Welcome to Nural's newsletter where you will find a compilation of articles, news and cool companies, all focusing on how AI is being used to tackle global grand challenges.

Our aim is to make sure that you are always up to date with the most important developments in this fast-moving field.

Packed inside we have

  • NVIDIA achieves success with federated learning
  • Google finds a new approach to helping radiologists
  • and Australian states plan to use facial recognition to enforce pandemic rules

If you would like to support our continued work from £1 then click here!

Graham Lane & Marcel Hedman


Key Recent Developments


Medical AI needs federated learning, so will every industry

Medical AI Needs Federated Learning, So Will Every Industry | NVIDIA Blog
Federated learning builds powerful AI models that generalize across healthcare institutions, and shows promise for further applications in energy, financial services, manufacturing and beyond.

What: Insufficient training data is a common problem in medical AI. For example, a model built in one hospital may not work well in a different hospital. NVIDIA was involved in a successful “federated learning” project predicting the oxygen needs of patients with Covid. Twenty hospitals across five continents developed their own models. These models were then aggregated into a super-model that works better than any individual model and generalizes better when applied to new hospitals and patient groups.

Key Takeaways: Firstly, hospitals maintain control of their own data thereby avoiding many patient confidentiality issues. Only the abstract computer models are shared. Secondly, the research suggests that the federated learning approach could be deployed in a wide range of areas beyond medical AI.

Paper: Federated learning for predicting clinical outcomes in patients with COVID-19.

Paper: The future of digital health with federated learning.


Google’s new deep learning system can give a boost to radiologists

Google’s new deep learning system can give a boost to radiologists
Google’s new AI system can help detect abnormal chest x-rays from normal ones.

What: Most AI systems that analyse medical images are trained to look for very specific conditions. Google researchers, by contrast,  have developed a deep learning system that detects whether a chest x-ray appears normal or abnormal.

Key Takeaway: Recent research indicates that AI-based systems do not greatly assist radiologists because they do not fit easily into well-established, robust clinical procedures. Radiologists seldom start a patient examination by looking for a very specific illness. The intuition of Google’s researchers was that abnormality detection can have a great impact on the work of radiologists, even if the trained model doesn't point out specific diseases.

Paper: Deep learning for distinguishing normal versus abnormal chest radiographs and generalization to two unseen diseases tuberculosis and COVID-19.


Australia's two largest states trial facial recognition software to police pandemic rules

Australia’s two largest states trial facial recognition software to police pandemic rules
Australia’s two most populous states are trialling facial recognition software that lets police check people are home during COVID-19 quarantine, expanding trials that have sparked controversy to the vast majority of the country’s population.

What: The software will police people who need to quarantine at home, for example after returning from abroad. At random times, the person in quarantine is required to take a selfie in their designated quarantine location. AI checks that the selfie is bona fide.

Key Takeaway: Without this system, police would need to visit in person to monitor compliance. In this sense the app is no more intrusive than the in-person alternative. Nonetheless, there has been a strong privacy backlash, in part fueled by lack of transparency of the states involved. In the absence of robust regulation such cases give rise to well-grounded concerns that this could be the start of a slippery slope towards mass surveillance.


AI Ethics

🚀 How do we use artificial intelligence ethically?

Succinct overview about how to apply AI ethically within an organization.

🚀 IEEE launches new standard to address ethical concerns during systems design

A new standard and methodology integrating human and social values into traditional systems engineering and design.

🚀 Urgent action needed over artificial intelligence risks to human rights

Following the Pegasus spyware scandal, UN Human Rights chief calls for a moratorium on the sale and use of certain AI systems.

Other interesting reads

🚀 GSK teams with King’s College to use AI to fight cancer

Five-year partnership will create a “digital biological twin” of a patient to test approaches and personalise treatments.

🚀 The UK AI Strategy: are we listening to the experts?

A new report highlights a fundamental disconnect between the AI Strategy roadmap and those actually building AI-based products.

🚀 Facebook aware of Instagram’s harmful effect on teenage girls, leak reveals

Facebook accused of being aware of the harmful mental health impact of Instagram on teenage girls but failing to take action.

Facebook has issued a strong rebuttal.


Cool companies found this week

Farming

Alphabet (the Google umbrella company) is developing a mobile unit about the size of a shipping container that can identify weeds, measure the ripeness of fruit and predict crop yields.

EarthSense, on the other hand, has developed a robust robot, small enough to fit in the trunk of a car, that collects plant information moving underneath the plant canopy

Healthcare

Rhino Health - specialises in healthcare AI using federated learning (for example across multiple hospitals), as described in the Recent Developments section of this newsletter.

Compliance

Genvis - an Australian startup developing "high impact software for public safety teams". The company specialises in facial recognition and is behind the controversial Covid quarantine compliance system used in the state of Western Australia (refered to in this newsletter).


In 2016 machine learning pioneer Geoffrey Hinton predicted that radiologists would be out of a job within 5 years. It didn't quite work out like that ...


AI/ML must knows

Foundation Models - any model trained on broad data at scale that can be fine-tuned to a wide range of downstream tasks. Examples include BERT and GPT-3. (See also Transfer Learning)
Few shot learning - Supervised learning using only a small dataset to master the task.
Transfer Learning - Reusing parts or all of a model designed for one task on a new task with the aim of reducing training time and improving performance.
Generative adversarial network - Generative models that create new data instances that resemble your training data. They can be used to generate fake images.
Deep Learning - Deep learning is a form of machine learning based on artificial neural networks.

Best,

Marcel Hedman
Nural Research Founder
www.nural.cc

If this has been interesting, share it with a friend who will find it equally valuable. If you are not already a subscriber, then subscribe here.

If you are enjoying this content and would like to support the work financially then you can amend your plan here from £1/month!