Google Voice Search Takes Another Quality Step Forward

Google Voice Search has taken another quality step forward. “Our improved acoustic models rely on Recurrent Neural Networks (RNN). RNNs have feedback loops in their topology, allowing them to model temporal dependencies: when the user speaks /u/ in the previous example, their articulatory apparatus is coming from a /j/ sound and from an /m/ sound before. Try saying it out loud – “museum” – it flows very naturally in one breath, and RNNs can capture that. The type of RNN used here is a Long Short-Term Memory (LSTM) RNN which, through memory cells and a sophisticated gating mechanism, memorizes information better than other RNNs. Adopting such models already improved the quality of our recognizer significantly.”

Mapping the Northern Lights With Twitter

Interesting: Mapping the Northern Lights with Twitter. “In the study, the use of Twitter as a measure of auroral activity is investigated for the first time. According to the researchers, studies have shown that Twitter users can provide real-time information about large-scale events and disasters such as earthquakes, influenza outbreaks and wildfires. The researchers’ study collates tweets and investigates the possibility of Twitter for both real-time analysis and mapping of an aurora, as has been done with other large-scale events such as natural disasters. “

Professor Receives Award for Indoor Mapping Research

So glad this aspect of mapping is getting more attention. Syracuse University assistant professor Dr. Yun Huang is getting an award from Google to continue research on indoor mapping. “Improving awareness about and potential uses of facilities to enhance learning is the basis for Dr. Huang’s concept for creating a system to better map indoor environments, capabilities, and resources, she said. In developing the idea, she wondered how much people know about and understand the unique details of their indoor environments, as well as being aware of all the resources that are available in their surroundings, such as the centers and buildings where they work and study on a daily basis, she explained.” Speaking from a commercial perspective, it’s also important to retailers, as you can only do so much with signage, and if you overdo it people get sign-blind.

Wikipedia Articles About Places Written by the West

Wikipedia’s articles about places written by the West, which shouldn’t surprise anybody. “Nearly half of all edits to articles about places on Wikipedia were made from just five countries, researchers at the University of Oxford have found. The UK, US, France, Germany and Italy are the source of 45% of the edits on ‘geocoded’ Wikipedia articles, which have a longitude and latitude associated with them to link them to a specific place in the world.”

Higher Social Status = Fewer International Facebook Friends?

People of a higher social status have fewer international friends? “A new study conducted in collaboration with Facebook using anonymised data from the social networking site shows a correlation between people’s social and financial status, and the levels of internationalism in their friendship networks – with those from higher social classes around the world having fewer friends outside of their own country”

How Hashtags And Symbols Affect Language on Twitter

Interesting: How hashtags and symbols affect language on Twitter. “Despite all the shortened words and slang seen on Twitter, it turns out that people follow many of the same communication etiquette rules on social media as they do in speech. Research from the Georgia Institute of Technology shows that when tweeters use hashtags — a practice that can enable messages to reach more people — they tend to be more formal and drop the use of abbreviations and emoticons. But when they use the @symbol to address smaller audiences, they’re more likely to use non-standard words such as ‘nah,’ ‘cuz’ and ‘smh.’”

Using Twitter Volume to Predict the Death of an App

Interesting: using Twitter volume to predict the death of apps. The researchers call it “social decay”. “Using Twitter data, BuzzFeed News analyzed the health of dozens of social apps to determine which ones might be fading away. To study a particular app, we tracked how the number of tweets linking to the app — for example, this tweet, which links to a live stream on Periscope, or this one linking to a live stream on Meerkat — changed over time. If the chart showed a steady decline over a number of months, we interpreted that as a warning sign. You might call it social decay.”

New Yorker Lady Hates New Google Logo

Do you hate the new Google logo? This lady does. “We loved the old logo, and we loved what Google was. Whatever it’s up to, whatever its intentions, Google should want to keep our love. So in the name of love, Google, give us back our serifs. Let this sans-serif building-block refrigerator-magnet silliness be the New Coke to your Coke, the Qwikster to your Netflix, the Freedom Tower to your One World Trade. Go back to your beautiful old serifs, and we’ll be that much likelier to let your self-driving cars drive us around.” Sorry, with Google there’s a lot more important stuff to care about than the logo (can you tell I’m not a designer).

Researchers Use Instagram Data to Predict Next Top Model

Indiana University Bloomington is using Instagram to make predictions about the next top model. Fashion model. Not train model or anything like that. “Researchers at Indiana University have predicted the popularity of new faces to the world of modeling with over 80 percent accuracy using advanced computational methods and data from Instagram. To conduct their analysis, IU scientists gathered statistics on 400 fashion models from the Fashion Model Directory, a major database of professional female fashion models, tracking hair and eye color; height; hip, waist, dress and shoes size; modeling agency; and runways walked.”

How Americans Use Twitter for News

The Knight Foundation did some research into how Americans use Twitter for news. “In order to better understand how Americans are engaging with news on Twitter, we built a small but representative sample of 176 Twitter users from an earlier national survey of 3,212 Americans conducted by Pew Research Center in association with the John S. and James L. Knight Foundation. We then analyzed the Twitter activity of these users, with their explicit permission.”

Google Training AI to Detect Pedestrians

It’s not as entertaining as Deep Dream, but Google is training its AI to detect pedestrians. Quickly. “We present a new real-time approach to object detection that exploits the efficiency of cascade classifiers with the accuracy of deep neural networks. Deep networks have been shown to excel at classification tasks, and their ability to operate on raw pixel input without the need to design special features is very appealing. However, deep nets are notoriously slow at inference time. In this paper, we propose an approach that cascades deep nets and fast features, that is both extremely fast and extremely accurate.”

Nature Publishes Survey Results on Open Access Publishing

Nature has published the results of a survey on open access publishing. “A survey of 22,000 academic researchers by Nature Publishing Group (NPG) and Palgrave Macmillan has found that a decreasing number of authors are concerned about perceptions of the quality of open access publications. In 2014, 40% of scientists who had not published open access in the last three years said ‘I am concerned about perceptions of the quality of OA publications.’ But this year, only 27% said they were concerned. In the humanities, business and social sciences (HSS), the drop was more marked; from 54% in 2014 to 41% in 2015. Nonetheless, concerns about perceptions of the quality of OA publications is still the leading factor in authors choosing not to publish OA.”