We present a method for explaining the image classification predictions of deep convolution neural networks, by highlighting the pixels in the image which influence the final class prediction. Our method requires the identification of a heuristic method to select parameters hypothesized to be most relevant in this prediction, and here we use Kullback-Leibler divergence to provide this focus. Overall, our approach helps in understanding and interpreting deep network predictions and we hope contributes to a foundation for such understanding of deep learning networks. In this brief paper, our experiments evaluate the performance of two popular networks in this context of interpretability.

Recently, deep learning approaches with various network architectures have achieved significant performance improvement over existing iterative reconstruction methods in various imaging problems. However, it is still unclear why these deep learning architectures work for specific inverse problems. To address these issues, here we show that the long-searched-for missing link is the convolution framelets for representing a signal by convolving local and non-local bases. The convolution framelets was originally developed to generalize the theory of low-rank Hankel matrix approaches for inverse problems, and this paper further extends the idea so that we can obtain a deep neural network using multilayer convolution framelets with perfect reconstruction (PR) under rectilinear linear unit nonlinearity (ReLU). Our analysis also shows that the popular deep network components such as residual block, redundant filter channels, and concatenated ReLU (CReLU) do indeed help to achieve the PR, while the pooling and unpooling layers should be augmented with high-pass branches to meet the PR condition. Moreover, by changing the number of filter channels and bias, we can control the shrinkage behaviors of the neural network. This discovery leads us to propose a novel theory for deep convolutional framelets neural network. Using numerical experiments with various inverse problems, we demonstrated that our deep convolution framelets network shows consistent improvement over existing deep architectures.This discovery suggests that the success of deep learning is not from a magical power of a black-box, but rather comes from the power of a novel signal representation using non-local basis combined with data-driven local basis, which is indeed a natural extension of classical signal processing theory.

And nowhere is that more true than in high performance computing. An insideHPC special report, sponsored by Intel, explores one of the most interesting and cutting-edge areas of AI, and that's the convergence of deep learning, data and HPC. This convergence is making AI technology more accessible to data scientists with no coding background required. In fact, according to the report, sponsored by Intel, HPC and the artificial intelligence communities are converging as they are both running similar types of data and compute intensive workloads on HPC hardware, including supercomputers, institutional clusters or the cloud. "The emerging AI community on HPC infrastructure is critical to achieving the vision of AI: machines that don't just crunch numbers, but help us make better and more informed complex decisions."

Doxel is a startup that came out of stealth this week with a US $4.5 million funding round. Their business is making construction cheaper, and their secret (as with so many startups now) is combining massive amounts of data with deep learning techniques. Using lidar-equipped robots, Doxel scans construction sites every day to monitor how things are progressing, tracking what gets installed and whether it's the right thing at the right time in the right place. You'd think that construction sites would be doing this by themselves anyway, but it turns out that they really don't, and in a recent pilot study on a medical office building, Doxel says it managed to increase labor productivity on the project by a staggering 38 percent. The concept behind Doxel is straightforward enough: Construction projects have plans and budgets and timelines.

In 2015, Google announced that it had added RankBrain to its algorithm, cementing the importance of artificial intelligence (AI) in search. Fast-forward to 2018, and search marketers are starting to use AI, machine learning and deep learning systems to uncover new insights, automate labor-intensive tasks and provide a whole new level of personalization to guide website visitors through their purchase funnel. We have now fully entered the AI revolution. Today's technology giants are all heavily invested in the potential of these AI methods to deliver better products and services, as they provide scale and computational power that humans alone could never offer. Of course, this technology has risen to prominence in the age of big data.

Which authors of this paper are endorsers? Disable MathJax (What is MathJax?)

Deep multi-task learning with low level tasks supervised at lower layers.

Machine Learning is a hot topic these days, as can be seen from search trends. It was the success of Deepmind and AlphaGo in 2016 that really brought machine learning to the attention of the wider community and the world at large. Yet it's a success that followed a long preamble that includes recent advances in three key areas: hardware, particularly GPUs (ideally suited to the vector and matrix based mathematics usually required in machine learning); data, due to the accessibility of larger and larger datasets; and algorithms and techniques, as deep learning research breakthroughs like those described in Krizhevsky, Sutskever and Hinton's landmark paper began to demonstrate best-of-breed results on benchmark challenges. So it's not just hype, and as IT engineers it's worth our while to gain better understanding of it. But the field can seem rather daunting to a newcomer due to all the math, statistics and algorithms involved.

Canada is gearing up for its second wave of brain gains in the hot field of artificial intelligence. The country's global leadership in AI began when deep learning pioneer Geoff Hinton decamped from Pittsburgh's Carnegie Mellon University (CMU) for the University of Toronto three decades ago, and this month Canada welcomed two big names from Dr. Hinton's alma mater. This week, Geoff Gordon, one of the top professors at CMU's renowned machine learning department, left his post to lead Microsoft Corp.'s fledgling AI research lab in Montreal, established when the software giant bought local startup Maluuba a year ago. Dr. Gordon expects others to follow him to the hometown of deep learning pioneer Yoshua Bengio, a University of Montreal professor, Microsoft adviser and co-founder of the Montreal Institute for Learning Algorithms (MILA). "As soon as I've told anybody that I'm becoming research director here, the first or second question they ask is: 'What are the opportunities to come work there?'" Dr. Gordon said, adding he intends to recruit CMU researchers as Microsoft aims to double the number of technical experts at its Montreal AI lab to 75 people within two years.

This video was based on Max's book "Life 3.0", which you can find at: http://amzn.to/2iEwe6w We live in an era of self driving cars, autonomous drones, deep learning algorithms, computers that beat humans at chess and go, and so on. So it's natural to ask, will artificial superintelligence replace humans, take our jobs, and destroy human civilization? Or will AI just become tools like regular computers. AI researcher Max Tegmark helps explain the myths and facts about superintelligence, the impending machine takeover, etc. MinutePhysics is on twitter - @minutephysics And facebook - http://facebook.com/minutephysics