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
- Microsoft AI surpasses human performance in commonsense Q&A
- AI predicts who will develop dementia in two years
- and AI creates better lightning forecasts
If you would like to support our continued work from £1 then click here!
Graham Lane & Marcel Hedman
Key Recent Developments
Microsoft AI surpasses human performance on Commonsense Q&A benchmark
Yichong XuWhat: A new Microsoft Azure AI model achieved a “groundbreaking milestone”, performing better than humans at a set of common sense reasoning tasks in English. It also outperformed other AI solutions across a set of 15 other languages. The model achieves this by integrating with external sources such as a knowledge base, a dictionary and large Q&A datasets when formulating answers.
Key Takeaways: AI will need a degree of “common sense” if it is to address grand challenges. Monolithic large language models can generate human-like text but their ability to answer questions correctly may actually decrease as they grow in size. The Microsoft solution works by integrating with external sources and is similar to research by OpenAI that supports autonomous web searching to improve answers.
AI accurately predicts who will develop dementia in two years
What: A large-scale study using machine learning was able to predict with 92 percent accuracy which people who attended memory clinics would develop dementia within two years. The model was trained using patient information routinely available in clinic, such as memory and brain function, performance on cognitive tests and specific lifestyle factors. Unexpectedly, the researchers also discovered that about 8% of the diagnoses in the training dataset were apparently wrong. The model subsequently identified about 80% of these inconsistent diagnoses.
Key Takeaways: Practical difficulties integrating AI models into day-to-day clinical practice can often be a key stumbling block to widespread adoption. However, it appears that this model could be embedded in memory clinics fairly seamlessly, and the researchers are now investigating practical use.
Paper: Performance of machine learning algorithms for predicting progression to dementia in memory clinic patients
AI can create better lightning forecasts
What: A new technique combines weather forecasts with a machine learning model based on analyses of past lightning events. This hybrid method can forecast lightning over geographic regions that are prone to lightning strikes at least two days earlier than the leading existing technique.
Key Takeaways: The researchers indicate that this development is possible because of the amount and quality of lightning strike data that is now available to train the AI model. A lightning strike database was set up in 2008 and this is supplemented with precise location data from commercial instruments and satellite monitoring. The researchers indicate that better lightning forecasts could help to prepare for potential wildfires, improve safety warnings for lightning and create more accurate long-range climate models.
Paper: A dynamical forecast-machine learning hybrid system for lightning prediction
AI Ethics
🚀 Chinese scientists develop AI ‘prosecutor’ that can press its own charges
"Machine is so far able to identify eight common crimes and assess how dangerous a suspect is to the public"
🚀 ‘Invisible’, often unhappy workforce that’s deciding the future of AI
Correctly labelled data is essential for AI, yet two reports find it is often undertaken by unhappy workers using suspect methodologies.
🚀 France latest to slap Clearview AI with order to delete data
France joins the UK in demanding controversial facial recognition company Clearview AI should delete user data due to breach of GDPR
Other interesting reads
🚀 Ring in the new
Andrew Ng provides an optimistic introduction to 2022, highlighting individuals who are seeking “to bring AI’s benefits to more people while ameliorating flaws that can lead to bad outcomes”.
🚀 Facebook AI’s FLAVA foundational model tackles vision, language, and vision & language tasks all at once
AI vision and language models handle the two elements separately but FLAVA is a single model for vision, language and combined tasks
🚀 California halts Pony.ai's driverless testing permit after accident
and
Toronto suspends self-driving bus pilot after disastrous Whitby crash
Two self-driving pilots suspended due to crashes
🚀 Microsoft introduces a few-shot learning molecular dataset to assist deep learning in early-stage drug discovery
The dataset addresses the problem of molecule-protein interaction prediction given a small amount of data.
Jobs
Data scientist - AxionRay
Axion are looking to hire a talented NLP DS lead as they enter hypergrowth. Axion is a stealth AI decision intelligence platform start-up working with electric vehicle engineering leaders to accelerate development, funded by top VCs.
Comp: $100k – $180k, meaningful equity!
If interested contact: marcel.hedman@axionray.com
Cool companies found this week
Computer Vision
Roboflow - the company provides an end-to-end vision platform and raised $20 million in round A funding to "continue to democratise computer vision".
Climate
Wonder Dynamics - aims to make “blockbuster-level” visual effects achievable by living-room-level creators using AI and cloud services and has raised $10 million in round A funding.
NLP as a Service
NLP Cloud - offers API access to open source natural language models at a competitive price. Unusually, the small company claims to be making a profit with a diverse customer base but has not sought venture capital funding.
And Finally ...
An AI algorithm has made more than $1 million selling six artworks as NFTs (Non-Fungible Tokens). Here is "Cauterize Commiserate" ...
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!