’s Waymo can simulate entire cities to refine their vehicles’ behavior in real-world settings.

Image Credits: Allen Institute for Artificial Intelligence

In the case of ProcTHOR, AI2 claims in an article that increasing the training locations continuously improves performance. This is good news for robots bound to work, homes, and other places.

Of course, the training of such systems takes lots of computing power. However, that may not be forever. Researchers from MIT claim to have created an “analog” processor that can be used to create an ultra-fast network composed of “neurons” and “synapses,” which could be utilized to complete tasks such as recognizing images and transliterating languages, and so on.

The processor used by the researchers uses “protonic programmable resistors” arranged in an array to “learn” skills. The increasing and decreasing of the conductivity of the resistors mimics the weakening and strengthening of synapses between neurons within the brain as a part of the learning process.

Conductance control is achieved by electrolytes, which regulate the protons’ movements. When more protons are moved into a channel within the resistance, the conductance amount rises. When the protons are eliminated from the resistor, the conductance decreases.

Processor on a computer circuit board

The processor on a computer circuit board

A non-organic phosphosilicate glass can make this MIT team’s process extremely speedy due to its nanometer-sized pores, whose surfaces offer the ideal pathways for the diffusion of proteins. Additionally, the glass can run at room temperature and isn’t affected by proteins as they travel through the pores.

“Once you have an analog processor, you will no longer be training networks everyone else is working on,” the lead writer and MIT postdoc Murat Onen quoted in the press release. “You can train networks with unimaginable complexity that no other could afford to beat, so you will be able to outdo the rest of them. This isn’t a speedier car it’s a spacecraft.”

When it comes to acceleration, machine learning is currently utilized to manage particle accelerators, at a minimum, in a practical way. The Lawrence Berkeley National Lab two teams have demonstrated that ML-based simulations of the full beam and machine provide the most precise predictions, up to 10 times more accurate than statistical analysis.

Image Credits: Thor Swift/Berkeley Lab

“If you can predict the beam properties with an accuracy that surpasses their fluctuations, you can then use the prediction to increase the performance of the accelerator,” said the laboratory’s Daniele Filippetti. It’s no easy task to model all the technology and physics involved, but it’s surprising that the different teams’ initial attempts to simulate this resulted in positive results.

In addition, the Oak Ridge National Lab, an AI-powered platform, allows them to perform Hyperspectral Computed Tomography by using neutron scattering to find the optimal… maybe we should let them tell us about it.

In the medical field, there’s a new method of using image-based machine learning in Neurology. Researchers from University College London have trained an algorithm to detect early indicators of brain lesions triggered by epilepsy.

MRIs of brains used to train the UCL algorithm.

MRIs of brains are used to train the UCL algorithm.

A common cause of epilepsy resistance to drugs is what is known as focal cortex dysplasia. This specific brain region formed abnormally, but for no reason; it doesn’t appear abnormal on MRI. Being able to detect it early is very beneficial. The UCL team developed an MRI inspection model, Multicentre Epilepsy Lesion Detection, on hundreds of healthy and affected FCD brain regions.

This model has identified two-thirds of FCDs that it detected, which is excellent considering the symptoms are not obvious. It identified 178 cases in which doctors could not determine an FCD. However, it could. Naturally, the final decision is left to the specialist; however, a computer indicating something is not suitable could be enough to get a closer look and make an accurate diagnosis.

“We focused on developing the A.I. algorithm that could be understood and could assist doctors in making choices. The ability to show doctors what the MELD algorithm came up with its predictions was an important element of the process,” claimed UCL’s Mathilde Ripart.

More articles

LEAVE A REPLY

Please enter your comment!
Please enter your name here

Latest article