By Jason Dorrier 

The skyrocketing scale of AI has been hard to miss in recent years. The most advanced algorithms now have hundreds of billions of connections, and it takes millions of dollars and a supercomputer to train them. But as eye-catching as big AI is, progress isn’t all about scale—work on the opposite end of the spectrum is just as crucial to the future of the field.

Some researchers are trying to make building AI faster, more efficient, and more accessible, and one area ripe for improvement is the learning process itself. Because AI models and the data sets they feed on have grown exponentially, advanced models can take days or weeks to train, even on supercomputers.

Might there be a better way? Perhaps.

DoraSorter will be used by FedEx to increase operational efficiency and construct agile logistic infrastructure. 

By Dipayan Mitra

Global transportation and eCommerce service providing company FedEx launches its new AI-powered sorting robot. FedEx collaborated with artificial intelligence-enabled robotic firm Dorabot to develop the new sorting robot named DoraSorter. 

This new development is FedEx’s step towards modernizing and automating the logistic network. In recent years, a considerable boom has been witnessed in the eCommerce industries leading to a vast number of shipments worldwide. 

DoraSorter will aid FedEx in meeting the demand of an ever-increasing number of shipments quickly, minimizing the need for human intervention in the sorting process involved in eCommerce transportation. 

According to FedEx, the AI-powered sporting robot will be initially deployed at the 5,200m2 FedEx South China E-Commerce Shipment Sorting Center in Guangzhou. The robot is already capable of handling multiple tasks, including managing small quantities of inbound and outbound shipments from customers. However, both the companies are still working to further increase and fine-tune the capabilities of DoraSorter. 

Though it almost certainly won’t work, it is a telling sign of where the field is heading

By Tate Ryan-Mosley

A police officer is at the scene of a murder. No witnesses. No camera footage. No obvious suspects or motives. Just a bit of hair on the sleeve of the victim’s jacket. DNA from the cells of one strand is copied and compared against a database. No match comes back, and the case goes cold. 

Corsight AI, a facial recognition subsidiary of the Israeli AI company Cortica, purports to be devising a solution for that sort of situation by using DNA to create a model of a face that can then be run through a facial recognition system. It is a task that experts in the field regard as scientifically untenable. 

Corsight unveiled its “DNA to Face” product in a presentation by chief executive officer Robert Watts and executive vice president Ofer Ronen intended to court financiers at the Imperial Capital Investors Conference in New York City on December 15. It was part of the company’s overall product road map, which also included movement and voice recognition. The tool “constructs a physical profile by analyzing genetic material collected in a DNA sample,” according to a company slide deck viewed by surveillance research group IPVM and shared with MIT Technology Review. 

The new technology can capture facial expressions from 3D images with an accuracy of over 80%. 

ByDipayan Mitra

Researchers from Korea’s one of the best science and technology universities, KAIST’s Department of Bio and Brain Engineering, have developed a new artificial intelligence-powered light field camera that can read 3D facial expressions. 

The highly capable camera uses a technique that uses infrared light to read facial expressions. Professors Ki-Hun Jeong and Doheon Lee led the research team which developed this artificial intelligence-enabled technology. 

The research named ‘Machine-Learned Light-field Camera that Reads Facial Expression from High-Contrast and Illumination Invariant 3D Facial Images’ was published in Advanced Intelligent Systems. 

Prof Ki-Hun Jeong said, “The sub-miniature light-field camera developed by the research team has the potential to become the new platform to quantitatively analyze the facial expressions and emotions of humans.” 

Using NVIDIA Clara, Google DeepVariant, and Oxford Nanopore Technologies sequencing, the Stanford University School of Medicine has been able to identify genetic diseases in as little as 7.5 hours. In just hours, a team of researchers from Stanford University was able to find a pathogenic variant and diagnose the rare seizure-causing genetic disorder.

This ultra-rapid sequencing research was published in New England Journal of Medicine. The proposed method set a new world record for fastest DNA sequencing technique. 

The research team accelerated both base calling and variant calling using NVIDIA GPUs on Google Cloud. They also sped up variant calling by turning it into an application with Clara Parabricks, a computational genomics application framework.

The team was able to optimize every step in the pipeline, including speeding up sample preparation and using nanopore sequencing on Oxford Nanopore’s PromethION Flow Cell. This allowed them generate more than 100 gigabytes worth of data per hour!

by Michael Wolf

This week restaurant chain Checkers & Rally’s announced a deal with Presto, a maker of restaurant technology, in which the drive-thru focused chain will roll out Presto’s AI-powered voice assistant technology chain-wide by the end of the year.

The announcement comes after early trials showed a 98% order accuracy for the voice assistant technology. And now, according to Checkers President and CEO Frances Allen, the chain plans to roll out Presto’s voice technology to all 267 store-owned and operated locations by the end of 2022. She also indicated that many franchise locations are expressing interest in the technology.

“We had a vision two years ago that we wanted what I would call ‘Alexa at the drive-thru,’” said Allen. 

And now, with Presto, they have it. According to Allen, about 80% of the chain’s total business comes through a drive-thru, so the company’s management felt transitioning to a high-accuracy automated drive-thru could significantly impact the business. “Anything we can do to improve operations,streamline for our guests and our employees, we wanted to do.”

Artificial intelligence drug design company Iktos, and South Korean clinical research biotech Astrogen announced today a collaboration with the goal of discovering small-molecule pre-clinical drug candidates for a specific, undisclosed, marker of Parkinson’s disease (PD).

Under the terms of the agreement, whose value was not disclosed, Iktos will apply its generative learning algorithms which seek to identify new molecular structures with the potential address the target in PD. Astrogen, which has a focus of the development of therapeutics for “intractable neurological diseases,” will provide in-vitro and in-vivo screening of lead compounds and pre-clinical compounds. While both companies will contribute to the identification of new small-molecule candidates, Astrgoen will lead the drug development process from the pre-clinical stages.

“Our objective is to expedite drug discovery and achieve time and cost efficiencies for our global collaborators by using Iktos’s proprietary AI platform and know-how,” noted Yann Gaston-Mathé, president and CEO of Paris-based Iktos in a press release. “We are confident that together we will be able to identify promising novel chemical matter for the treatment of intractable neurological diseases. Our strategy has always been to tackle challenging problems alongside our collaborators where we can demonstrate value generation for new and on-going drug discovery projects.”

The Second-generation AI systems focus on progress where rather than analysing data for diagnosis assistance, prediction, or tailoring therapy, the second-generation platforms help in improving the biological processes.

BY MEETA RAMNANI

Improving global health requires drugs to be more effective and affordable. While there are multiple branded and generic drugs available, partial or complete loss of response to chronic medications is a major cause that leads to ineffectiveness. Combining this with a lack of adherence by patients leads to even more healthcare issues.  

The first-generation AI systems did not address these needs, which led to a low adoption rate. But the second-generation AI systems are focused on a single subject – improving patients’ clinical outcomes. The digital pills combine a personalised second-generation AI system along with the branded or generic drug and improve the patient response as it increases adherence and overcomes the loss of response to chronic medications. It works on improving the effectiveness of drugs and therefore reducing healthcare costs and increasing end-user adoption.

There are many examples to prove that there is a partial or complete loss of response to chronic medications. Cancer drug resistance is a major obstacle for the treatment of multiple malignancies, one-third of epileptics develop resistance to anti-epileptic drugs; also, a similar percentage of patients with depression develop resistance to anti-depressants. Other than the loss of response to chronic medications, low adherence is also a common problem for many NCDs. A little less than 50% of severely asthmatic patients adhere to inhaled treatments, while 40% of hypertensive patients show non-adherence.

The second-generation systems are aimed at improving outcomes and reducing side effects. To overcome the hurdle of biases induced by big data, these systems implement an n = 1 concept in a personalised therapeutic regimen. This focus of the algorithm improves the clinically meaningful outcome for an individual subject. The personalised closed-loop system used by the second-generation system is designed to improve the end-organ function and overcome tolerance and loss of effectiveness.

Aiseed-developed drones. Credit: Aiseed

By Ambrose Huang and Kevin Cheng

The pace of global drone development has increased over the last few years with smart drones being used more commonly for consumer markets, rescue operations, visual inspections and logistics. Taiwan-based startup Aiseed has combined edge computing with self-developed artificial intelligence (AI) to develop advanced drone products with better features and higher speed. With relatively low costs, the company hopes to make advanced drones more accessible across different markets.

Chinese AI-driven satellite takes ultra-high res images of US city from orbit at speeds ‘never before seen’

China’s Beijing-3 satellite was able to take hi-res images of the San Francisco Bay area

By Jon Rogers

CHINA has developed a powerful AI satellite capable of taking hi-res images of cities in a matter of seconds.

Beijing-3, a small one-tonne commercial satellite, can take images clear enough to be able to identify a military vehicle on the street and tell what type of weapon it might be carrying, according to researchers.

The satellite was launched by China in June and performed an in-depth scan of the central San Francisco Bay area, covering 1,470 square miles, the journal Spacecraft Engineering reports.

Most satellites observing the Earth must be stable when taking image because altitude control mechanisms can produce vibrations that blur the images.

But in the Chinese experiment on June 16, the satellite was able to change the angle of its camera’s line of sight to the ground when passing over the US.

The movement meant it could capture a larger area than satellites have been previously able to.

The pictures were taken at an altitude of 310 miles and had a resolution of 50 centimetres per pixel with the test showing the satellite could take images while its body was twisting at up to 10 degrees per second, a speed not seen on a satellite before.

By Edd Gent – Dec 20, 202131,578

Bigger is better—or at least that’s been the attitude of those designing AI language models in recent years. But now DeepMind is questioning thisrationale, and says giving an AI a memory can help it compete with models 25 times its size.

When OpenAI released its GPT-3 model last June, it rewrote the rulebook for language AIs. The lab’s researchers showed that simply scaling up the size of a neural network and the data it was trained on could significantly boost performance on a wide variety of language tasks.

Since then, a host of other tech companies have jumped on the bandwagon, developing their own large language models and achieving similar boosts in performance. But despite the successes, concerns have been raised about the approach, most notably by former Google researcher Timnit Gebru.

In the paper that led to her being forced out of the company, Gebru and colleagues highlighted that the sheer size of these models and their datasets makes them even more inscrutable than your average neural network, which are already known for being black boxes. This is likely to make detecting and mitigating bias in these models even harder.