In this paper, we address the problem of dataset quality in the context of Machine Learning (ML)-based critical systems. We briefly analyse the applicability of some existing standards dealing with data and show that the specificities of the ML context are neither properly captured nor taken into ac-count. As a first answer to this concerning situation, we propose a dataset specification and verification process, and apply it on a signal recognition system from the railway domain. In addi-tion, we also give a list of recommendations for the collection and management of datasets. This work is one step towards the dataset engineering process that will be required for ML to be used on safety critical systems.

When Apple unveiled the HomePod mini last month, the company showed off a long-awaited feature: the ability to use it with third-party music services. And as of now, Pandora is the first third-party music app to work with Apple's existing HomePod as well as the upcoming HomePod mini. An update to Pandora's iOS app is rolling out that lets you add the app the HomePod. If you've updated the Pandora app, you'll find a new item in the settings menu: Connect with HomePod. From there, it just takes a few taps to give Pandora permission to work with the speaker. This means you can ask Siri to do things like "play New Indie Radio on Pandora" and it'll start playing your station directly on the speaker.

Light strips are a fun and easy way to add smart lighting to just about anywhere in your home, like under kitchen cabinets, around TVs, along baseboards, and more. To get the job done right, you need a light strip that's fast and simple to set up, stays securely in place, and comes loaded with fun and useful features to light up any room. For these reasons, the Lifx Z LED Strip 6.6' Kit is the best smart light strip you can buy. In what felt like a blink of an eye, we had these dimmable lights connected to Alexa, Google Assistant, and Siri. They also work with IFTTT, SmartThings, Nest, Arlo, Flic, and more.

Jabra has been surprising us with category-busting earbuds since the brand's first "so-called" true wireless pair (that is, those without wires at all), the Elite and Elite Active 65t. Elite they were, and the follow-up Elite and Elite Active 75t are even better. In fact, Jabra's decision to retroactively add noise canceling to the already stellar 75t series via firmware makes them excellent competition for the new 85t, at a much lower price point. However, while the 85t may look a lot like the 75t--with some extra bulk--Jabra's engineers have drawn up something new here, aimed at long-term wearability and improved sound quality to take on the AirPods Pro, Bose's QuietComfort Earbuds, and other premium-tier rivals. And apart from one notable flaw (for which a fix is reportedly coming) they're about as close to perfect as true wireless gets.

When we covered Leena AI as a member of the Y Combinator Summer 2018 cohort, the young startup was firmly focused on building HR chatbots, but in the intervening years it has expanded the vision to a broader HR policy platform. Today, the company announced an $8 million Series A led by Greycroft with help from several individual industry investors. Company CEO and co-founder Adit Jain says that in 2018 the company was concentrating on building an intelligent virtual assistant for HR-related questions. It allowed employees to ask the bot questions like how many vacation days they have left or what holidays they have off this year. Over the last couple of years since leaving Y Combinator, the company has moved into broader HR service delivery.

Researchers are studying whether artificial-intelligence tools that analyze things like typing speed, sleep patterns and speech can be used to help clinicians better identify patients with early-stage dementia. Huge quantities of data reflecting our ability to think and process information are now widely available, thanks to watches and phones that track movement and heart rate, as well as tablets, computers and virtual assistants such as Amazon Echo that can record the way we type, search the internet and pay bills.

The time has come for Artificial Intelligence (AI)-driven teaching assistants to help ease a human teacher's workload in the age of online learning, however, such virtual machines have to be effective and communicate well to be accepted by the society in a broad way, argue researchers. The increase in online education has allowed a new type of teacher to emerge -- an artificial one. But just how accepting students are of an artificial instructor remains to be seen, said researchers at the University of Central Florida's Nicholson School of Communication and Media who are working to examine student perceptions of AI-based teachers. Some of their findings, published in the'International Journal of Human-Computer Interaction', indicated that for students to accept an AI teaching assistant, it needs to be effective and easy to talk to. "The hope is that by understanding how students relate to AI-teachers, engineers and computer scientists can design them to easily integrate into the education experience," said Jihyun Kim, an associate professor in the school and lead author of the study.

How to Do It is Slate's sex advice column. Send it to Stoya and Rich here. Every week, the crew responds to a bonus question in chat form. I recently met a guy on Tinder, where I usually don't have much luck because I'm not conventionally attractive and want to date, not just hook up. But after talking to this guy for a few days we seem practically perfect for each other!

Searching for and making decisions about products is becoming increasingly easier in the e-commerce space, thanks to the evolution of recommender systems. Personalization and recommender systems have gone hand-in-hand to help customers fulfill their shopping needs and improve their experiences in the process. With the growing adoption of conversational platforms for shopping, it has become important to build personalized models at scale to handle the large influx of data and perform inference in real-time. In this work, we present an end-to-end machine learning system for personalized conversational voice commerce. We include components for implicit feedback to the model, model training, evaluation on update, and a real-time inference engine. Our system personalizes voice shopping for Walmart Grocery customers and is currently available via Google Assistant, Siri and Google Home devices.

Recommendation techniques are important approaches for alleviating information overload. Being often trained on implicit user feedback, many recommenders suffer from the sparsity challenge due to the lack of explicitly negative samples. The GAN-style recommenders (i.e., IRGAN) addresses the challenge by learning a generator and a discriminator adversarially, such that the generator produces increasingly difficult samples for the discriminator to accelerate optimizing the discrimination objective. However, producing samples from the generator is very time-consuming, and our empirical study shows that the discriminator performs poor in top-k item recommendation. To this end, a theoretical analysis is made for the GAN-style algorithms, showing that the generator of limit capacity is diverged from the optimal generator. This may interpret the limitation of discriminator's performance. Based on these findings, we propose a Sampling-Decomposable Generative Adversarial Recommender (SD-GAR). In the framework, the divergence between some generator and the optimum is compensated by self-normalized importance sampling; the efficiency of sample generation is improved with a sampling-decomposable generator, such that each sample can be generated in O(1) with the Vose-Alias method. Interestingly, due to decomposability of sampling, the generator can be optimized with the closed-form solutions in an alternating manner, being different from policy gradient in the GAN-style algorithms. We extensively evaluate the proposed algorithm with five real-world recommendation datasets. The results show that SD-GAR outperforms IRGAN by 12.4% and the SOTA recommender by 10% on average. Moreover, discriminator training can be 20x faster on the dataset with more than 120K items.