Recent work suggests that convolutional neural networks of different architectures learn to classify images in the same order. To understand this phenomenon, we revisit the over-parametrized deep linear network model. Our analysis reveals that, when the hidden layers are wide enough, the convergence rate of this model's parameters is exponentially faster along the directions of the larger principal components of the data, at a rate governed by the corresponding singular values. We term this convergence pattern the Principal Components bias (PC-bias). Empirically, we show how the PC-bias streamlines the order of learning of both linear and non-linear networks, more prominently at earlier stages of learning. We then compare our results to the simplicity bias, showing that both biases can be seen independently, and affect the order of learning in different ways. Finally, we discuss how the PC-bias may explain some benefits of early stopping and its connection to PCA, and why deep networks converge more slowly with random labels.

Britons' trust in science is at an all-time high after the Covid pandemic, a new report reveals – but misinformation on social media continues to present a'threat to scientific credibility'. The 3M State of Science Index, published on Tuesday, reveals that 90 per cent of UK residents trust science in 2022, compared with 85 per cent in 2019. This stat also compares with 88 per cent of Europeans and 89 per cent of people globally who trust science in 2022. In the UK, 57 per cent of Brits say they are now more appreciative of science after the pandemic, likely due to the efforts of scientists in creating Covid vaccines. However, misinformation'is widespread' on social media and threatens the future of the public's understanding of science, the report says.

At Apple we use machine learning to teach our products to understand the world more as humans do. Of course, understanding the world better means building great assistive experiences. Machine learning can help our products be intelligent and intuitive enough to improve the day-to-day experiences of people living with disabilities. We can build machine-learned features that support a wide range of users including those who are blind or have low vision, those who are deaf or are hard of hearing, those with physical motor limitations, and also support those with cognitive disabilities. Mobile devices and their apps have become ubiquitous.

Australia will need to double the pace of its renewable energy uptake to meet the 2030 emissions target set by the Albanese government, even without any increase in demand, according to Bruce Mountain, head of the Victoria Energy Policy Centre. Labor's main energy policy, Rewiring the Nation, includes the creation of a special corporation to funnel $20bn into new transmission links to accelerate the uptake of more clean energy. The plan is part of Labor's pledge to cut Australia's 2005-level greenhouse gas emissions 43% by 2030, projecting renewables reach an 82% share of renewables in the National Electricity Market by then. Excluding hydro power, renewable energy has increased its share of the market 3% annually in the past five years, Mountain says. "Deducting 10% from hydro, the target is 72%," he says of Labor's goal.

For the past three decades, geologist Carlos Souza has worked at the Brazil-based nonprofit Imazon, exploring ways he and the teams he coordinates can use applied science to protect the Amazon rainforest. For much of that time, satellite imagery has been a big part of his job. In the early 2000s, Souza and colleagues came to understand that 90 percent of deforestation occurs within 5 kilometers of newly created roads. While satellites have long been able to track road expansion, the old way of doing things required people to label those findings by hand, amassing what would eventually become training data. Those years of labor paid off last fall with the release of an AI system that Imazon says reveals 13 times more roadway than the previous method, with an accuracy rate of between 70 and 90 percent.

Artificial Intelligence (AI) is much more than just a buzzword nowadays. It powers facial recognition in smartphones and computers, translation between foreign languages, systems which filter spam emails and identify toxic content on social media, and can even detect cancerous tumours. These examples, along with countless other existing and emerging applications of AI, help make people's daily lives easier, especially in the developed world. As of October 2021, 44 countries were reported to have their own national AI strategic plans, showing their willingness to forge ahead in the global AI race. These include emerging economies like China and India, which are leading the way in building national AI plans within the developing world. Oxford Insights, a consultancy firm that advises organisations and governments on matters relating to digital transformation, has ranked the preparedness of 160 countries across the world when it comes to using AI in public services.

One of Europe's largest business software providers, with 14,000 employees, 1,135,000 private and public sector customers across the Nordics, Benelux, Central and Eastern Europe and Latin America, Visma, has adopted Luminance's AI-powered legal process automation tool as part of its M&A due diligence reviews, as the Oslo-headquartered company looks to continue its rapid growth trajectory that has seen a net revenue of €1.74 billion in 2020. Norway's Visma made a record 42 acquisitions in 2021. This increased M&A activity means the company's in-house legal department must review large volumes of data when assessing acquisition opportunities and analysing areas of risk and opportunity. Hjalmar Florijn, Head of Legal at Visma Benelux, commented that the rapid insight from Luminance will also help them to keep more M&A work in-house and reduce reliance on external counsel: "As Visma continues to experience rapid growth, it is critical that we can instantly understand the contractual landscape of an acquisition target. With Luminance, our in-house legal teams will be able to uncover critical findings earlier in the review process, giving us confidence that we have been appraised of all possible risks and allowing us to plan our business strategy accordingly," said Florijn.

Atmospheric Extreme Events (EEs) cause severe damages to human societies and ecosystems. The frequency and intensity of EEs and other associated events are increasing in the current climate change and global warming risk. The accurate prediction, characterization, and attribution of atmospheric EEs is therefore a key research field, in which many groups are currently working by applying different methodologies and computational tools. Machine Learning (ML) methods have arisen in the last years as powerful techniques to tackle many of the problems related to atmospheric EEs. This paper reviews the ML algorithms applied to the analysis, characterization, prediction, and attribution of the most important atmospheric EEs. A summary of the most used ML techniques in this area, and a comprehensive critical review of literature related to ML in EEs, are provided. A number of examples is discussed and perspectives and outlooks on the field are drawn.

Monitoring whales in remote areas is important for their conservation; however, using traditional survey platforms (boat and plane) in such regions is logistically difficult. The use of very high-resolution satellite imagery to survey whales, particularly in remote locations, is gaining interest and momentum. However, the development of this emerging technology relies on accurate automated systems to detect whales, which are currently lacking. Such detection systems require access to an open source library containing examples of whales annotated in satellite images to train and test automatic detection systems. Here we present a dataset of 633 annotated whale objects, created by surveying 6,300 km2 of satellite imagery captured by various very high-resolution satellites (i.e. WorldView-3, WorldView-2, GeoEye-1 and Quickbird-2) in various regions across the globe (e.g. Argentina, New Zealand, South Africa, United States, Mexico). The dataset covers four different species: southern right whale (Eubalaena glacialis), humpback whale (Megaptera novaeangliae), fin whale (Balaenoptera physalus), and grey whale (Eschrichtius robustus).

The combination of Reinforcement Learning (RL) with deep learning has led to a series of impressive feats, with many believing (deep) RL provides a path towards generally capable agents. However, the success of RL agents is often highly sensitive to design choices in the training process, which may require tedious and error-prone manual tuning. This makes it challenging to use RL for new problems and also limits its full potential. In many other areas of machine learning, AutoML has shown that it is possible to automate such design choices, and AutoML has also yielded promising initial results when applied to RL. However, Automated Reinforcement Learning (AutoRL) involves not only standard applications of AutoML but also includes additional challenges unique to RL, that naturally produce a different set of methods. As such, AutoRL has been emerging as an important area of research in RL, providing promise in a variety of applications from RNA design to playing games, such as Go. Given the diversity of methods and environments considered in RL, much of the research has been conducted in distinct subfields, ranging from meta-learning to evolution. In this survey, we seek to unify the field of AutoRL, provide a common taxonomy, discuss each area in detail and pose open problems of interest to researchers going forward.