Apple is building a car

The Guardian revealed last week that Apple is looking for a secure vehicle test track facility near SiliconValley. Therefore confirming the rumours that Apple is working on building a self-driving car. However, an article in The Verge warns Apple fans not to hold their breath. The development cycl e time from concept to production, for established car manufacturers, is in excess of 5 years. Apple has no experience in this market and consequently may be expected to take longer. Of course this may be like Apple Maps where Apple belatedly realised they strategically couldnt gift location support to Google and their established mapping product. Google, of course, has been developing driverless cars for ages. Apple has perhaps realised it cant leave this market segment to Google or risk just partnering with one car manufacturer.

from The Universal Machine http://universal-machine.blogspot.com/

Put the internet to work for you.

Delete or edit this Recipe

Recommended for you

Read More..

Building A Visual Planetary Time Machine

Posted by Randy Sargent, Google/Carnegie Mellon University; Matt Hancher and Eric Nguyen, Google; and Illah Nourbakhsh, Carnegie Mellon University

When a societal or scientific issue is highly contested, visual evidence can cut to the core of the debate in a way that words alone cannot — communicating complicated ideas that can be understood by experts and non-experts alike. After all, it took the invention of the optical telescope to overturn the idea that the heavens revolved around the earth.

Last month, Google announced a zoomable and explorable time-lapse view of our planet. This time-lapse Earth enables you explore the last 29 years of our planet’s history — from the global scale to the local scale, all across the planet. We hope this new visual dataset will ground debates, encourage discovery, and shift perspectives about some of today’s pressing global issues.

This project is a collaboration between Google’s Earth Engine team, Carnegie Mellon University’s CREATE Lab, and TIME Magazine — using nearly a petabyte of historical record from USGS’s and NASA’s Landsat satellites. And in this post, we’d like to give a little insight into the process required to build this time-lapse view of our planet.

Previews of the phenomena visible in these time-lapses.

First well describe Google’s Earth Engine system for deriving the time-series imagery. Second, well tell you more about CMU’s open-source “Time Machine” software for creating and streaming large, explorable time-series imagery.

Annual Composites: Distilling a Massive Dataset

Google Earth Engine brings together the worlds scientific satellite imagery — over a petabyte of multispectral imagery recording over 40 years of history — and makes it available online with tools that scientists, independent researchers, and nations can use to mine this massive warehouse of data to detect changes, map trends and quantify differences on the Earths surface using Google’s computational infrastructure. Today, the platform is used to monitor the Amazon and estimate forest carbon in Tanzania, among hundreds of other partners developing new uses for the technology.

Using Earth Engine, we first built annual global mosaics at a resolution of 30 meters per pixel for each year from 1984 through 2012. We started with a total of 2,068,467 scenes from the Landsat 4, 5, and 7 satellites, comprising 909 terabytes of data. The Earth’s atmosphere is a constantly-shifting sea of clouds, so in order to assemble a seamless cloud-free view of each year we analyzed all the images available at each location and used a simple cloud model to separate out the clouds from the ground. To help correct for atmospheric and seasonal effects, we used an additional 20TB of data from the MODIS MCD43A4 product to build a cloud-free low-resolution model of the Earth over time. We combined all this to produce a statistical estimate of the color of each pixel for every year for which data was available. Producing the final 29 global mosaics took a bit less than a day and consumed approximately 260,000 core-hours of CPU.

Some areas of the planet are almost perpetually cloudy, obscuring satellite views. In addition, before the more capable Landsat 7 began operating in 1999, coverage in some areas of the world was sparse, particularly in Asia, for various operational and technological reasons. We wrestled with how best to visualize areas with missing or cloud-obscured images from each year. In the end, after much experimentation, we chose to simply interpolate between valid image years. Other techniques, such as greying out invalid data, created distractingly large artifacts, visually drowning out the valid information. However, the downside with the approach we have taken is that it can be difficult to tell which data is original and which is interpolated. We are exploring the possibility of including a view that allows drilling down into the non-interpolated, original mosaics.

"Time Machine": An HTML5 Time-Series Exploration Tool

Once we had produced the final global images, we adapted the Carnegie Mellon CREATE Lab’s open-source “Time Machine” software, which enables authoring, streaming, and exploring very-high-resolution videos. Time Machine videos take advantage of the power of HTML5 and modern web browsers: they are streamed as multiresolution, overlapping video tiles and displayed in a web page by manipulating the HTML5 <video> tag, in much the same way that Google Maps first demonstrated using the HTML <img> tag.

Examples of zoomable timelapses with hundreds of millions or billions of pixels per frame include documenting plant growth, bee colony collapse, and very-large-scale simulations of the universe. Time-lapse Earth, however, sets a new record for giant videos: each frame of the video is a global Mercator-projected map with a resolution of 30 meters per pixel at the equator, for a total of 1.78 trillion pixels per frame. That’s about a million times larger than a standard HD video stream. In order to scale to such large videos, we needed to integrate Time Machine’s data production pipeline into Earth Engine and the rest of Google’s infrastructure. Encoding the final video tiles consumed approximately 1.4 million core-hours of CPU in Google’s data centers over the course of about a day. For CMUs researchers, this would have been impossible without Googles resources.

Combining all three phases of product generation:

• Total processing time: 3 days
• Total CPU usage: 1.8 million core-hours
• Peak CPU usage: 66,000 simultaneous cores

Destination locations of top 1500 share links, weighted by number of visits.

Time-lapse Earth is powerful because it helps us to access and construct the story of our planet. That story will become richer with each release, as we continue to improve fidelity and add data. The story-teller is everyone — scientists and citizens alike provide the real value by interacting, exploring, layering their knowledge upon the globe, and sharing their insights so that we can all better understand our world.

We are especially proud of the collaboration that made time-lapse Earth possible, and believe it to be an exemplar of how industry, academia, government, and the press can benefit from working together deeply over a period of years. By drawing on the strengths of each member of the collaborative community, Google strives to integrate the worlds technical expertise and knowledge in order to tackle innovative and groundbreaking projects. In doing so, it is our goal to deliver an impactful service, one that can put a focus on the dramatic effect we are having on our planet.

Read More..

Text to Speech for low resource languages episode 2 Building a parametric voice

Posted by Alexander Gutkin, Google Speech Team

This is the second episode in the series of posts reporting on the work we are doing to build text-to-speech (TTS) systems for low resource languages. In the previous episode, we described the crowdsourced data collection effort for Project Unison. In this episode, we describe our work to construct a parametric voice based on that data.

In our previous episode, we described building TTS systems for low resource languages, and how one of the objectives of data collection for such systems was to quickly build a database representing multiple speakers. There are two main justifications for this approach. First, professional voice talents are often not available for under-resourced languages, so we need to record ordinary people who get tired reading tedious text rather quickly. Hence, the amount of text a person can record is rather limited and we need multiple speakers for a reasonably sized database that can be used by others as well. Second, we wanted to be able to create a voice that sounds human but is not identifiable as a real person. Various concatenative approaches to speech synthesis, such as unit selection, are not very suitable for this problem. This is because the selection algorithm may join acoustic units from different speakers generating a very unnatural sounding result.

Adopting parametric speech synthesis techniques is an attractive approach to building multi-speaker corpora described above. This is because in parametric synthesis the training stage of the statistical component will take care of multiple-speakers by estimating an averaged out representation of various acoustic parameters representing each individual speaker. Depending on number of speakers in the corpus, their acoustic similarity and ratio of speaker genders, the resulting acoustic model can represent an average voice that is indistinguishable from human and yet cannot be traced back to any actual speakers recorded during the data collection.

We decided to use two different approaches to acoustic modeling in our experiments. The first approach uses Hidden Markov Models (HMMs). This well-established technique was pioneered by Prof. Keiichi Tokuda at Nagoya Institute of Technology, Japan and has been widely adopted in academia and industry. It is also supported by a dedicated open-source HMM synthesis toolkit. The resulting models are small enough to fit on mobile devices.

The second approach relies on Recurrent Neural Networks (RNNs) and vocoders that jointly mimic the human speech production system. Vocoders mimic the vocal apparatus to provide a parametric representation of speech audio that is amenable to statistical mapping. RNNs provide a statistical mapping from the text to the audio and have feedback loops in their topology, allowing them to model temporal dependencies between various phonemes in human speech. In 2015, Yannis Agiomyrgiannakis proposed Vocaine, a vocoder that outperforms the state-of-the-art technology in speed as well as quality. In 2013, Heiga Zen, Andrew Senior and Mike Schuster proposed a neural network-based model that mimics deep structure of human speech production for speech synthesis. The model has further been extended into a Long Short-Term Memory (LSTM) RNN. This allows long term memorization, which is good for speech applications. Earlier this year, Heiga Zen and Hasim Sak described the LSTM RNN architecture that has been specifically designed for fast speech synthesis. The LSTM RNNs are also used in our Automatic Speech Recognition (ASR) systems recently mentioned in our blog.

Using the Hidden Markov Model (HMM) and LSTM RNN synthesizers described above, we experimented with a multi-speaker Bangla corpus totaling 1526 utterances (waveforms and corresponding transcriptions) from five different speakers. We also built a third system that utilizes LSTM RNN acoustic model, but this time we made it small and fast enough to run on a mobile phone.

We synthesized the following Bangla sentence "??? ???? ????? ??????? ??????" translated from “This is an example sentence in Bangla”. Though HMM synthesizer output can sound intelligible, it does exhibit some classic downsides with a voice that sounds buzzy and muffled. With the LSTM RNN configuration for mobile devices, the resulting audio sounds clearer and has improved intonation over the HMM version. We also tried a LSTM RNN configuration with more network nodes (and thus not suitable for low-end mobile devices) to generate this waveform - the quality is slightly better but is not a huge improvement over the more lightweight LSTM RNN version. We hypothesize that this is due to the fact that a neural network with many nodes has more parameters and thus requires more data to train.

These early results are encouraging for several reasons. First, they confirm that natural-sounding speech synthesis based on multiple speakers is practically possible. It is also significant that the total number of recordings used was relatively small, yet were able to build intelligible parametric speech synthesis. This means that it is possible to collect training data for such a speech synthesizer by engaging the help of volunteers who are not professional voice artists, for a short period of time per person. Using multiple volunteers is an advantage: it results in more diverse data, and the resulting synthetic voice does not represent any specific individual. This approach may well be the foundation for bringing speech technology to many more traditionally under-served languages.

NEXT UP: But can it say, “Google”? (Ep.3)

Read More..

Building a deeper understanding of images

Posted by Christian Szegedy, Software Engineer

The ImageNet large-scale visual recognition challenge (ILSVRC) is the largest academic challenge in computer vision, held annually to test state-of-the-art technology in image understanding, both in the sense of recognizing objects in images and locating where they are. Participants in the competition include leading academic institutions and industry labs. In 2012 it was won by DNNResearch using the convolutional neural network approach described in the now-seminal paper by Krizhevsky et al.^[4]

In this year’s challenge, team GoogLeNet (named in homage to LeNet, Yann LeCuns influential convolutional network) placed first in the classification and detection (with extra training data) tasks, doubling the quality on both tasks over last years results. The team participated with an open submission, meaning that the exact details of its approach are shared with the wider computer vision community to foster collaboration and accelerate progress in the field.

The competition has three tracks: classification, classification with localization, and detection. The classification track measures an algorithm’s ability to assign correct labels to an image. The classification with localization track is designed to assess how well an algorithm models both the labels of an image and the location of the underlying objects. Finally, the detection challenge is similar, but uses much stricter evaluation criteria. As an additional difficulty, this challenge includes a lot of images with tiny objects which are hard to recognize. Superior performance in the detection challenge requires pushing beyond annotating an image with a “bag of labels” -- a model must be able to describe a complex scene by accurately locating and identifying many objects in it. As examples, the images in this post are actual top-scoring inferences of the GoogleNet detection model on the validation set of the detection challenge.

This work was a concerted effort by Wei Liu, Yangqing Jia, Pierre Sermanet, Scott Reed, Drago Anguelov, Dumitru Erhan, Andrew Rabinovich and myself. Two of the team members -- Wei Liu and Scott Reed -- are PhD students who are a part of the intern program here at Google, and actively participated in the work leading to the submissions. Without their dedication the team could not have won the detection challenge.

This effort was accomplished by using the DistBelief infrastructure, which makes it possible to train neural networks in a distributed manner and rapidly iterate. At the core of the approach is a radically redesigned convolutional network architecture. Its seemingly complex structure (typical incarnations of which consist of over 100 layers with a maximum depth of over 20 parameter layers), is based on two insights: the Hebbian principle and scale invariance. As the consequence of a careful balancing act, the depth and width of the network are both increased significantly at the cost of a modest growth in evaluation time. The resultant architecture leads to over 10x reduction in the number of parameters compared to most state of the art vision networks. This reduces overfitting during training and allows our system to perform inference with low memory footprint.

For the detection challenge, the improved neural network model is used in the sophisticated R-CNN detector by Ross Girshick et al.^[2], with additional proposals coming from the multibox method^[1]. For the classification challenge entry, several ideas from the work of Andrew Howard^[3] were incorporated and extended, specifically as they relate to image sampling during training and evaluation. The systems were evaluated both stand-alone and as ensembles (averaging the outputs of up to seven models) and their results were submitted as separate entries for transparency and comparison.

These technological advances will enable even better image understanding on our side and the progress is directly transferable to Google products such as photo search, image search, YouTube, self-driving cars, and any place where it is useful to understand what is in an image as well as where things are.

References:

[1] Erhan D., Szegedy C., Toshev, A., and Anguelov, D., "Scalable Object Detection using Deep Neural Networks", The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2014, pp. 2147-2154.

[2] Girshick, R., Donahue, J., Darrell, T., & Malik, J., "Rich feature hierarchies for accurate object detection and semantic segmentation", arXiv preprint arXiv:1311.2524, 2013.

[3] Howard, A. G., "Some Improvements on Deep Convolutional Neural Network Based Image Classification", arXiv preprint arXiv:1312.5402, 2013.

[4] Krizhevsky, A., Sutskever I., and Hinton, G., "Imagenet classification with deep convolutional neural networks", Advances in neural information processing systems, 2012.

Read More..

A picture is worth a thousand coherent words building a natural description of images

Posted by Google Research Scientists Oriol Vinyals, Alexander Toshev, Samy Bengio, and Dumitru Erhan

“Two pizzas sitting on top of a stove top oven”
“A group of people shopping at an outdoor market”
“Best seats in the house”

People can summarize a complex scene in a few words without thinking twice. It’s much more difficult for computers. But we’ve just gotten a bit closer -- we’ve developed a machine-learning system that can automatically produce captions (like the three above) to accurately describe images the first time it sees them. This kind of system could eventually help visually impaired people understand pictures, provide alternate text for images in parts of the world where mobile connections are slow, and make it easier for everyone to search on Google for images.

Recent research has greatly improved object detection, classification, and labeling. But accurately describing a complex scene requires a deeper representation of what’s going on in the scene, capturing how the various objects relate to one another and translating it all into natural-sounding language.

Automatically captioned: “Two pizzas sitting on top of a stove top oven”

Many efforts to construct computer-generated natural descriptions of images propose combining current state-of-the-art techniques in both computer vision and natural language processing to form a complete image description approach. But what if we instead merged recent computer vision and language models into a single jointly trained system, taking an image and directly producing a human readable sequence of words to describe it?

This idea comes from recent advances in machine translation between languages, where a Recurrent Neural Network (RNN) transforms, say, a French sentence into a vector representation, and a second RNN uses that vector representation to generate a target sentence in German.

Now, what if we replaced that first RNN and its input words with a deep Convolutional Neural Network (CNN) trained to classify objects in images? Normally, the CNN’s last layer is used in a final Softmax among known classes of objects, assigning a probability that each object might be in the image. But if we remove that final layer, we can instead feed the CNN’s rich encoding of the image into a RNN designed to produce phrases. We can then train the whole system directly on images and their captions, so it maximizes the likelihood that descriptions it produces best match the training descriptions for each image.

The model combines a vision CNN with a language-generating RNN so it can take in an image and generate a fitting natural-language caption.

Our experiments with this system on several openly published datasets, including Pascal, Flickr8k, Flickr30k and SBU, show how robust the qualitative results are -- the generated sentences are quite reasonable. It also performs well in quantitative evaluations with the Bilingual Evaluation Understudy (BLEU), a metric used in machine translation to evaluate the quality of generated sentences.

A selection of evaluation results, grouped by human rating.

A picture may be worth a thousand words, but sometimes it’s the words that are most useful -- so it’s important we figure out ways to translate from images to words automatically and accurately. As the datasets suited to learning image descriptions grow and mature, so will the performance of end-to-end approaches like this. We look forward to continuing developments in systems that can read images and generate good natural-language descriptions. To get more details about the framework used to generate descriptions from images, as well as the model evaluation, read the full paper here.

Read More..