Multi-task learning (MTL) refers to the paradigm of learning multiple related tasks together. By contrast, single-task learning (STL) learns each individual task independently. MTL often leads to better trained models because they can leverage the commonalities among related tasks. However, because MTL algorithms will "transmit" information on different models across different tasks, MTL poses a potential security risk. Specifically, an adversary may participate in the MTL process through a participating task, thereby acquiring the model information for another task. Previously proposed privacy-preserving MTL methods protect data instances rather than models, and some of them may underperform in comparison with STL methods. In this paper, we propose a privacy-preserving MTL framework to prevent the information on each model from leaking to other models based on a perturbation of the covariance matrix of the model matrix, and we study two popular MTL approaches for instantiation, namely, MTL approaches for learning the low-rank and group-sparse patterns of the model matrix. Our methods are built upon tools for differential privacy. Privacy guarantees and utility bounds are provided. Heterogeneous privacy budgets are considered. Our algorithms can be guaranteed not to underperform comparing with STL methods. Experiments demonstrate that our algorithms outperform existing privacy-preserving MTL methods on the proposed model-protection problem.

Yes, the robots are definitely coming for the jobs of America's 3.5 million cashiers. Just ask the retail workers who've already been displaced by automated checkout machines. Robots may also be coming for radiologists, whose expertise diagnosing diseases through X-rays and MRIs is facing stiff competition from artificial intelligence. And robots are starting to do some of the work in professions as diverse as chef, office clerk and tractor-trailer operator. For most of us, though, the robot invasion will simply change the tasks we do, not destroy our jobs altogether.

In a recent study, two researchers at the University of Washington used deep learning techniques to estimate obesity levels in 6 US cities. Adyasha Maharana and Dr. Elaine Okanyene Nsoesie used a convolutional neural network to extract information from Google Maps images which they found had a close relationship with obesity levels in the area. Features extracted by the convolutional neural network. The network seems to focus on natural features such as lakes and parks.Adyasha Maharana; Elaine Okanyene Nsoesie. The research suggests that the predictive power for obesity rates came from the presence of natural features such as lakes and parks detected by the neural network. The left side shows the true obesity rates from the Behavioral Risk Factor Surveillance System.

In the China model, a nation leverages its large population and low costs to build a base of blue-collar manufacturing. It then steadily works its way up the value chain by producing better and more technology-intensive goods. In the India model, a country combines a large English-speaking population with low costs to become a hub for outsourcing of low-end, white-collar jobs in fields such as business-process outsourcing and software testing. If successful, these relatively low-skilled jobs can be slowly upgraded to more advanced white-collar industries. Both models are based on a country's cost advantages in the performance of repetitive, non-social and largely uncreative work -- whether manual labor in factories or cognitive labor in call centers.

For decades now, Sam Wineburg, a professor of education and history at Stanford, has been studying the way history is taught. His new book, Why Learn History (When It's Already on Your Phone?), is about the way historical thinking--habits of mind that emphasize the careful assessment of evidence and the presumption of uncertainty, among other things--can help us navigate the information-rich environment of the web. The book, which zips along in a chatty, essayistic mode, details the work Wineburg and his colleagues have done to see how different groups of people--students, professional historians, scientists, and fact-checkers for magazines--process information in online and analog environments. In one entertaining chapter, which we've excerpted on Slate, Wineburg dissects Howard Zinn's A People's History of the United States, showing how the beloved book, so often taken as "the real truth" by people turned on to history when they read it, privileges a compelling narrative over the interrogation of evidence. Good historical thinking is by no means a magical solution to our information woes, as demonstrated by Wineburg's reports of what trained historians do while trying to navigate the web to find information about nonhistorical topics.

We also need to recognise that there will be specific points in our lives where our priorities, and therefore interests, might change. We are all used to talking about a mid-life crisis where we impulsively make rash decisions (such as buying a new sports car). However, research by LinkedIn has confirmed that we now have quarter-life crises.

In press materials for "Gradient Descent," Nature Morte stated that the works are created "entirely by AI in collaboration with artists." Obvious even signed their work with the mathematical equation for the algorithm they used, rather than the collective's name. As much as artists and gallerists may enjoy attributing authorship to AI, and emphasize that they cannot anticipate just what an AI algorithm will produce, legally, there is no doubt as to whether it's the human artist or the AI who owns the finished work. AI is simply a tool artists use, the way a photographer uses a camera or Adobe Photoshop in the creation of their images, says Jessica Fjeld, assistant director of the Cyberlaw Clinic at Harvard Law School. "Humans are deeply involved with every aspect of the creation and training of today's AI technologies, and this will continue to be true tomorrow and for the foreseeable future," Fjeld says.

Utopians believe that once AI far surpasses human intelligence, it will provide us with near-magical tools for alleviating suffering and realizing human potential. In this vision, super-intelligent AI systems will so deeply understand the universe that they will act as omnipotent oracles, answering humanity's most vexing questions and conjuring brilliant solutions to problems such as disease and climate change. But not everyone is so optimistic. The best-known member of the dystopian camp is the technology entrepreneur Elon Musk, who has called super-intelligent AI systems "the biggest risk we face as a civilization," comparing their creation to "summoning the demon." This group warns that when humans create self-improving AI programs whose intellect dwarfs our own, we will lose the ability to understand or control them.

Welcome to this series on reinforcement learning! We'll first start out by introducing the absolute basics to build a solid ground for us to run. We'll then progress onto more advanced and sophisticated topics that integrate artificial neural networks and deep learning into reinforcement learning. We'll also be getting our hands dirty by implementing some super cool reinforcement learning projects in code! Without further ado, let's get to it!

We propose graph-dependent implicit regularisation strategies for distributed stochastic subgradient descent (Distributed SGD) for convex problems in multi-agent learning. Under the standard assumptions of convexity, Lipschitz continuity, and smoothness, we establish statistical learning rates that retain, up to logarithmic terms, centralised statistical guarantees through implicit regularisation (step size tuning and early stopping) with appropriate dependence on the graph topology. Our approach avoids the need for explicit regularisation in decentralised learning problems, such as adding constraints to the empirical risk minimisation rule. Particularly for distributed methods, the use of implicit regularisation allows the algorithm to remain simple, without projections or dual methods. To prove our results, we establish graph-independent generalisation bounds for Distributed SGD that match the centralised setting (using algorithmic stability), and we establish graph-dependent optimisation bounds that are of independent interest. We present numerical experiments to show that the qualitative nature of the upper bounds we derive can be representative of real behaviours. Keywords: Distributed machine learning, implicit regularisation, generalisation bounds, algorithmic stability, multi-agent optimisation.