Deep neural networks (DNNs) perform well on a variety of tasks despite the fact that most networks used in practice are vastly overparametrized and even capable of perfectly fitting randomly labeled data. Recent evidence suggests that developing compressible representations is key for adjusting the complexity of overparametrized networks to the task at hand. In this paper, we provide new empirical evidence that supports this hypothesis by identifying two types of units that emerge when the network's width is increased: removable units which can be dropped out of the network without significant change to the output and repeated units whose activities are highly correlated with other units. The emergence of these units implies capacity constraints as the function the network represents could be expressed by a smaller network without these units. In a series of experiments with AlexNet, ResNet and Inception networks in the CIFAR-10 and ImageNet datasets, and also using shallow networks with synthetic data, we show that DNNs consistently increase either the number of removable units, repeated units, or both at greater widths for a comprehensive set of hyperparameters. These results suggest that the mechanisms by which networks in the deep learning regime adjust their complexity operate at the unit level and highlight the need for additional research into what drives the emergence of such units.

We give a complete characterization of the sampling complexity of best Markovian arm identification in one-parameter Markovian bandit models. We derive instance specific nonasymptotic and asymptotic lower bounds which generalize those of the IID setting. We analyze the Track-and-Stop strategy, initially proposed for the IID setting, and we prove that asymptotically it is at most a factor of four apart from the lower bound. Our one-parameter Markovian bandit model is based on the notion of an exponential family of stochastic matrices for which we establish many useful properties. For the analysis of the Track-and-Stop strategy we derive a novel concentration inequality for Markov chains that may be of interest in its own right.

The field of machine learning has focused, primarily, on discretized sub-problems (i.e. vision, speech, natural language) of intelligence. While neuroscience tends to be observation heavy, providing few guiding theories. It is unlikely that artificial intelligence will emerge through only one of these disciplines. Instead, it is likely to be some amalgamation of their algorithmic and observational findings. As a result, there are a number of problems that should be addressed in order to select the beneficial aspects of both fields. In this article, we propose leading questions to guide the future of artificial intelligence research. There are clear computational principles on which the brain operates. The problem is finding these computational needles in a haystack of biological complexity. Biology has clear constraints but by not using it as a guide we are constraining ourselves.

Plenty of force was with "Star Wars: The Rise of Skywalker" on its first night in theaters, but it fell short of the two previous films in the trilogy. "The Rise of Skywalker" brought in an estimated $40 million in North America in its Thursday night previews, the fifth biggest Thursday opening ever. The first film, "Star Wars: The Force Awakens," earned $57 million its Thursday night previews in 2015, at the time the biggest of all time. It was topped by the $60 million take of "Avengers: Endgame" earlier this year. The second film in the trilogy, "Star Wars: The Last Jedi," had a $45 million tally on its opening Thursday night in 2017.

President Trump and Defense Secretary Mark Esper visit the Army-Navy locker rooms to deliver words of encouragement before the 120th Army-Navy football game in Philadelphia. A probe into hand gestures flashed by West Point cadets and Naval Academy midshipmen at last weekend's televised Army-Navy college football in Philidelphia game were not racist, separate military investigations conducted by the military academies found. Clips of the "OK" hand gestures by the service-academy students during a Dec. 14, ESPN College GameDay broadcast game went viral and raised concerns over whether the signs were associated with white nationalism. The gesture, which features the thumb and forefinger that touch in a circle with the other fingers outstretched, has been appropriated as a signal for white supremacy in recent years. The Naval Academy found that two of its midshipmen were participating in a "sophomoric game" and had no racist intent behind the hand signs.

The history of Convolutional neural networks have a remote origin. It is actually in 1979, when Professor Kunihiko Fukushima proposed a hierarchical, multilayered artificial neural network called The neocognitron. The neocognitron has been used for solving the problem of handwritten character recognition and some other pattern recognition tasks, and served as the inspiration for convolutional neural networks. But, if you asked about the history of the neocognitron, we simply can tell you that it was inspired by the model proposed by Hubel & Wiesel in 1959. They found two types of cells in the visual primary cortex called simple cells and complex cells, and also proposed a cascading model of these two types of cells for use in pattern recognition tasks.

Audio Player failed to load. Try to Download directly (2.19 MB) Space to play or pause, M to mute, left and right arrows to seek, up and down arrows for volume. Australia's Human Rights Commission is calling for a moratorium on the introduction of some new artificial intelligence technologies, until the rights of humans can be safeguarded. And many of those inside the industry agree that the technology is taking off too fast for our legal system to keep up. The commission wants to better regulate artificial intelligence like facial recognition to protect people's privacy and to prevent society's most vulnerable from being further disadvantaged.

A U.S. government study released this week found that 189 facial recognition algorithms from 99 developers "falsely identified African-American and Asian faces 10 to 100 times more often than Caucasian faces." This should be the last such study. We are long overdue for federal governments to regulate or outright ban facial recognition. This year, the NYPD ran a picture of actor Woody Harrelson through a facial recognition system because officers thought the suspect seen in drug store camera footage resembled the actor. This year China used facial recognition to track its Uighur Muslim population.

KPMG International has been ranked #3 overall among the world's leading providers of Enterprise Artificial Intelligence (AI) services - and #1 for its technology innovation capabilities - in a new HFS Research report. The HFS Enterprise AI Services Top 10 Report (PDF 1.3 MB) examines the impact that leading global services providers are having in today's complex and rapidly expanding AI landscape. Noting AI's growing significance as "a key change agent" in how enterprises operate - and calling AI's potential to drive revenue growth "immense" - HFS Research assessed and ranked 21 firms based on Innovation Capability, Ability to Execute and Voice of the Customer criteria. The report praises KPMG's "industry-leading approach to consolidating IA assets and creating an ecosystem for AI talent." The report also calls out KPMG for demonstrating both an "integrated view of intelligent automation technologies" and "a globally consistent approach" under KPMG's Lighthouse Centre of Excellence and the KPMG Ignite AI platform.

Ian Wilson, former Head of AI at HSBC, presently runs an independent enterprise AI strategy advisory firm. Lee Smallwood is COO of Markets and Securities Services (North America) at Citi. Dr. Nishant Chandra is Senior Director of Data Products at VISA, and previously served as Data Science Leader at AIG. Thank you - your brief is being sent to you by email.