Allstate deployed Amelia as a test in September and, to date, has used the platform in more than 3 million customer conversations. With an instant message interface, Amelia leads call center employees through step-by-step procedures to help answer a variety of customer questions, including policy and policyholder information. "Amelia is quickly becoming an important component of our customer service strategy," said Allstate Senior Vice President Carla Zuniga. "She provides our call center personnel with the information and procedures they need to address our customers' questions and concerns." Amelia's machine learning capability allows her to improve her ability to answer customer questions by "listening" to interactions she doesn't understand and with Allstate experts' help quickly expanding her knowledge.

Sonic Style is hoping to shake up the way music is traditionally categorized, moving past the typical overarching genres like rock or hip-hop to classify each song on a granular level. To that end, Gracenote has amassed nearly 450 Sonic Style descriptors that create a "style profile" of each recording and can pair machine learning descriptors such as tempo and mood with editorial ones, like artist genre, era, and origin. For example, the press release suggests the service can mine through Taylor Swift's catalogue to figure out which songs are more pop, more country, more pop electronica, or god help us, more R&B. That level of musical understanding will hopefully help smart speakers, voice assistants, streaming services, and, yes, even measly humans create better, more personalized playlists. "Sonic Style applies neural network-powered machine learning to the world's music catalogs, enabling Gracenote to deliver granular views of musical styles across complete music catalogs," says Brian Hamilton, Gracenote's general manager of music and auto, in a statement.

The spectrum of desire that drives the design and architecture worlds has shifted. Where homes were once considered a showcase of wealth and access to the world's most elusive treasures, they are now a comfortable hub – a chamber within which we find space to cleanse ourselves of life, and clasp moments of calm to our chests, snatched from the chaos and demand of a connected world. For many, luxury has been redefined. It now revolves around adding convenience, function and intuition to the home, inevitably culminating in society entering a state of comatose artificial intelligence reliance, or at least that's what the film industry would have you believe. "The future is big business but we have had it in our lives forever and always take it in stride," says Academy Award-nominated production designer KK Barrett.

Apple's "Amazing Stories" revival finally has new showrunners on board. The tech giant has inked a deal with "Once Upon a Time" co-creators Adam Horowitz and Edward Kitsis in order for the two to take on the role and executive produce the project. Variety first reported that Horowitz and Kitsis have now assumed the role that Bryan Fuller left in February. The announcement of their involvement in the project comes days after ABC aired the finale of "Once Upon a Time" season 7, which is the fantasy series' final run. As what IBTimes learned last February, Fuller exited the reboot of Steven Spielberg's science fiction horror series due to creative differences.

Facebook AI Research (FAIR) scientists yesterday unveiled a neural network capable of translating music from one style, genre, and set of instruments to another. Soon, you won't have to blow your own horn; you can just whistle to an AI and it'll turn your song into the symphony or dance hit of your dreams. The AI takes one input, such as a symphony orchestra playing Bach, and translates it into something else, like the same song played on a piano in the style of Beethoven, for example. FAIR becomes the first AI research team to create an unsupervised learning method for recreating high-fidelity music with a neural network. Our results present abilities that are, as far as we know, unheard of. Asked to convert one musical instrument to another, our network is on par or slightly worse than professional musicians.

In recommender systems, cold-start issues are situations where no previous events, e.g. ratings, are known for certain users or items. In this paper, we focus on the item cold-start problem. Both content information (e.g. item attributes) and initial user ratings are valuable for seizing users' preferences on a new item. However, previous methods for the item cold-start problem either 1) incorporate content information into collaborative filtering to perform hybrid recommendation, or 2) actively select users to rate the new item without considering content information and then do collaborative filtering. In this paper, we propose a novel recommendation scheme for the item cold-start problem by leverage both active learning and items' attribute information. Specifically, we design useful user selection criteria based on items' attributes and users' rating history, and combine the criteria in an optimization framework for selecting users. By exploiting the feedback ratings, users' previous ratings and items' attributes, we then generate accurate rating predictions for the other unselected users. Experimental results on two real-world datasets show the superiority of our proposed method over traditional methods.

Multi-armed bandit algorithms have become a reference solution for handling the explore/exploit dilemma in recommender systems, and many other important real-world problems, such as display advertisement. However, such algorithms usually assume a stationary reward distribution, which hardly holds in practice as users' preferences are dynamic. This inevitably costs a recommender system consistent suboptimal performance. In this paper, we consider the situation where the underlying distribution of reward remains unchanged over (possibly short) epochs and shifts at unknown time instants. In accordance, we propose a contextual bandit algorithm that detects possible changes of environment based on its reward estimation confidence and updates its arm selection strategy respectively. Rigorous upper regret bound analysis of the proposed algorithm demonstrates its learning effectiveness in such a non-trivial environment. Extensive empirical evaluations on both synthetic and real-world datasets for recommendation confirm its practical utility in a changing environment.

We present a method for translating music across musical instruments, genres, and styles. This method is based on a multi-domain wavenet autoencoder, with a shared encoder and a disentangled latent space that is trained end-to-end on waveforms. Employing a diverse training dataset and large net capacity, the domain-independent encoder allows us to translate even from musical domains that were not seen during training. The method is unsupervised and does not rely on supervision in the form of matched samples between domains or musical transcriptions. We evaluate our method on NSynth, as well as on a dataset collected from professional musicians, and achieve convincing translations, even when translating from whistling, potentially enabling the creation of instrumental music by untrained humans.

Training a task-completion dialogue agent via reinforcement learning (RL) is costly because it requires many interactions with real users. One common alternative is to use a user simulator. However, a user simulator usually lacks the language complexity of human interlocutors and the biases in its design may tend to degrade the agent. To address these issues, we present Deep Dyna-Q, which to our knowledge is the first deep RL framework that integrates planning for task-completion dialogue policy learning. We incorporate into the dialogue agent a model of the environment, referred to as the world model, to mimic real user response and generate simulated experience. During dialogue policy learning, the world model is constantly updated with real user experience to approach real user behavior, and in turn, the dialogue agent is optimized using both real experience and simulated experience. The effectiveness of our approach is demonstrated on a movie-ticket booking task in both simulated and human-in-the-loop settings.

I do think that beating TensorFlow on tasks like logistic regression is not particularly hard. A student asked me once to help optimize his Tf code for a large scale linear regression model on multiple GPUs. It was magnitudes slower than the single-core scikit-learn implementation. We spent hours trying to get the best performance out of it, including various experiments with the data loading directly to the GPU tensors bypassing the Python runtime. TensorFlow is just not optimized for this kind of stuff because of various overheads, I assume. People underestimate how fast scikit-learn is for generalized linear models thanks to BLAS and LIBLINEAR.