Google Brain Team Research Articles

Style Migration Based on the Loss Function of Location Information

Using the improved Johnson et al.’s style migration network as a starting point, this paper proposes a new loss function based on the position information Gram matrix. The new method adds the chunked Gram matrix with position information, and simultaneously, the structural similarity between the style map and the resultant image is added to the style training. The style position information is given to the resultant image, and finally, the resolution of the resultant image is improved with the SRGAN. The new model can effectively migrate the texture structure as well as the color space of the style image, while the data of the content image are kept intact. The simulation results reveal that the image processing results of the new model improve those of the classical Johnson et al.’s method, Google Brain team method, and CCPL method, and the SSIM values of the resulting map and style image are all greater than 0.3. As a comparison, the SSIM values of Johnson et al., Google Brain team, and CCPL are 0.14, 0.11, and 0.12, respectively, which is an obvious improvement. Moreover, with deeper training, the new method can improve the similarity of certain resulting images and style images to more than 0.37256. In addition, training other arbitrary content images on the basis of the trained model can quickly yield satisfactory results.

DESCNet: Developing Efficient Scratchpad Memories for Capsule Network Hardware

Deep Neural Networks (DNNs) have been established as the state-of-the-art algorithm for advanced machine learning applications. Recently proposed by the Google Brain's team, the Capsule Networks (CapsNets) have improved the generalization ability, as compared to DNNs, due to their multi-dimensional capsules and preserving the spatial relationship between different objects. However, they pose significantly high computational and memory requirements, making their energy-efficient inference a challenging task. This paper provides, for the first time, an in-depth analysis to highlight the design and management related challenges for the (on-chip) memories deployed in hardware accelerators executing fast CapsNets inference. To enable an efficient design, we propose an application-specific memory hierarchy, which minimizes the off-chip memory accesses, while efficiently feeding the data to the hardware accelerator. We analyze the corresponding on-chip memory requirements and leverage it to propose a novel methodology to explore different scratchpad memory designs and their energy/area trade-offs. Afterwards, an application-specific power-gating technique is proposed to further reduce the energy consumption, depending upon the utilization across different operations of the CapsNets. Our results for a selected Pareto-optimal solution demonstrate no performance loss and an energy reduction of 79% for the complete accelerator, including computational units and memories, when compared to a state-of-the-art design executing Google's CapsNet model for the MNIST dataset.

A Systematic Review About Use of TensorFlow for Image Classification and Word Embedding in the Brazilian Context

A great deal of data are available on the Internet, and it is possible to extract any type of implicit knowledge using Artificial Intelligence (AI) tools to support decision making. The open-source TensorFlow framework, developed by the Google Brain Team in 2015, is, presently, the most used tool for several AI applications, such as image classification, word embedding, and chatbot development. This paper presents results of a systematic review of the use of the TensorFlow framework for image classification and word embedding applications written in Portuguese language and in the Brazilian context. We used Google Scholar as Academic Search Engine and 90 were retrieved initially. However, just 12 were remained for reading and obtaining of the main information. Title, publication year, used domain, type of application and covered scope were collected from papers retrieved to accelerate studies in the AI area and to disseminate the potential of this framework for emerging challenges.

Fire induced progressive collapse potential assessment of steel framed buildings using machine learning

In this paper, a new Machine Learning framework is developed for fast prediction of the failure patterns of simple steel framed buildings in fire and subsequent progressive collapse potential assessment. This pilot study provides a new tool of fire safety assessment for engineers in an efficient and effective way in the future. The concept of Critical Temperature Method is used to define the failure patterns for each structural member which is incorporated into a systematic methodology employing both Monte Carlo Simulation and Random Sampling to generate a robust and sufficient large dataset for training and testing, hence guarantees the accurate prediction. A comparative study for different machine learning classifiers is made. Three classifiers are chosen for failure patterns prediction of buildings under fire: Decision Tree, KNN and Neural Network using Google Keras with TensorFlow which is specially used for Google Brain Team. The Machine Learning framework is implemented using codes programmed by the author in VBA and Python language. A case study of a 2 story by 2 bay steel framed building was made. Two different fire scenarios were chosen. The procedure gives satisfactory prediction of the failure pattern and collapse potential of the building under fire.

What are the limits of deep learning?

The much-ballyhooed artificial intelligence approach boasts impressive feats but still falls short of human brainpower. Researchers are determined to figure out what’s missing. There’s no mistaking the image: It’s a banana—a big, ripe, bright-yellow banana. Yet the artificial intelligence (AI) identifies it as a toaster, even though it was trained with the same powerful and oft-publicized deep-learning techniques that have produced a white-hot revolution in driverless cars, speech understanding, and a multitude of other AI applications. That means the AI was shown several thousand photos of bananas, slugs, snails, and similar-looking objects, like so many flash cards, and then drilled on the answers until it had the classification down cold. And yet this advanced system was quite easily confused—all it took was a little day-glow sticker, digitally pasted in one corner of the image. Apparent shortcomings in deep-learning approaches have raised concerns among researchers and the general public as technologies such as driverless cars, which use deep-learning techniques to navigate, get involved in well-publicized mishaps. Image credit: Shutterstock.com/MONOPOLY919. This example of what deep-learning researchers call an “adversarial attack,” discovered by the Google Brain team in Mountain View, CA (1), highlights just how far AI still has to go before it remotely approaches human capabilities. “I initially thought that adversarial examples were just an annoyance,” says Geoffrey Hinton, a computer scientist at the University of Toronto and one of the pioneers of deep learning. “But I now think they're probably quite profound. They tell us that we're doing something wrong.” That’s a widely shared sentiment among AI practitioners, any of whom can easily rattle off a long list of deep learning’s drawbacks. In addition to its vulnerability to spoofing, for example, there is its gross inefficiency. “For a child to learn to recognize a cow,” says Hinton, “it's not like their …

TensorFlow: learning functions at scale

TensorFlow is a machine learning system that operates at large scale and in heterogeneous environments. Its computational model is based on dataflow graphs with mutable state. Graph nodes may be mapped to different machines in a cluster, and within each machine to CPUs, GPUs, and other devices. TensorFlow supports a variety of applications, but it particularly targets training and inference with deep neural networks. It serves as a platform for research and for deploying machine learning systems across many areas, such as speech recognition, computer vision, robotics, information retrieval, and natural language processing. In this talk, we describe TensorFlow and outline some of its applications. We also discuss the question of what TensorFlow and deep learning may have to do with functional programming. Although TensorFlow is not purely functional, many of its uses are concerned with optimizing functions (during training), then with applying those functions (during inference). These functions are defined as compositions of simple primitives (as is common in functional programming), with internal data representations that are learned rather than manually designed. TensorFlow is joint work with many other people in the Google Brain team and elsewhere. More information is available at tensorflow.org.