Hi! I've just released keras-surgeon, a network pruning library for Keras implemented in python. A few months ago I wanted to experiment with pruning neural network channels and discovered that there wasn't really a straightforward way to do that. Seeing as I wanted to get some more python experience under my belt, I have implemented a solution. The aim is to facilitate repeating academic studies, applying their findings to real problems or doing new research using Keras. Source code documentation (see readme): https://github.com/BenWhetton/keras-surgeon

For those who criticize pop music for being manufactured and predictable--you're in for a treat. In the past, it's taken scores of writers to come up with a decent pop music tune. Six people were on hand to pen Beyoncé's "Run the World (Girls)". Six more were necessary for Flo Rida's "Whistle". Now, it might not need any at all.

To commemorate the silver jubilee of FICO's use of artificial intelligence and machine learning, we asked FICO employees a question: What does the future of AI look like? The post below is one of the thought-provoking responses, from Shafi Rahman, a principal scientist at FICO, working in San Diego. The most memorable scene in the film 2001: A Space Odyssey is when HAL 3000 turns off the life-support functions of the crew in an attempt to keep itself from getting disconnected. This was my first encounter with what we call artificial intelligence. Then in Terminator, we were introduced to Skynet, a neural network based AI, which raised an army of robots for self-preservation.

It's 2050 and the world revolves around you. From the contents of your fridge to room temperature, digital assistants ensure your home runs smoothly. Your screens know your taste and show channels you want to see as you enter the room. Your car is driverless and your favourite barman may just be an Android. Predictions for an AI-dominated future are increasingly common, but Antoine Blondeau has experience in reading, and arguably manipulating, the runes, he helped develop technology that evolved into predictive texting and Apple's Siri.

Summary: we looked at how an attacker might go about backdooring a CNN when the training is outsourced. The upshot is that it's pretty easy to get a network to learn to treat the presence of a "backdoor trigger" in the input specially without affecting the performance of the network on inputs where the trigger is not present. We also looked at transfer learning: if you download a backdoored model from someplace like the Caffe Model Zoo and fine-tune it for a new task by retraining the fully connected layers, it turns out that the backdoor can survive the retraining and lower the accuracy of the network when the trigger is present! It appears that retraining the entire network does make the backdoor disappear, but we have some thoughts on how to get around that that didn't make it into the paper. We argue that this means you need to treat models you get off the internet more like software and be careful about making sure you know where they came from and how they were trained.

For Amazon Echo owners in the UK, the arrival of Music Unlimited provided the likes of Apple and Spotify with some music needed competition. The company immediately made it the default music service on the smart speaker, allowing users to instantly queue music using their voice once they'd signed up. Understandably, Amazon is keen to differentiate its service from its more popular rivals, so it's banking on Alexa "activities," which have today arrived in the UK. Activities are basically Amazon-built voice controls that make it easier to find music for a specific event or setting. Users have apparently requested ways to find a perfect meditation playlist, or music that matches the theme of a party or a workout and the company is only too keen to oblige.

On the very first day of Wolfram Camp, I called Stephen Wolfram "Steve." "It's Stephen, actually," said the world's most controversial physicist in his crisp-yet-droll British accent. In another life, the creator of Wolfram Alpha would have made an excellent BBC Radio News announcer. "No one under the age of 50 calls me Steve," he added. Katie Orenstein is a New York City-based writer, programmer, and thespian who moonlights as a high school senior.

Science fiction gets a bad rap in some circles for being a genre with nothing but a bunch of spaceships and robots. It's also got drama, action, and even some romance. It's also full of spaceships and robots and if you're not really into those things, well, the door is over there. If you do adore droids and contraptions that fly through the cosmos, though, we have something here that's just for you. Below is a selection of the best sci-fi movies you can currently stream on Netflix, Hulu, and Amazon.

Qualcomm announced that it has purchased Dutch artificial intelligence research company Scyfer to boost its expertise in machine learning. The price was not disclosed. Scyfer has built artificial intelligence systems for companies in manufacturing, healthcare, finance and other industries. Founded in 2013, Scyfer is a spinoff of the University of Amsterdam. Professor Max Welling co-founded the company and will continue to work both as a professor and an artificial intelligence researcher for Qualcomm.

As film composer Drew Silverstein, one of Amper's founders, explained to TechCrunch, Amper isn't meant to act totally on its own, but was designed specifically to work in collaboration with human musicians: "One of our core beliefs as a company is that the future of music is going to be created in the collaboration between humans and AI. We want that collaborative experience to propel the creative process forward."