Google Tech Talks
July 25, 2008
ABSTRACT
Today many workers find themselves caught betwe
Google Tech Talks
July 25, 2008
ABSTRACT
Today many workers find themselves caught between (a) a message centric world of email filled with natural language and (b) the form and report centric world of IT systems filled with structured information. Workers spend many hours navigating interfaces in order to complete tedious update and retrieval tasks to connect these worlds together. For example, an office assistant will be asked via email to book a trip to Paris. First he will retrieve information needed for the task, such as frequent flyer account numbers and approved corporate hotels. Then he will use the e-mail and retrieved information to complete the task. Fortunately, the repetitive nature of many of these tasks makes them ripe for automation using machine learning.
This talk will detail the design and evaluation of VIO, a system that extracts information from email, automating update tasks. VIO monitors emails and makes task suggestions. VIO allows users to quickly identify and repair any suggestion errors, improving the machine learning performance as it learns to automate tasks. The talk will also share more recent work on MIXER, an interface that allows users to automate retrieval tasks triggered by incoming email messages. Once created, MIXER will augment an incoming email with the retrieved information needed to complete the task. Our initial evaluation demonstrates that this user-constructed agent can significantly reduce task completion time, freeing workers from mundane tasks.
Speaker: Anthony Tomasic
Anthony Tomasic is the Director of the Carnegie Mellon University Master of Science in Information Technology, Very Large Information Systems (VLIS) program. His research currently focuses on applying machine learning to the desktop as part of the Carnegie Mellon University RADAR research program. Anthony has also published research on internet scale database systems, federated databases, and the performance of information retrieval systems. He has worked at Stanford University and INRIA (Rocquencourt). He received his Ph.D. in 1994 from Princeton University.
(más)
(menos)
Añadido: hace 10 horas
Reproducciones: 402
Google Tech Talks
July 23, 2008
ABSTRACT
Every material has a distinctive spectrum. The
Google Tech Talks
July 23, 2008
ABSTRACT
Every material has a distinctive spectrum. The spectrum of a material tells us about its chemistry. Hyperspectral images produce a spectrum (represented as several hundred numbers) at each pixel in an image. So hyperspectral images enable us to map variations in chemistry.
The first hyperspectral scanners, built in the 1980's and 1990's, were designed for airborne applications, primarily for mineral, environmental and military applications. However, in recent years, hyperspectral microscopes and cameras have been developed and are being used for terrestrial applications in areas such as medical diagnosis, burns analysis and skin cancer, biosecurity, pharmaceuticals, forensics and in agribusiness.
A significant issue in hyperspectral imaging is that the spectra at many pixels in an image are actually mixtures of the spectra of the pure ingredients. My main focus over a number of years has been on developing fast and sophisticated algorithms and software for "unmixing" these spectra into their pure ingredients, both when the pure ingredients are known and when they are unknown. This has resulted in two software packages:
The Spectral Assistant (TSA), which has been incorporated into another CSIRO package, The Spectral Geologist, which itself has been sold to over 100 (mainly exploration and mining) companies around the world; and Iterated Constrained Endmembers (ICE), which has yet to be commercialized.
I will give an overview of the algorithms underlying TSA and ICE, and demonstrate their application to some mineral, remotely sensing and biological data sets. Finally, I will discuss some unsolved statistical and computational problems associated with these packages.
Speaker: Mark Berman
Mark Berman received the B.Sc.(Hons.) degree and University Medal in mathematical statistics from the University of New South Wales in 1974, and the Master of Statistics degree from the same institution in 1976. In 1978, he was awarded the Ph.D. and D.I.C. degrees in mathematical statistics by the Imperial College of Science and Technology, London.
He was a visiting lecturer in the Department of Statistics at the University of California, Berkeley during 1978-1979. Most of his time since then has been with the CSIRO Division of Mathematical and Information Sciences (CMIS), Sydney, where he is now a Chief Research Scientist. He led CMIS' Image Analysis Group from 1989 to 2000. He spent 1988 at the Melbourne Research Laboratories of Broken Hill Proprietary Ltd. where he established the Image Processing and Data Analysis Group. His research interests are in image analysis (especially hyperspectral), spectroscopy and spatial data analysis.
Since 2007, Dr. Berman has been working part time at CMIS. During this period, he has also given Ph.D courses in spectroscopy and hyperspectral image analysis at the Technical University of Denmark and Stanford University.
(más)
(menos)
Añadido: hace 10 horas
Reproducciones: 241
Google Tech Talks
July, 21 2008
ABSTRACT
Cryptography is now much more than keeping cre
Google Tech Talks
July, 21 2008
ABSTRACT
Cryptography is now much more than keeping credit card numbers safe from packet sniffers and laptop thieves. We combine several advances in cryptography theory to construct a model of computation in which a third party securely performs computations for a set of parties who do not trust one another. These parties encrypt their inputs to a computation, then circulate the encrypted values. Our third party decrypts them, performs the computation and then issues correctness proofs of the results: we cryptographically "tie its hands" to do the right thing. In addition to this model of provably correct computation, we will discuss other techniques to control the information flow of secret data to and from the party, so even the third party cannot profitably abuse the secret data before or after it knows it during the computation. Finally, we will illustrate the power of these techniques in e-commerce, and our design decisions, through secure electronic auctions and securities exchanges.
Speaker: Dc. Christopher Thorpe
Dr. Christopher Thorpe is a computer scientist and entrepreneur who recently completed a Ph.D. in computer science from Harvard, advised by Michael Rabin and David Parkes. Prior to obtaining his Ph.D., he led internationalization at Tellme Networks, after being the internationalization engineer for Yahoo!'s commerce properties and software engineer for Yahoo! Merchant Solutions. Chris joined Yahoo in 1998 when it acquired Viaweb as its Yahoo! Store property; Viaweb was founded by graduate students he knew as an undergraduate at Harvard. He also holds an A.B. from Harvard in Computer Science and Music.
(más)
(menos)
Añadido: hace 3 días
Reproducciones: 1772
|
Google Tech Talks
July, 16 2008
ABSTRACT
Many of the applications you develop are applic
Google Tech Talks
July, 16 2008
ABSTRACT
Many of the applications you develop are applications you would use. This makes it easy to know what will work and what won't. At some point, however, you'll find yourself developing something that you would only occasionally use, and suddenly you're treading in dark places. You know user research is important, you know the experience of using the product should be positive, if not delightful. But sometimes the findings you get are pretty difficult to translate into a decision about the software.
Mental models are diagrams that represent the underlying philosophies and emotions that drive people's behavior, matched up with the ways you think you can support them with your software. Rather than knowing "I like to go to movies alone," you'll learn the myriad reasons why. (E.g. "I like to give the director the attention and respect he deserves, because when I wrote a play in college, people didn't pay attention very well, they didn't get the point, and I felt frustrated.") Knowing the motivating philosophy opens up different avenues for supporting the behavior. You could, for example, offer additional means for this type of moviegoer to "get the point" of the movie. Mental models are useful as structures for attaching these ideas to sets of philosophies and for generating new ideas in places where there are gaps.
In this presentation, author Indi Young will introduce you to mental models and show you one that was developed at Google for the Analytics product. Indi will show you how to use the mental model to expand your perspective and create applications that reach beyond the basic requirements.
Speaker: Indi Young
Indi's work spans a number of decades, from the mid-80's when the desktop metaphor was replacing command line and menu-based systems, to the mid-90's when the Web first toddled onto the scene, to now, when designers are intent on crafting good experiences. After 10 years of consulting, Indi helped found Adaptive Path with six other partners, all hoping to spread good design around the world, making things easier for people everywhere. Indi's mental models have helped both start-ups and large corporations discover and support customer behaviors they didn't think to explore at first. She has written a book about the mental model method, Mental Models - Aligning design strategy with human behavior, published by Rosenfeld Media.
(más)
(menos)
Añadido: hace 4 días
Reproducciones: 1558
Google Tech Talks
July, 17 2008
ABSTRACT
Three companies recently collaborated to use DT
Google Tech Talks
July, 17 2008
ABSTRACT
Three companies recently collaborated to use DTrace, a powerful open source process introspection tool to find and fix a substantial Rails latency issue.
Teams from Joyent and Twitter and DTrace developer Bryan Cantrill from Sun joined forces to spend a day looking in detail at how Ruby processes behave within a Rails production environment. The purpose of the collaboration was to use the dynamic tracing framework to fix a latency issue observed in Twitter. DTrace is one of the components of the open source project OpenSolaris. It is designed for forensic investigation of processes, and as such is perfectly suited for the inspection and monitoring of Ruby processes running Rails applications.
During their analysis, the joint team discovered that the raising and catching of particular set of exceptions within Rails caused large amounts of CPU time to be consumed generating back-traces hundreds of frames long. Through the detection and removal of these exceptions, the latency of a particular class of Rails request-response cycles was substantially improved.
Joyent CTO and co-founder Jason Hoffman is vocal about the benefits of DTrace:
We use DTrace all the time in identifying performance issues in our customer's and in our own applications.
However, he believes that there is still room for improvement, and that the introspection into Ruby processes that DTrace offers is still not deep enough. Patches for versions 1.8.5 and 1.8.6 of Ruby are currently being worked on that will afford DTrace with an even greater ability to peer inside Ruby's internals. Jason also believes that DTrace will continue to generate a tremendous amount of insight into production Rails and Ruby processes. With ports in development for FreeBSD and planned for OS X Leopard, DTrace is no longer an exclusive tool for Solaris users.
Interestingly, not only is this a success story for open source tools being used to improve open source frameworks, it is also a success story for the process of open source software development. Within 11 hours of a ticket being filed, it had been accepted by Rails creator David Heinemeier Hansson himself, and incorporated into the latest version of Rails for immediate use.
Speaker: Jason Hoffman
Jason A. Hoffman is a founder and the CTO of Joyent, an on-demand infrastructure and cloud computing company that serves billions of page views and traffics hundreds of millions of emails per month.
Joyent is dedicated to the singular mission that developers should be able to start at a small scale and flex to a global scale with minimal friction. Joyent is among the world's largest OpenSolaris installations and while supporting all unix-based languages and data stores, Joyent's products have included the first production support of Ruby on Rails, inclusion of the ZFS file system, and Joyent's DTrace-enabled Ruby ships on MacOS X Leopard and soon on OpenSolaris.
Jason is a systems scientist with BS and MS degrees from UCLA, and a PhD from UCSD, and is an expert in scalable architectures. He has applied his knowledge and experience from the Web to Games to Computational Chemistry, Proteomics and Cancer biology.
(más)
(menos)
Añadido: hace 4 días
Reproducciones: 1306
Google Tech Talks
May 5, 2008
ABSTRACT
Technology has transformed investment and tr
Google Tech Talks
May 5, 2008
ABSTRACT
Technology has transformed investment and trading over the past 30 years. Markets have become computer networks, brokers are disintermediated by direct access and algo trading. Reporters are disintermediated when investors have access to primary sources at the same time they do. An ever larger view of exploitable economic and business activity can found on the web. Dr. Leinweber brings an unusually broad and deep view to these issues, from both a sell- and buy-side perspective.
Speaker: David Leinweber
David Leinweber is Haas Fellow in Finance. His professional interests focus on how moderninformation technologies are best applied in trading and investing. As the founder of two financial technology companies, and a quantitative investment manager he is an active participant in today's transformation of markets.
He is an advisor to investment firms, stock exchanges, brokerages, and technology firms in areas related to financial markets , and a frequent speaker and author on these subjects. His book, "Nerds on Wall Street" will be published by Wiley in 2008.
He graduated from MIT, in physics and computer science and also has a Ph.D. in Applied Mathematics from Harvard. But on a good day, it's hard to tell.
(más)
(menos)
Añadido: hace 4 días
Reproducciones: 1109
|
Google Tech Talks
July, 14 2008
ABSTRACT
We'll go over what the new Maps API for Flash c
Google Tech Talks
July, 14 2008
ABSTRACT
We'll go over what the new Maps API for Flash can do, and show how to easily integrate it with other libraries for parsing XML/GeoRSS/KML/JSON. We'll also show how to use it with the non-JS version of the Google AJAX libraries.
The talk will include an introduction to ActionScript3 (like JS but better!), Flex development, and Flex controls.
Speaker: who/pamelafox
(más)
(menos)
Añadido: hace 4 días
Reproducciones: 886
Google Tech Talks
July, 15 2008
ABSTRACT
The Xbox 360 game platform was released with a
Google Tech Talks
July, 15 2008
ABSTRACT
The Xbox 360 game platform was released with a great deal of fanfare surrounding its new product design and user interface. The developers saw the need for a tight brand consistency with print, packaging, and product interface, and chose to commission a new typeface family for use throughout the brand. Steve presents the path to the Xbox 360's new look from the type designer's perspective: the evolution of the product, the design brief, the creative process, and the unique challenges of developing a font for less than optimal screen displays. Steve also discusses the business and technology of fonts, the work for the Android platform and, most importantly, why type matters in an age of multimedia and text messaging.
Speaker: STEVE MATTESON
STEVE MATTESON is the Director of Type Design for Ascender Corporation and has created fonts for use in various screen display environments and print publishing since 1987. A graduate of Rochester Institute of Technology, Steve has an extensive background in typography, design and printing which he has applied to his development of high quality typefaces.
His work can be found in user interface designs (such as Windows Vista, Xbox 360 and Google's Android Platform); in publishing (such as Pescadero Pro, Andy and Endurance Pro); and for corporate branding (such as Symantec, Microsoft and Alcon Labs). He resides in Louisville, CO with his wife, 2 kids and 2 Labrador Retrievers.
A partial showing of Steve' Portfolio can be found here: http://ascendercorp.com/portfolio_commercial.html
and here: http://ascendercorp.com/portfolio_custom.html
(más)
(menos)
Añadido: hace 1 semana
Reproducciones: 2438
Google Tech Talks
July, 15 2008
ABSTRACT
KML is the default language of virtual globes,
Google Tech Talks
July, 15 2008
ABSTRACT
KML is the default language of virtual globes, however it only delivers static data with pre-set rules of behavior. Using EarthBrowser as an example, new ideas for creating, presenting and controlling earth based information will be explored.
http://www.earthbrowser.com/
http://www.jskml.org/
Speaker: Matt Giger
Matt Giger is the founder of Lunar Software and creator of EarthBrowser. Matt was born and has lived in Oregon for most of his life. He received a B.A in Physics from Reed College, B.S. in Engineering and Applied Science from CalTech and a M.S. in Computer Science from The University of Oregon.
(más)
(menos)
Añadido: hace 1 semana
Reproducciones: 1791
|
|
Ver los 856 vídeos
|
