Radical Reads

Machine learning and Covid-19: Oxford scientists develop extremely rapid diagnostic test for Covid-19   (University of Oxford)

“Scientists from Oxford University’s Department of Physics have developed an extremely rapid diagnostic test that detects and identifies viruses in less than five minutes…

Working directly on throat swabs from COVID-19 patients, without the need for genome extraction, purification or amplification of the viruses, the method starts with the rapid labelling of virus particles in the sample with short fluorescent DNA strands. A microscope is then used to collect images of the sample, with each image containing hundreds of fluorescently-labelled viruses.

Machine-learning software quickly and automatically identifies the virus present in the sample. This approach exploits the fact that distinct virus types have differences in their fluorescence labeling due to differences in their surface chemistry, size, and shape.”

Radical Commentary: AI may hold the key to mass, real-time testing for COVID-19. Oxford University scientists have been working to label virus particles in throat swabs from COVID-19 patients. Comprehensive labeled data enables machine learning to quickly identify the presence of COVID-19 in test swabs. Equipped with rapid testing capabilities, public health officials would have a proactive means to mitigate outbreaks in places like schools, hospitals, airports, restaurants, workplaces, etc. If this particular solution works, it will not be widely available until at least mid-2021.

AI and manufacturing: Exporters turn to automation to boost productivity and profits  (Globe and Mail)

“The pandemic has made many things painfully obvious to Canadian exporters – including the fact that investing in automation and technology is the future of exporting.

Firms that invested in automation are now reaping the benefits while those that didn’t are playing catch up, says Dennis Darby, president and chief executive officer of Canadian Manufacturers and Exporters (CME).”

Radical Commentary: Last year, CME published a report warning about the lack of investment by Canadian manufacturers in technology and innovation. Now with COVID-19 resulting in physical distancing and worker absenteeism, Canadian manufacturers are rethinking operations and increasingly looking to automation to supplement their workforces and reinforce their supply chains.

But technology adoption takes time. While repetitive and predictable tasks like the movement of goods can be successfully automated, other repetitive but more complex and unpredictable tasks that require human hands, such as picking, placing and unloading, have not yet been widely reinforced by automation. Automating this work requires advanced AI that can understand, learn and adapt.

One of the few companies addressing this enormous challenge is Covariant, a Radical portfolio company. Covariant is building a universal AI — the “Covariant Brain” — that enables robots to see, reason and act autonomously in the real world. Robots powered by Covariant learn general abilities such as robust 3D perception, physical affordances of objects, few-shot learning and real-time motion planning which gives them the intelligence to learn how to manipulate new objects they have never seen before in environments where they have never operated. This kind of AI technology is reshaping the future of supply chain management and offers a glimpse of what is possible for manufacturers looking to intelligently reinforce operations on the factory and warehouse floor.

AI research: A radical new technique that lets AI learn with practically no data  (MIT Technology Review)

“Machine learning typically requires tons of examples. To get an AI model to recognize a horse, you need to show it thousands of images of horses. This is what makes the technology computationally expensive—and very different from human learning…

Now a new paper from the University of Waterloo in Ontario suggests that AI models should also be able to do this—a process the researchers call “less than one”-shot, or LO-shot, learning. In other words, an AI model should be able to accurately recognize more objects than the number of examples it was trained on.” 

Radical Commentary: Think about how quickly we learn to identify animals. After seeing one image of a cow in a picture book, a child might spot a real life cow in a farmer’s field. Humans often need just a brief exposure to an object — sometimes we can just read about it — in order to identify it. That capacity to rapidly learn on limited data is one of the things that distinguishes humans from AI systems.

Research breakthroughs on few-shot learning could make a big difference for AI research and applications which depend on large datasets and the expensive cloud compute infrastructures required to train AI. The less data needed to train AI, the more accessible the technology will be for companies and industries, and the lower the costs of computation. There may also be privacy benefits as less information must be collected to create useful models. While this research is still progressing, advances in this area could have a profound effect on the development and broader adoption of AI across many industries and businesses which lack proprietary data moats and the financial means to access expensive cloud computation at scale.

AI and healthcare: Artificial Intelligence May Improve Heart Transplant Outcomes  (Health IT Analytics)

“A $3.2 million grant from NIH will support researchers in using artificial intelligence to improve outcomes for heart transplant patients.

Researchers from the Perelman School of Medicine at the University of Pennsylvania, Case Western Reserve University, Cleveland Clinic, and Cedars-Sinai Medical Center will use the four-year grant to determine the likelihood of cardiac patients accepting or rejecting a new heart.”

Radical commentary: A patient’s body rejecting the donor organ is one of the most significant risks of a heart transplant, with rejections occurring in 30 to 40 per cent of patients in the first year after a transplant. However, it is difficult for healthcare professionals to determine the probability of whether a patient will reject the donor organ or not and why.

This grant from the National Institutes of Health will provide these researchers an opportunity to use AI to analyze imaging and detect patterns of rejection. Arming healthcare professionals with this information could lead to reduced rates of infection and rejection of donor organs. AI is also being used to enhance the accuracy of breast cancer screenings and cervical cancer screenings to improve patient outcomes.

AI and Earth-observation research: Satellites could soon map every tree on Earth  (Nature)

“Convolutional networks rely on the availability of training data, which in this case consisted of satellite images in which the visible outlines of tree and shrub canopies were manually traced. Through training using these samples, the computer learnt how to identify individual tree canopies with high precision in other images. The result is a wall-to-wall mapping of all trees larger than 2 m in diameter across the whole of southern Mauritania, Senegal and southwestern Mali.”

Radical Commentary: The ability to map individual trees starts us on the long journey towards obtaining a full inventory of our planet’s biodiversity. Historically, the quality of inventory mapping varied depending on funding, political will, and accessibility, often determined by state lines. AI techniques may reduce gaps in capacity to create complete maps based on climatic regions rather than state and territorial boundaries.

Understanding biodiversity and tree canopy cover is a necessary tool for ensuring plant conservation and sustainable use. Scaling this technique to map the size and location of every tree on earth is limited by the quality of the satellite data. Individual image pixels generally correspond to areas on the ground that are larger than 100 square metres, and often larger than one square kilometre. However, during the past two decades, a variety of commercial satellites have begun to collect data at a higher spatial resolution, capable of capturing ground objects measuring one square metre or less. As this image data improves, AI will bring a fuller accounting of our planet’s true bio health.

***
Editor’s Note: We will continue to use this platform to share without commentary articles focused on data and the use of it to illustrate and illuminate racial injustice. Because you cannot fix problems you cannot see or understand.

Racial Disparities in Voting Wait Times: Evidence from Smartphone Data  (arXiv)

“Equal access to voting is a core feature of democratic government. Using data from millions of smartphone users, we quantify a racial disparity in voting wait times across a nationwide sample of polling places during the 2016 US presidential election. Relative to entirely-white neighborhoods, residents of entirely-black neighborhoods waited 29% longer to vote and were 74% more likely to spend more than 30 minutes at their polling place. This disparity holds when comparing predominantly white and black polling places within the same states and counties, and survives numerous robustness and placebo tests. Our results document large racial differences in voting wait times and demonstrates that geospatial data can be an effective tool to both measure and monitor these disparities.”