We describe the class of convexified convolutional neural networks (CCNNs), which capture the parameter sharing of convolutional neural networks in a convex manner. By representing the nonlinear convolutional filters as vectors in a reproducing kernel Hilbert space, the CNN parameters can be represented as a low-rank matrix, which can be relaxed to obtain a convex optimization problem. For learning two-layer convolutional neural networks, we prove that the generalization error obtained by a convexified CNN converges to that of the best possible CNN. For learning deeper networks, we train CCNNs in a layer-wise manner. Empirically, CCNNs achieve performance competitive with CNNs trained by backpropagation, SVMs, fully-connected neural networks, stacked denoising auto-encoders, and other baseline methods.

We consider the problem of provably optimal exploration in reinforcement learning for finite horizon MDPs. We show that an optimistic modification to value iteration achieves a regret bound of order (HSAT)1/2 (up to a logarithmic factor) where H is the time horizon, S the number of states, A the number of actions and T the number of time-steps. This result improves over the best previous known bound HS(AT)1/2 achieved by the UCRL2 algorithm of [Jaksch, Ortner, Auer, 2010]. The key significance of our new results is that for large T, the sample complexity of our algorithm matches the optimal lower bound of Ω(HSAT)1/2. Our analysis contains two key insights.

Francesco is Data Scientist at Janssen Pharmaceutical Companies of Johnson & Johnson and a Science writer. He is committed to "A World Without Disease" paradigm shift in healthcare, leveraging Artificial Intelligence and Data Science to predict risk and intercepting diseases. He is focused on putting machine learning at the service of human beings. Do you know why you can't hear the ugly ahem sounds on the podcast Data Science at Home? Let me introduce the ahem detector, a deep convolutional neural network that is trained on transformed audio signals to recognize "ahem" sounds. The network has been trained to detect such signals on the episodes of Data Science at Home, the podcast about data science at worldofpiggy.com/podcast.

Deep Learning involves the interplay of Computer Science, Physics, Biology and Psychology. In addition to that, it has the potential to be extremely disruptive. Not just disruptive technologically, but also in how we view society and even our own humanity. I seriously don't think that you can stumble upon a subject as captivating and as consequential as this. It is indeed surprising that there is very little that is written about this in a more comprehensive and cohesive way. This book is an opinionated take on this.

The next paper from our reading group is by Forrest N. Iandola, Matthew W. Moskewicz, Khalid Ashraf, Song Han, William J. Dally and Kurt Keutzer. This paper introduces a small CNN architecture called "SqueezeNet" that achieves AlexNet-level accuracy on ImageNet with 50x fewer parameters. As you may have noticed with one of our recent posts we're really interested in learning more about the compression of neural network architectures and this paper really stood out. It's no secret that much of deep learning is tied up in the hell that is parameter tuning. This paper makes a case for increased study into the area of convolutional neural network design in order to drastically reduce the number of parameters you have to deal with.

At the 2017 GPU Technology Conference NVIDIA announced CUDA 9, the latest version of CUDA's powerful parallel computing platform and programming model. In this post I'll provide an overview of the awesome new features of CUDA 9. The CUDA 9 release candidate is now available. To learn more you can watch the recording of my talk from GTC 2017, "CUDA 9 and Beyond". The soul of CUDA 9 is support for the powerful new Volta Architecture, specifically the new Tesla V100 GPU accelerator which was launched at GTC 2017.

Neural networks and deep learning are big topics in Computer Science and in the technology industry, they currently provide the best solutions to many problems in image recognition, speech recognition and natural language processing. Recently many papers have been published featuring AI that can learn to paint, build 3D Models, create user interfaces(pix2code), some create images given a sentence and there are many more incredible things being done everyday using neural networks. I'm writing this series of posts about Neural Networks and Deep learning, where I'm going to guide you from learning the basic concepts of Artificial Neural Networks (ANN), show you examples from simple Network to mimic the AND gate, to solving Image recognition tasks using Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNN) and more. The code will always be written in python, some times with the help of Tensorflow (I don't expect you to be guru using Tensorflow as I will try to explain the code in details). The definition of a neural network, more properly referred to as an'artificial' neural network (ANN), is provided by the inventor of one of the first neurocomputers, Dr. Robert Hecht-Nielsen.

The practice of Deep Learning is vastly outpacing theory. This is despite the incredible number of Deep Learning papers that are published every day on Arxiv. To develop good theoretical results, researchers have to settle with simplified models that are tractable with our current investigative tools. More advanced models that use the latest state-of-the-art techniques are at a level of complexity that are beyond our current mathematical toolbox to understand. The practice therefore of Deep Learning, despite all the heavy math that is employed, is actually more like alchemy than that of chemistry.

Identifying 3 moss species by deep learning, using the "chopped picture" method Graduate School of Agriculture, Kyoto University, Japan * corresponding author: ise@kais.kyoto-u.ac.jp Abstract In general, object identification tends not to work well on ambiguous, amorphous objects such as vegetation. In this study, we developed a simple but effective approach to identify ambiguous objects and applied the method to several moss species. As a result, the model correctly classified test images with accuracy more than 90%. Using this approach will help progress in computer vision studies. Introduction Especially in recent years, deep learning has become a very effective tool for object identification (Krizhevsky et al. 2012, Szegedy et al. 2015).

Convolutional Neural Networks (CNNs) have achieved state-of-the-art performance on a variety of computer vision tasks, particularly visual classification problems, where new algorithms reported to achieve or even surpass the human performance. In this paper, we examine whether CNNs are capable of learning the semantics of training data. To this end, we evaluate CNNs on negative images, since they share the same structure and semantics as regular images and humans can classify them correctly. Our experimental results indicate that when training on regular images and testing on negative images, the model accuracy is significantly lower than when it is tested on regular images. This leads us to the conjecture that current training methods do not effectively train models to generalize the concepts. We then introduce the notion of semantic adversarial examples - transformed inputs that semantically represent the same objects, but the model does not classify them correctly - and present negative images as one class of such inputs.