We revisit the importance of the individual units in Convolutional Neural Networks (CNNs) for visual recognition. By conducting unit ablation experiments on CNNs trained on large scale image datasets, we demonstrate that, though ablating any individual unit does not hurt overall classification accuracy, it does lead to significant damage on the accuracy of specific classes. This result shows that an individual unit is specialized to encode information relevant to a subset of classes. We compute the correlation between the accuracy drop under unit ablation and various attributes of an individual unit such as class selectivity and weight L1 norm. We confirm that unit attributes such as class selectivity are a poor predictor for impact on overall accuracy as found previously in recent work \cite{morcos2018importance}. However, our results show that class selectivity along with other attributes are good predictors of the importance of one unit to individual classes. We evaluate the impact of random rotation, batch normalization, and dropout to the importance of units to specific classes. Our results show that units with high selectivity play an important role in network classification power at the individual class level. Understanding and interpreting the behavior of these units is necessary and meaningful.

Sentiment classification involves quantifying the affective reaction of a human to a document, media item or an event. Although researchers have investigated several methods to reliably infer sentiment from lexical, speech and body language cues, training a model with a small set of labeled datasets is still a challenge. For instance, in expanding sentiment analysis to new languages and cultures, it may not always be possible to obtain comprehensive labeled datasets. In this paper, we investigate the application of semi-supervised and transfer learning methods to improve performances on low resource sentiment classification tasks. We experiment with extracting dense feature representations, pre-training and manifold regularization in enhancing the performance of sentiment classification systems. Our goal is a coherent implementation of these methods and we evaluate the gains achieved by these methods in matched setting involving training and testing on a single corpus setting as well as two cross corpora settings. In both the cases, our experiments demonstrate that the proposed methods can significantly enhance the model performance against a purely supervised approach, particularly in cases involving a handful of training data.

As smart speakers gained popularity, a big name in the space seemed to have fallen by the wayside. Sonos pioneered a high-quality connected home-theater speaker system in the early 2000s. In 2016, it began making a shift to support streaming music services on its products. It shipped its first assistant-laden speaker, the Alexa-enabled Sonos One, in 2017, but has now taken that idea even further. The company, on Wednesday, unveiled Beam, a smart sound bar, which will soon be virtual assistant agnostic: It's shipping with Alexa, gaining Siri control with AirPlay 2 in July, and adding Google Assistant compatibility later this year.

Sonos took the wraps off its latest smart speaker at an event in San Francisco earlier today. The $399 Sonos Beam is smaller and less expensive than either the Sonos Playbase or the older Sonos Playbar, and it will support Apple's AirPlay 2 multi-room audio technology when it ships in July. "All the tech giants are building smart speakers," said Chris Kallai, Sonos' VP of Hardware Product Management, when introducing the Sonos Beam. "But they're not doing it so you can listen to all the music you want. Kallai said the Beam combines three products into one: "Alexa, Apple Music, and a compact soundbar."

Close your eyes and try to picture a journalist. Well, what if all of that was replaced … by robots? Okay, our show isn't about to be hosted by a machine (yet). But artificial intelligence is already being used in newsrooms today. For instance, there's Heliograf, a bot developed by The Washington Post.

On Tuesday, Tesla held its annual shareholder meeting. As expected, none of the controversial shareholder votes passed--there will be no independent CEO replacing Musk, and his brother and the other two candidates were reelected to the board with no drama or fireworks. Even Tesla's chief counsel Todd Maron referred to the opening agenda items as "the boring bits." Things got more interesting once Musk took to the stage for a Q&A session, answering queries submitted in advance via Twitter and then from the audience. It was an odd performance, often feeling more like a Netflix comedy special than a shareholder meeting.

Between Boston Dynamics' headless, door-opening cyborg dogs, Omron Automation's empathy chips for robots, and MIT Media Lab's new'psychopath' AI Norman (named for Norman Bates, the murderer from Alfred Hitchcock's Psycho), it looks like we've got everything we need to cobble together a darker, more violent version of I, Robot, or at least another episode of Black Mirror Season 5. Seriously, listen to this: A team of MIT employees took a normal image-captioning AI (designed to look at pictures and provide a written description of what it sees) and fed it a steady stream of images from an unnamed Reddit board where people exclusively post horrifying, morbid images of murder and death. Afterward, the team showed this AI (now dubbed Norman) a series of Rorschach inkblots, which are used by psychiatrists and psychoanalysts to judge a patient's mental state. The team then compared Norman's captions to a normal AI that had not been traumatized with images of death and found a disturbing pattern. For example, Norman captioned one inkblot "man is murdered by machine gun in broad daylight", while the other AI captioned the same image, "a black and white photo of a baseball glove." Norman's morbidity plays out again and again in the tests.

Spotify is gearing up to launch its own range of hardware products fuelling rumours the company plans to launch its own smart speaker. The music streaming service has registered with the Federal Communications Commission (FCC), paving the way for it to sell hardware products for the first time. Without proper certification from the FCC, it is impossible to market or sell wireless products in the United States. Spotify has yet to submit any hardware for certification from the US government agency. The company has long been rumoured to be developing its own hardware, however, its registration with the FCC is the first evidence it is considering bringing products to market.

Scientists at the Massachusetts Institute of Technology (MIT) trained an artificial intelligence algorithm dubbed "Norman" to become a psychopath by only exposing it to macabre Reddit images of gruesome deaths and violence, according to a new study. Nicknamed Norman after Anthony Perkins' character in Alfred Hitchcock's 1960 film Psycho, the artificial intelligence was fed only a continuous stream of violent images from various pernicious subreddits before being tested with Rorschach inkblot tests. The imagery detected by Norman produced spooky interpretations of electrocutions and speeding car deaths where a standard AI would only see umbrellas and wedding cakes. Scientists at the Massachusetts Institute of Technology (MIT) trained an AI algorithm, "Norman," to become a psychopath by only exposing it to macabre Reddit images of gruesome deaths and violence. MIT scientists Pinar Yanardag, Manuel Cebrian and Iyad Rahwan specifically trained the AI to perform image captioning, a "deep learning method" for artificial intelligence to cull through images and produce corresponding descriptions in writing.

Smart speaker and home audio company Sonos is holding an event in San Francisco today, and the first bit of news is that AirPlay 2 support is coming in July. As previously announced Apple's latest streaming protocol will work with the Sonos One, Playbase, and Play:5 speakers -- anything older than that won't be supported, unfortunately. The update will come via software update, and it'll let you group together multiple Sonos speakers through Apple's apps. This means you can start up songs in Apple Music and send them directly to Sonos without having to use the dedicated Sonos app. Sonos gave a quick demo showing off what AirPlay 2 can do.