Microsoft Research Blog

Artificial intelligence

  1. Deep Roots: Improving CNN Efficiency with Hierarchical Filter Groups 

    June 30, 2017 | Yani Ioannou, Duncan Robertson, Roberto Cipolla, and Antonio Criminisi

    We propose a new method for creating computationally efficient and compact convolutional neural networks (CNNs) using a novel sparse connection structure that resembles a tree root. This allows a significant reduction in computational cost and number of parameters compared to state-of-the-art deep CNNs, without compromising…

  2. SmartPaste: Learning to Adapt Source Code 

    May 21, 2017 | Miltos Allamanis and Marc Brockschmidt

    Deep Neural Networks have been shown to succeed at a range of natural language tasks such as machine translation and text summarization. While tasks on source code (ie, formal languages) have been considered recently, most work in this area does not attempt to capitalize on…

  4. Optimizing CNNs on Multicores for Scalability, Performance and Goodput 

    April 3, 2017

    Convolutional Neural Networks (CNN) are a class of Artificial Neural Networks (ANN) that are highly efficient at the pattern recognition tasks that underlie difficult AI problems in a variety of domains, such as speech recognition, object recognition, and natural language processing. CNNs are, however, computationally…

  5. StyleBank: An Explicit Representation for Neural Image Style Transfer 

    March 26, 2017 | Jing Liao, Nenghai Yu, and Gang Hua

    We propose StyleBank, which is composed of multiple convolution filter banks and each filter bank explicitly represents one style, for neural image style transfer. To transfer an image to a specific style, the corresponding filter bank is operated on top of the intermediate feature embedding…

  6. Deep Roots: Improving CNN Efficiency by Learning a Basis for Filter Dependencies 

    February 28, 2017 | Yani Ioannou, Duncan Robertson, Roberto Cipolla, and Antonio Criminisi

    We propose a new method for creating computationally efficient and compact convolutional neural networks (CNNs) using a novel sparse connection structure that resembles a tree root. This allows a significant reduction in computational cost and number of parameters compared to state-of-the-art deep CNNs, without compromising…

  7. Sparse Bayesian registration of medical images for self-tuning of parameters and spatially adaptive parametrization of displacements. 

    January 31, 2017 | Loic Le Folgoc, Hervé Delingette, Antonio Criminisi, and Nicholas Ayache

    We extend Bayesian models of non-rigid image registration to allow not only for the automatic determination of registration parameters (such as the trade-off between image similarity and regularization functionals), but also for a data-driven, multiscale, spatially adaptive parametrization of deformations. Adaptive parametrizations have been used…

  8. Multi-layer generalized linear estimation 

    January 25, 2017 | Andre Manoel, Florent Krzakala, Marc Mezard, and Lenka Zdeborova

    We consider the problem of reconstructing a signal from multi-layered (possibly) non-linear measurements. Using non-rigorous but standard methods from statistical physics we present the Multi-Layer Approximate Message Passing (ML-AMP) algorithm for computing marginal probabilities of the corresponding estimation problem and derive the associated state evolution…

  9. Towards Better Analysis of Deep Convolutional Neural Networks 

    December 31, 2016

    Deep convolutional neural networks (CNNs) have achieved breakthrough performance in many pattern recognition tasks such as image classification. However, the development of high-quality deep models typically relies on a substantial amount of trial-and-error, as there is still no clear understanding of when and why a…

  10. Towards Better Analysis of Deep Convolutional Neural Networks 

    December 31, 2016

    Deep convolutional neural networks (CNNs) have achieved breakthrough performance in many pattern recognition tasks such as image classification. However, the development of high-quality deep models typically relies on a substantial amount of trial-and-error, as there is still no clear understanding of when and why a…

  11. Natural Language Understanding and Intelligent Applications 

    December 2, 2016

    This book constitutes the joint refereed proceedings of the5th CCF Conference on Natural Language Processing and Chinese Computing, NLPCC 2016, and the 24th International Conference on Computer Processing of Oriental Languages, ICCPOL 2016, held in Kunming, China, in December 2016. The 48 revised full papers…

  12. Helping HPC users specify job memory requirements via machine learning 

    November 12, 2016 | Eduardo Rocha Rodrigues, Renato L. de F. Cunha, Marco A.S. Netto, and Michael Spriggs

    Resource allocation in High Performance Computing (HPC) settings is still not easy for end-users due to the wide variety of application and environment configuration options. Users have difficulties to estimate the number of processors and amount of memory required by their jobs, select the queue…