Rapid Advances in Computer Science and Opportunities for Society - - PowerPoint PPT Presentation

Rapid Advances in Computer Science and Opportunities for Society

European CS Presentation, October 2010

Alfred Spector VP, Research and Special Initiatives

Rapid Advances in Computer Science & Opportunities for Society

Information and Communication Technologies have had a rapid impact on society, and –amazingly—the pace of innovation continues to

• accelerate. This innovation is catalyzed by ever-increasing hardware and

networking capabilities, the growth in internet usage, as well as important advances in basic and applied computer science. In this talk, I will describe some of the research that Google is undertaking (for example, in machine translation, semantic processing, and information management), and discuss some of the likely beneficial impacts on our society – for example, in science, the humanities, education, philanthropic activities, and more. I’ll conclude my presentation with some interesting challenges from both a technology and policy point of view.

Abstract

Outline

Google Prodigiousness Advances in the Field: examples Translation Speech Vision Cloud-based collaboration around structured-data Operations Research Semantic Processing Beneficial Societal Impacts: examples Earth Engine Google Health Other Health Efforts Crisis Response Digital Humanities Education A Technical Themes Challenges

Mission

Organizing the world’s information and Making it universally accessible and useful.

Google and Commerce

Over 1 million AdWords advertisers worldwide Over 1 million AdSense publishers worldwide Via the Google Ad Network, AdSense publishers reach over 80% of global internet users in 100 countries and 20 languages YouTube is monetizing over a billion video views per week globally In 2009, Google generated $54 billion of economic activity for American businesses, website publishers, and non-profits

Prodigiousness

Giga 109, Tera 1012, Peta 1015, Exa 1018, Zetta1021 Publicized: Bigtable of 70 petabytes, 10M ops/sec. Warehouse computing possibilities? 100 x 10 x 20 x 20 x 40 = 16,000,000 nodes… Some representative numbers: Storage: 1018 -> 1020-21 Users: 109 -> 1010 Devices: 10? -> 1012 Network: 1020, now, ->1021/yr 32 KB/sec. for 1B people Apps: 105 -> 106-7 or more E.g., embedded car systems: 30-50 ECUs, 100M lines of code

A variety of science engineering challenges

Focus on Innovation that Benefits our Users Focus on Research and Engineering

Commitment to advancing technology Rich domain of work due to our mission Grand challenge problems Internal consensus that production issues are

• ften as challenging/fun as pure invention

Technical leverage

• 1. Google Common Distributed System
• 2. A Focus on Services
• 3. Empiricism and a Holistic Approach to Design

Our Innovation Culture

Focus on talent Distributed across the organization Impacting Google necessitates broad, diverse involvement in science and engineering Research is done both in our research team and in

• ur engineering organization, organized
• pportunistically

Teams benefit greatly From mutual talent From Google’s comparative advantages to our scale and broad use From service-based architecture (“ease” of working in vivo)

Ideal Distributed Computing

Device s

Research Challenges in Ideal Distributed Computing

Alternative designs that would give better energy efficiency at lower utilization Server O.S. design aimed at many highly-connected machines in

• ne building

Unifying abstractions for exploiting parallelism beyond inter- transaction parallelism and map-reduce Latency reduction A general model of replication including consistency choices, explained and codified Machine learning techniques applied to monitoring/controlling such systems Automatic, dynamic world-wide placement of data & computation to minimize latency and/or cost, given constraints on Building retrieval systems that efficiently and usably deal with ACLs Holistic models of privacy The user interface to the user’s diverse processing and state

Totally Transparent Processing

D: The set of all end-user access devices L: The set of all human languages M: The set of all modalities C: The set of all corpora

Personal Computers Phone Media Players/Readers Telematics Set-top Boxes Appliances Health devices … Current languages Historical languages Other forms of human notation Possible language specialization Formal languages … Text Image Audio Video Graphics Other sensor-based data … The normal web The deep web Periodicals Books Catalogs Blogs Geodata Scientific datasets Health data …

For all d in D, all l in L, all m in M, and all c in C

Totally Transparent Processing

“Hybrid” Intelligence

To extend the capability of people, not in isolation Aggregation of empirical signal is exceedingly valuable Ex: Feedback in Information Retrieval; e.g., in ranking or spelling correction Machine learning; e.g., image content analysis, speech recognition with semi-supervised learning

Research Challenges in Transparent Computing & Hybrid Intelligence

Endless applications, with very new user interface implications Addressing limits to data Techniques to integrate user-feedback in acceptable fashions Approaches to new signal Explanation, scale, and variance minimization in machine learning Information fusion/learning across diverse signals – The Combination Hypothesis, more generally Usability: devices and subpopulations Privacy

Domains of Application

Search engines Translation Speech recognition Vision Remedial Education Personal health Epidemiology Economic prediction Societal/environmental optimization Social Networking in ever more clever/useful ways Humanities and Social Sciences Multi-player gaming

Translation

Machine Translation @ Google

Statistical Machine Translation Model translation process with a statistical model Learning from data: monolingual & bilingual More data: better translation quality Computationally expensive approach Models have many hundreds of Gigabyte of data (Moore's law helps here) Applying syntax information as a signal Results: Much better translation quality Ongoing progress More research groups, ... 58 languages (so far) recently: Haitian Creole, Urdu, Georgian, ..., Latin

Grand Challenges

Morphology: translating into morphologically rich languages e.g. Russian, Hungarian need: morphology-aware translation models Reliability: some translation mistakes more severe than others:

hotel - Montreal Heath Ledger - Tom Cruise

Research: How to detect 'crazy' translations? Long-distance reordering: 'simple' case: SVO, SOV (one) approach: parse source & reorder issue: parsing accuracy for

• ut-of-domain texts

Finding all Training Data

How about Poetry?

Paper at EMNLP 2010 conference: “Poetic” Statistical Machine Translation: Rhyme and Meter, D.Genzel, J.Uszkoreit, F.Och, EMNLP, 2010. Approach: Enforce meter and rhyme as extra constraints (similar to language model) E.g. iambic pentameter: stress pattern 0101010101 Produce most 'probable' translation that obeys constraints ("Function follows form") Example output (couplet in amphibrachic tetrameter An officer stated that three were arrested and that the equipment is currently tested.

Speech

Goals for Speech Technology at Google

Much of the world’s information is spoken – we need to recognize it before we can organize it:

YouTube transcription and translation (breaking the language barrier for YouTube access) Voicemail transcription

Mobile is the fastest growing and most widespread platform for communication and services that has ever existed

Spoken input and output is key to usability Our goal is completely ubiquitous availability of speech i/o (every application/service, every usage scenario, every language)

How do we get there?

Delivery from the cloud – support constant iteration and refinement Operating at large scale – train huge statistical models on huge amounts of data

Learning from use - without human transcription Challenges:

How do we grow the model to take advantage of the data? (richer models of accent, speaker, noise, etc.) Huge computational demands Infrastructure demands – parallelization – leverage Google software environment

Training Acoustic Models w/Unsupervised Learning

Supervised vs. unsupervised training - hours of data vs. error rate

Vision

Computer Vision

Advance state-of-the art in 3 key areas of image/audio/video analysis and apply results to our multimedia products. Semantic Interpretation: Generate human understandable description of content. (eg. auto-tagging videos on YouTube, Image annotation, porn classification etc.) Matching: Find similar entities from a large corpus. (eg. "find similar" on image search, video fingerprinting for YouTube etc ). Synthesis: Generate better images/video by understanding the statistics of a large corpus of images. (eg. better facades in 3D building on Google Earth, automatic shadow removal from areal images etc.)

Semantic Interpretation sample problem - Video Annotation

Video metadata has a cognitive cost on the user because they have to type it in, be careful about what keywords they use, and in general try to make their video searchable Many uploaders don’t have the motivation, or energy to provide proper metadata Noisy metadata hurts everyone – spam, misspellings, 1337, acronyms, etc.

Cloud-based Computing Structured Data

Structured Data on the Web

Discovery and search for structured data: The deep Web -- significant gap in coverage Structured tables on the Web -- not leveraged in search Enable easy creation, management, sharing and publishing of structured data: Fusion Tables: www.google.com/fusiontables

Google Fusion Tables

host, manage, collaborate on, visualize, and publish data tables online

What can I do with Fusion Tables? Host data online - and stay in control control can be at the level of columns or rows Re-use data without making copies Collaborate on the details Merge data from multiple tables Comment on individual rows, columns or cells Make a map (or chart or timeline) in minutes! Manage data via our site or an API

Fusion Table Example Gallery

Easy Data Upload, Attribution recorded

Easily Create Informative Maps

Easily Create Informative Maps

baby steps towards the dream platform DEMO:

circle of blue

Cloud-based Computing Prediction

• 1. Upload
• 2. Train

Upload your training data to Google Storage Build a model from your data Make new predictions

• 3. Predict

Machine learning as a web service

Smart Apps for every developer

• RESTful HTTP service
• Simple integration

Prediction API

Under the hood, many classifiers/regressors. Recent research, efficient and theoretically principled methods for distributed learning (NIPS-09, HLT-10):

Network costs can be reduced by an order of magnitude with minimal loss in classifier accuracy.

Under the API

Operations Reseach and Optimization

Operations Research Challenges

Size: Optimization is often NP Complete Increasing the size by 1 doubles the search space. The tools are barely keeping up with the problems. Uncertainty: Data is often fuzzy. How do you route cars when there are roadblocks, new

• ne-ways, traffic jams?

Can you use optimization in on-line algorithm connected to users? How well can you optimize against forecasted data, how do you react if the forecast is bad? User expectation and requirements: The definition of problems is also unclear. What is the

• bjective? What is a good solution? Can I violate this

requirement? By how much?

Operations Research Opportunities

Machine Learning can help us in two ways: By providing guidance towards good solutions. By qualifying valid solutions. By reducing the search space. Large computing resources means we can try a bit harder. Crowd-Sourcing means better data, better feedback, better evaluations of algorithms and solutions. Having all our code open-source means we can collaborate on building the best set of tools. See http://code.google.com/p/or-tools

Semantic Processing

Web Inference and Learning

Goal: better understanding of Web content and user intent Method: algorithms that draw reliable semantic inferences from the wealth of evidence implicit in massive Web data

How to interpret this term in this context? Does this sentence answer that question? Will this user click on that ad?

Learning: create concise representations to support good inferences

Meaning from the Web

Elementary semantic inference: what are the possible classes for each instance?

Combination wins

Combined graph: 1.4M nodes, 75M edges

Applications to Society Follow

Earth Engine

Motivation - Carbon Forest Tracking

UNEP: "Atlas of our Changing Environment"

1975 1989 2001

Rondonia, Brazil

A Sampling of Other Use Cases

Disease Early Warning: Remote surveillance of

disease and prediction of epidemics.

Population Census: Supplements traditional census

and mapping in developing regions.

Humanitarian Crisis Mapping: Can detect and

monitor a growing range of crisis types.

Water Resources: Monitor water quality and

availability and alleviate water shortage problems.

Food Security: Famine early warning, rainfall and

water requirements estimations, agr production estimates and irrigation and fertilizer supply & demand.

Global Education Programs

Parallel Geo-Processing “in the cloud” (a brief illustration)

Original image

Original image ... is divided into 256px sub-units.

Sub-units are distributed

Sub-units are distributed ... to separate machines.

Sub-units are distributed ... to separate machines ... where they can be processed in parallel.

Thousands can be processed simultaneously

Result is reassembled

Result is reassembled ... into a finished image

Global-scale earth observation and informatics platform

For public benefit, and to support emerging green economy Help science come out of research lab and into operational use, at scale Unprecedented catalog of earth observation data for mining and analysis Promote transparency, reproducibility, collaboration, “open science”

Very fast computation of scientific map products

Intrinsically-parallel pixel processing system Built-in Google algorithms as well as user-supplied Earth Engine API for 3rd party algorithm development Access control, versioning, provenance Online and desktop versions (open source desktop version)

On a lot of useful data

Every available Landsat and MODIS scene (more satellites coming) Commercial datasets (very high resolution satellite imagery) Environmental data (atmospheric, ocean, terrestrial) User-supplied (ex: in-situ data collected via Android phones)

Overview

Scale of Data

US Satellite imagery dataset: Landsat

Peru: 60 Landsat scenes (3Gpix, 20GB) per covering World: 8000 Landsat scenes (2TB) per covering Complete global coverage every 16 days Operating since 1972, historical archive holds ~4PB US, NASA EOS approaching ~10PB of Earth images

Europe: European Space Agency (ESA)

ESA satellite missions: MERIS, Envisat, others SpotImage (France): 20M SPOT images since 1986, 10,000 new images collected daily, 5+ PB archive ESA Launching Sentinel-1 in 2011

Representative Examples

Envisat : Gulf Oil Spill June 2010 (ESA)) MERIS: Hurricane Isabel Sept 2003 (ESA) Spot Image: Xingu Brazilian Amazon

View on YouTube

Health (US)

Google Health personal health dashboard

Launched in May 2008 Major update Sept 2010 User controls and

• wns his/her data

A platform that…

Provides a dashboard for wellness information & medical records Allows user to connect and interact with a broad group of “add on” services Includes a non- tethered PHR

Crisis Response

Person Finder

Pre-Earthquake - Aug 26, 2009

1 Day After Earthquake - Jan 13, 2010

13 Days After Earthquake - Jan 25, 2010

Digital Humanities

Illuminating the Humanities

Q: What can you do with: 12 million books in

• ver 400 languages

comprised of 5 billion pages and 2 trillion words ...all digitized?

A: Look to the humanities for new questions...

How would you (re)define Victorian literature? What are the differences between the English and Latin editions of Hobbes’ Leviathan? How have places changed over the course of history?

Digital Humanities Awards

Research program supporting university research taking a computational approach to traditional humanist questions. US program, Summer 2010 12 projects 23 researchers 15 universities European program, Winter 2010 10 projects planned

$1M total funding

Education

Curriculum Development

Seeding and supporting computing curriculum development

Exploring computational thinking in K12 (launch late Oct) CS4HS: High school computer science (cs4hs.com) Undergraduate open source CS curriculum: Google Code University (code.google.com.edu) Lantern platform: Wiki for open source curriculum development (in collaboration with Khan Academy)

Talent Development: Google Summer of CodeTM

Program Genesis ”Flip bits not burgers” during summer holidays Exposure to real-world software development Students paired with mentor from OS community Execute to milestones laid out in accepted application Stipend allows students to concentrate on OS development 2010 1026 students 150 organizations 69 countries

Technology Leadership: App Inventor for Android

Visual programming environment for Android mobile devices Helping people become creators (rather than consumers) of technology Launched in Google Labs July 12, 2010 http://appinventor.googlelabs.com/about/

CS4HS European Workshops

École Polytechnique Fédérale de Lausanne (EPFL), Switzerland: 'Building and Programming Robots' ETH Zurich, Switzerland: 'ABZ Ausbildung - und Beratungszentrum fuer Informatikunterricht' Makerere University, Uganda: 'Grassroot approach to improve the quality of applicants of computing programs at Makerere University' Manchester University, United Kingdom: 'Animation11' Oslo University, Norway: 'TENK' Queen Mary University, United Kingdom: 'cs4fn magazine' RWTH Aachen, Germany: 'Bright Brains in Computer Science' Sapienza University of Rome, Italy: 'Challenge and Fun with the CS Olympiads' University of Stuttgart, Germany: 'UniS2010' Technion, Israel: 'High School Computer Science Female Students', Visits in Google: 'Impressions, Conceptions and Influences' Trinity College Dublin, Ireland: 'Computer Programming Outreach, B2C' University of Cape Town, South Africa: 'Project Umonya' University College Dublin, Ireland: 'CS Summer School' University of Warsaw, Poland: 'Mastering Programing Skills, Workshops for Teachers'

Technology Leadership: Google Code University

Course content on current computing technologies and paradigms CS Curriculum Search Tech Talks on CS Topics Tutorials, lecture slides, problem sets for a variety of topic areas: AJAX Programming Algorithms Distributed Systems Web Security Languages Practical Skills (MySQL, Linux) http://code.google.com/edu/

Supporting our Academic Institutions

Research Awards Programs - 230+ projects funded in the last year Next Due dates August 15 (CS Awards), October 15 (Marketing Awards) Research-awards@google.com Focused Grant Program Visiting Faculty Program - 20 faculty (ongoing) University-relations@google.com Ph.D Fellowship Program 2009: 13 students supported in North America 2010: 15 in North America, 16 outside North America Over 150 other scholarships ~1000 interns worldwide CS4HS: 1500+ teachers (~100,000 students): US & EMEA

Final Thoughts

Scale of Communication and Computing is profound Endless opportunity for technical growth Some large themes Major new application domains

Google rapidly to innovate in science/technology and value to consumers We are providing increased support for academic institutions in computer science and related areas It's a most exciting area in which to innovate