The results also provide further evidence that embeddings have a major impact on the performance of such architectures.

Binarized Neural Network (BNN) removes bitwidth redundancy in classical CNN by using a single bit (-1/+1) for network parameters and intermediate representations, which has greatly reduced the off-chip data transfer and storage overhead. However, a large amount of computation redundancy still exists in BNN inference. By analyzing local properties of images and the learned BNN kernel weights, we observe an average of $\sim$78% input similarity and $\sim$59% weight similarity among weight kernels, measured by our proposed metric in common network architectures. Thus there does exist redundancy that can be exploited to further reduce the amount of on-chip computations. Motivated by the observation, in this paper, we proposed two types of fast and energy-efficient architectures for BNN inference. We also provide analysis and insights to pick the better strategy of these two for different datasets and network models. By reusing the results from previous computation, much cycles for data buffer access and computations can be skipped. By experiments, we demonstrate that 80% of the computation and 40% of the buffer access can be skipped by exploiting BNN similarity. Thus, our design can achieve 17% reduction in total power consumption, 54% reduction in on-chip power consumption and 2.4$\times$ maximum speedup, compared to the baseline without applying our reuse technique. Our design also shows 1.9$\times$ more area-efficiency compared to state-of-the-art BNN inference design. We believe our deployment of BNN on FPGA leads to a promising future of running deep learning models on mobile devices.

In this paper, we introduce McTorch, a manifold optimization library for deep learning that extends PyTorch. It aims to lower the barrier for users wishing to use manifold constraints in deep learning applications, i.e., when the parameters are constrained to lie on a manifold. Such constraints include the popular orthogonality and rank constraints, and have been recently used in a number of applications in deep learning. McTorch follows PyTorch's architecture and decouples manifold definitions and optimizers, i.e., once a new manifold is added it can be used with any existing optimizer and vice-versa. McTorch is available at https://github.com/mctorch .

In the context of Europe, the UK is one of the leading places for AI adoption. "I would say it's the UK and then Germany and France," explains Ben Lorica, chief data scientist at O'Reilly Media. "If you were to look at the world as a whole, I think at this moment at least, because of the nature of the technology involved, the US and China are considered the leaders." This is because a lot of the emerging technologies, such as artificial intelligence, rely on massive amounts of data and computing to be successful. In these two markets (US and China), companies can scale up to many users right away, because of the availability of larger datasets and computational power.

In the current AI Spring, many people and corporations are betting big that the capabilities of deep learning algorithms will continue to improve as the algorithms are fed more data. Their faith is backed by the miracles performed by such algorithms: they can see, listen and do a thousand other things that were previously considered too difficult for AI. Our guest for the third episode of the AGI Podcast, Pascal Kaufmann, is amongst those who believe such faith in deep learning is misplaced. Rather than putting his faith in deep learning and other popular methods that seek to mimic the workings of the human brain, Pascal is taking a different route to create AGI. History seems to be on the side of Pascal.

A new research article, THE THERMODYNAMIC ANALYSIS OF NEURAL COMPUTATION, examines the energy-information cycle of sensory processing in the brain. However, the work has relevance for artificial intelligence, especially deep learning. The cortical brain is an evolutionary marvel which interacts with the outside world via self-regulation, based on its resting or ground state. The resting state, which is disturbed during stimulus and sensory processing, is recovered by automatic operations. The brain's energy need multiplies in the complex brain of warm-blooded animals.

We can use Bing on Spark to quickly create our own machine learning datasets featuring anything we can find online. To create a custom snow leopard dataset takes only two distributed queries. The first query creates the "positive class" by pulling the first 80 pages of the "snow leopard" image results. The second query creates the "negative class" to compare our leopards against. We can perform this search in two different ways, and we plan to explore them both in upcoming posts.

Artificial intelligence is going to have a significant impact on the future of insurance. This revolution is not far off, and the industry is on the verge of a monumental, tech-driven shift. The deep learning technologies needed for this are already here -- think neural networks or machine learning techniques. All of these fall under the umbrella of artificial intelligence, and will help the insurance industry move from a'detect and repair' model to a'predict and prevent' model. It's a similar story with other industries which are considering implementing AI. 'The pace of change will also accelerate as brokers, consumers, financial intermediaries, insurers, and suppliers become more adept at using advanced technologies to enhance decision making and productivity, lower costs and optimise the customer experience,' writes Ramnath Balasubramanian, partner at McKinsey, Ari Libarikian, senior partner at McKinsey, and Doug McElhaney, associate partner at McKinsey. "Currently, the greatest benefit we see is an enhanced and prioritised customer experience, followed by streamlined back-office processing and improved fraud detection," explains Harald Gölles, CTO at omni:us.

Facebook today announced the release of deep learning framework PyTorch 1.0 in developer preview, which includes a series of tools and integrations to make it more compatible with popular services from Google Cloud, Amazon Web Services, and Microsoft's Azure Machine Learning. Arm, Nvidia, Qualcomm, and Intel are also adding PyTorch support for things like kernel library integrations and tools to track inference runtime. PyTorch was released to the public in January 2017 and has been downloaded more than 1 million times. PyTorch 1.0 was first announced in May at the F8 developer conference, and includes deeper integration with Facebook's Caffe2 and ONNX. Back in May, Facebook VP Bill Jia and CTO Mike Schroepfer promised PyTorch 1.0 would launch with new pretrained models, tools, and libraries to give developers more flexibility and options.

Yann Lecun is the scientist leading Facebook's AI efforts. Facebook is on course to double the size of the Facebook Artificial Intelligence Research (FAIR) division in the next two years, according to the company's chief AI scientist, Yann LeCun. FAIR currently has approximately 180-200 staff, but the division is expected to grow to around 400 people by 2020 as Facebook continues to put AI at the heart of its platforms. "I don't know everybody's name anymore and I don't recognise everybody either," LeCun admitted in an interview at Facebook's New York office. Asked whether FAIR is likely to double in size in the next couple of years, LeCun said: "Yes, probably. Members of FAIR carry out fundamental research in the field of AI. Some of their breakthroughs are applied to Facebook's platforms (Facebook, Instagram, and Whatsapp) by an applied machine learning (AML) team and other engineers, but the majority of their research is purely academic. So far they've developed algorithms that can analyse MRI scans and play games like "Starcraft," among other things. But competition for AI talent is intense. Facebook and Google are locked in a battle to hire the smartest minds in the field, while others like Apple, Amazon, and Microsoft are also trying to poach the best PhD students and other academics, leading to brain drain concerns. Facebook is hiring many of these people through FAIR, while Google, or Alphabet, as Google's parent company is known, is hiring through DeepMind, and Google Brain to some extent. DeepMind's team has swelled from less than 100 to over 700 since it was acquired by Google in 2014 for £400 million. Asked why FAIR hasn't grown at the same rate as some other AI labs, Rob Fergus, head of FAIR in New York, said: "It's a market supply.