Advances in artificial intelligence are happening in big ways, but the progress is all from the most technical minds and companies on Earth. Those in the liberal arts and humanities, however, are now thinking they need to get in on AI before it drastically changes how we live. To do AI research for the public interest from an entirely new perspective, LinkedIn founder Reid Hoffman, the Omidyar Network (a philanthropic investment firm) and the Knight Foundation (which invests in journalism and arts) have put together a $27 million fund for AI research. Called the "Ethics and Governance of Artificial Intelligence Fund," it applies the humanities, social sciences and other disciplines to the development of AI. "Artificial intelligence and complex algorithms in general, fueled by big data and deep-learning systems, are quickly changing how we live and work…" reads the announcement from the Knight Foundation. "Because of this pervasive but often concealed impact, it is imperative that AI research and development be shaped by a broad range of voices--not only by engineers and corporations, but also by social scientists, ethicists, philosophers, faith leaders, economists, lawyers and policymakers."

Pinterest is overhauling its related pins feature using artificial intelligence. With approximately 75 billion pins on its platform, curation is a critical part of the service Pinterest offers its 150 million users. In order to help users discover new items, recipes, and ideas that relate to their interests, Pinterest is increasingly turning to deep learning -- an AI that is fed vast amounts of data through a neural network that simulates how a human brain works until it can autonomously recognize other data. More: Facebook's new'M Suggestions' could bring more AI to Messenger The platform's related pins feature pops up beneath a Pin with similar recommendations. In the past these suggestions were surfaced from boards that the same Pin had been saved to by other users.

Forbes Shutterstock Image READ MORE 6. "There an estimated 3,000 AI startups worldwide, and many of them are building on NVIDIA's platform. They're using NVIDIA's GPUs to put AI into apps for trading stocks, shopping online and navigating drones." Read more … Aaron Tilley Writer 7. The retail sector is now best positioned to leverage AI and Deep Learning, as these new technologies are developing… 8. READ MORE AI software such as Computer Vision is being developed by startups to help retail consumers find the perfect and individualized fit. THIRD LOVE A app that enables women to find the right fitting bra from home using a mobile device and deep learning. VOLUMENTAL Offers computer vision applications for sizing shoes and eyewear to create a individualized retail experience for customers.

This could a little late given that we have already embarked upon a new year. But it could be worthwhile looking back for a moment... If that's a little far fetched, considering the wide use of drones, advances in VR/AR and blockchain, that's because of the'bias' (read enthusiasm) in my neurons. I haven't been for long in this field but after Deepmind's paper a few years back, this year was among the first to show commercial viability of AI and showed how well poised it is for a few established problems. I have tried here to distil some major events that happened earlier in 2016 but, you know, like all networks, my brain might have missed out on some signals.

First of all, I am a technologist living in an academic environment as a professor of computer science at Northwestern University. But I am also a co-founder of Narrative Science, an advanced natural language generation company that is a thriving business. And my work in both of these worlds has lead to a substantial amount of time spent with different government organizations that are considering how and when A.I. should be utilized. Each world has its own concerns and it is no secret that they occasionally look at each other with some skepticism. Academics and businesses each see the other as completely missing the point.

We study large-scale kernel methods for acoustic modeling in speech recognition and compare their performance to deep neural networks (DNNs). We perform experiments on four speech recognition datasets, including the TIMIT and Broadcast News benchmark tasks, and compare these two types of models on frame-level performance metrics (accuracy, cross-entropy), as well as on recognition metrics (word/character error rate). In order to scale kernel methods to these large datasets, we use the random Fourier feature method of Rahimi and Recht (2007). We propose two novel techniques for improving the performance of kernel acoustic models. First, in order to reduce the number of random features required by kernel models, we propose a simple but effective method for feature selection. The method is able to explore a large number of non-linear features while maintaining a compact model more efficiently than existing approaches. Second, we present a number of frame-level metrics which correlate very strongly with recognition performance when computed on the heldout set; we take advantage of these correlations by monitoring these metrics during training in order to decide when to stop learning. This technique can noticeably improve the recognition performance of both DNN and kernel models, while narrowing the gap between them. Additionally, we show that the linear bottleneck method of Sainath et al. (2013) improves the performance of our kernel models significantly, in addition to speeding up training and making the models more compact. Together, these three methods dramatically improve the performance of kernel acoustic models, making their performance comparable to DNNs on the tasks we explored.

In this paper, we propose an efficient transfer leaning methods for training a personalized language model using a recurrent neural network with long short-term memory architecture. With our proposed fast transfer learning schemes, a general language model is updated to a personalized language model with a small amount of user data and a limited computing resource. These methods are especially useful for a mobile device environment while the data is prevented from transferring out of the device for privacy purposes. Through experiments on dialogue data in a drama, it is verified that our transfer learning methods have successfully generated the personalized language model, whose output is more similar to the personal language style in both qualitative and quantitative aspects.

A developer spent a couple of days over his winter break training an artificial neural network to play the classic racing game Mario Kart 64 and documented his results to share what he learned in the process. "It had been a few years since I'd done any serious machine learning, and I wanted to try out some of the new hotness (aka TensorFlow) I'd been hearing about," said Kevin Hughes who works as a developer at Shopify. Using a GeForce GTX 1060 GPU and the cuDNN-accelerated TensorFlow deep learning framework, Hughes trained his model on only four races on Luigi Raceway, two races on Kalimari Desert and two races on Mario Raceway. He mentioned that even with a small training set, the model was able to drive a new untrained section of the Royal Raceway (below). Hughes' adds that he plans on adding a reinforcement layer to the project so the AI can start to teach itself.

How is machine learning transforming bioscience research? The relationship between biology and machine learning is not new and has existed for decades, even before data science and machine learning became fashionable. Fields like protein structure prediction, homology modeling [1] [2] [3] and cheminformatics [4] frequently employ tools from machine learning. For a long time machine learning was defined by the ability to choose effective features, which is often (a) labor intensive and (b) requires a need to understand or have an idea about solutions, which limited the application of machine learning. It is also important to keep in mind that biological data derived from experiments are prone to error, hence domain specific knowledge is almost always required, and biological or-omics data tend to be high dimensional and sparse.