Audio textures are a subset of environmental sounds, often defined as having stable statistical characteristics within an adequately large window of time but may be unstructured locally. They include common everyday sounds such as from rain, wind, and engines. Given that these complex sounds contain patterns on multiple timescales, they are a challenge to model with traditional methods. We introduce a novel modelling approach for textures, combining recurrent neural networks trained at different levels of abstraction with a conditioning strategy that allows for user-directed synthesis. We demonstrate the model's performance on a variety of datasets, examine its performance on various metrics, and discuss some potential applications.

You've thought long and hard about it, and you're finally ready to buy a smart speaker. They're all being heavily marketed for Black Friday sales, and you'd like to ask Siri, Alexa or the Google Assistant to play music, tell you the weather, turn off your TV or lights on command. But you're stuck: Which brand and model to buy? Amazon has five Echo speakers currently available, Google has six and Apple has two. Which one is right for you? And then there are the concerns about security and snooping.

I've been experimenting with flipout layers with tensorflow and tensorflow probability, but I keep having numerical stability issues in late epochs using negative log likelihood losses with adam. It also seems like the network has a lot of difficulty with learning scale parameters, (e.g. the standard deviation if the probability model is normal) particularly when the n is smallest, and returns a lot of samples far, far outside the range of the training data.

This idea could be interpreted as being rather bleak; are we doomed to repeat the errors of the past until we correct them? We certainly do need to learn and re-learn life lessons--whether in our work, relationships, finances, health, or other areas--in order to grow as people. Zooming out, the same phenomenon exists on a much bigger scale--that of our collective human history. We like to think we're improving as a species, but haven't yet come close to doing away with the conflicts and injustices that plagued our ancestors. What might happen over the course of this year, and what information would we use to make educated guesses about it? The editorial team at The Economist took a unique approach to answering these questions.

A wind of innovation is blowing in the artificial intelligence sector. As artificial intelligence develops, its use cases diversify. Many companies are emerging and exploiting this technology in a relevant and innovative way. Artificial intelligence and machine learning are increasingly popular among companies in all industries. However, AI algorithms tend to overwork processors and GPUs.

Errol Baran, the senior vice president for business development and innovation at BBC Global News gives The Drum a peek into Project Songbird, the publisher's AI-powered synthetic voice which'reads' articles from BBC.com. Project Songbird is BBC News' new smart text-to-speech based commercial proposition for its digital properties. The Beeb hopes will allow its audiences to listen to their favorite feature articles hands-free without the need to actively click and browse. Built with cognitive and behavioral integrated tech software, the project will capitalize on consumer demand for audio content and leverage on BBC's editorial coverage. Built with the ability to download in the background, the audio product aims to offer a dynamic advertiser experience.

The obvious method is to cross-reference a claim with existing facts to verify whether that claim is true. This is accurate and highly explainable, but it also comes at some significant costs. Not only does it assume that the claim is checkable (e.g., what if a world leader decided to say "the entire universe was created last Thursday"), but it also requires a huge database of evidence because the AI needs sufficient evidence on so many different areas in order to be able to verify an acceptable proportion of claims. The alternate method is to use the context of a claim to try to make an assumption on whether or not it's true. For example, if the claim was made by The Onion, we could probably assume it's not true.

The artificial intelligence applied research startup Abzu identifies false news with its proprietary QLattice. In the latest Reuters Institute Digital News Report, less than four in ten people said that they trust most news most of the time (that's 38% surveyed in January 2020, a fall of four percentage points from 2019)¹. Today's global crises make it all too obvious the necessity for dependable and factual journalism, yet we are exposed to a continuum of information authored by innumerable sources with debatable credentials. Slaves to our most basic emotions -- fear, disgust, and surprise² -- we are inflamed by an addictive negative feedback loop of our own creation. We crave the truth, but data shows we force-feed ourselves lies.

This week we'll celebrate Thanksgiving in the US, and nostalgia will be all over TV screens and Peacock resurrects Saved by the Bell with a cast that includes new younger actors and several of the stars from the original show. Other throwbacks include Mad Max on Ultra HD Blu-ray and a Buck Rogers box set, but for something newer you can check out Peninsula, a sequel to the excellent Korean zombie movie Train to Busan as it arrives on Ultra HD Blu-ray. Netflix's latest feature film is Mosul, along with its Hillbilly Elegy movie. Criterion is also releasing a special edition of Netflix's Martin Scorsese feature The Irishman, however it's sadly only available in 1080p Blu-ray without the benefit of 4K and Dolby Vision HDR. For an all-new option, try Superintelligence on HBO Max, where Melissa McCarthy stars as a woman chosen by an all-powerful AI for surveillance via her various connected devices.

High-dimensional representations for words, text, images, knowledge graphs and other structured data are commonly used in different paradigms of machine learning and data mining. These representations have different degrees of interpretability, with efficient distributed representations coming at the cost of the loss of feature to dimension mapping. This implies that there is obfuscation in the way concepts are captured in these embedding spaces. Its effects are seen in many representations and tasks, one particularly problematic one being in language representations where the societal biases, learned from underlying data, are captured and occluded in unknown dimensions and subspaces. As a result, invalid associations (such as different races and their association with a polar notion of good versus bad) are made and propagated by the representations, leading to unfair outcomes in different tasks where they are used. This work addresses some of these problems pertaining to the transparency and interpretability of such representations. A primary focus is the detection, quantification, and mitigation of socially biased associations in language representation.