Google Tech Talk
October 7, 2009

ABSTRACT

Presented by Dror Baron, Visiting Scientist, Technion - Israel Institute of Technology.

Traditional signal acquisition techniques sample band-limited analog signals above the Nyquist rate, which is related to the highest analog frequency in the signal. Compressed sensing (CS) is based on the revelation that optimization routines can reconstruct a sparse signal from a small number of linear projections of the signal. Therefore, CS-based techniques can acquire and process sparse signals at much lower rates. CS offers tremendous potential in applications such as broadband analog-to-digital conversion, where the Nyquist rate exceeds the state of the art.

Information theory has numerous insights to offer CS; I will describe several investigations along these lines. First, distributed compressed sensing (DCS) provides new distributed signal acquisition algorithms that exploit both intra- and inter-signal correlation structures in multi-signal ensembles. DCS is immediately applicable in sensor networks. Next, we leverage the remarkable success of graph reduction algorithms and LDPC channel codes to design low-complexity CS reconstruction algorithms.

Linear measurements play a crucial role not only in compressed sensing but in disciplines such as finance, where numerous noisy measurements are needed to estimate various statistical characteristics. Indeed, many areas of science and engineering seek to extract information from linearly derived measurements in a computationally feasible manner. Advances toward a unified theory of linear measurement systems will enable us to effectively process the vast amounts of data being generated in our dynamic world.

Read more about this video…

Google Tech Talk
March 19, 2010

ABSTRACT

Presented by Geoff Hinton, University of Toronto.

Deep networks can be learned efficiently from unlabeled data. The layers of representation are learned one at a time using a simple learning module that has only one layer of latent variables. The values of the latent variables of one module form the data for training the next module. Although deep networks have been quite successful for tasks such as object recognition, information retrieval, and modeling motion capture data, the simple learning modules do not have multiplicative interactions which are very useful for some types of data.

The talk will show how to introduce multiplicative interactions into the basic learning module in a way that preserves the simple rules for learning and perceptual inference. The new module has a structure that is very similar to the simple cell/complex cell hierarchy that is found in visual cortex. The multiplicative interactions are useful for modeling images, image transformations, and different styles of human walking.

Speaker bio: http://www.cs.toronto.edu/~hinton/bio.html

Read more about this video…

Google Tech Talk
May 5, 2010

ABSTRACT

Presented by Justin Ma.

We explore online learning approaches for detecting malicious Web sites (those involved in criminal scams) using lexical and host-based features of the associated URLs. We show that this application is particularly appropriate for online algorithms as the size of the training data is larger than can be efficiently processed in batch and because the distribution of features that typify malicious URLs is changing continuously. Using a real-time system we developed for gathering URL features, combined with a real-time source of labeled URLs from a large Web mail provider, we demonstrate that recently-developed online algorithms can be as accurate as batch techniques, achieving daily classification accuracies up to 99% over a balanced data set.

Slides: http://cseweb.ucsd.edu/~jtma/google_talk/jtma-google10.pdf

Justin Ma is a PhD candidate at UC San Diego advised by Stefan Savage, Geoff Voelker and Lawrence Saul. His research interests are in systems and networking with an emphasis on network security, and his current focus is the application of machine learning to problems in security. He will be joining UC Berkeley as a postdoc after graduation. [Home page: http://www.cs.ucsd.edu/~jtma/ ]

Read more about this video…

Google Tech Talk
August 11, 2010

ABSTRACT

Presented by Alice Oh.

Probabilistic topic models are useful for uncovering the underlying semantic structure of a collection of documents. We take a simple and widely used topic model, the Latent Dirichlet Allocation (LDA, Blei et al. 2003) and extend it in two ways. First, we construct topic chains to understand the dynamic semantic structure of an online news corpus, and second, we develop a unification model of sentiment and aspect to discover the detailed semantics of user generated reviews.

For the topic chains research, we present a framework for comparing and clustering the topics discovered by LDA. We discuss how to interpret the resulting topic chains to understand the general topic trends, temporary issues, and the focus shifts within the topic chains. We applied the topic chains framework to 9 months of online news and present the results.

For the aspect-sentiment unification model (ASUM) research, we base our model on an observation that users evaluate various aspects of a product in a review, such as the lens of the camera or the LCD display of a laptop, and each sentence usually represents one aspect and a corresponding sentiment toward that aspect. So we propose a sentence-LDA (SLDA) with a constraint that all words in a single sentence are generated from one aspect. Then we extend SLDA to Aspect and Sentiment Unification Model (ASUM) to jointly discover pairs of {aspect, sentiment} which we call senti-aspects. We applied SLDA and ASUM to reviews of electronic devices and restaurants and present the results.

Alice Oh is an Assistant Professor of Computer Science at Korea Advanced Institute of Science and Technology. She leads her research group, Users and Information Lab, with the vision of delivering information to satisfy the user. To that end, she studies and employs methods from machine learning, human-computer interaction, and statistical natural language processing. Alice completed her M.S. in Language and Information Technologies at CMU and her Ph.D. in Computer Science at MIT.

Read more about this video…

Replace ads with social media links

Read more about this video…

See thumbnails of all your open tabs

Read more about this video…

Create a virtual drive from any existing folder

Are you looking for a fast and easy way to create a virtual drive on your PC? On today's Tekzilla Daily, Veronica shows you how to use vSubst, a free tool that makes virtual drive creation a snap.

Read more about this video…

A new way to discover interesting add-ons

Tired of sifting through pages and pages of Firefox addons to find the best ones? Find the ones most relevant to you as suggested by other Mozilla users, with Firefox Collections. Veronica shows you how it works on today's Tekzilla Daily.

Read more about this video…

Use a virtual ruler to calculate the size of your windows.

Read more about this video…

Read more about this video…

Read more about this video…

Save your system from doom!

Worried about your Windows 7 machine dying and losing all your data? Veronica shows you how to take steps NOW to prevent the unthinkable on today's Tekzilla Daily.

Read more about this video…

Optimize your screen display

Read more about this video…

Prevent unwanted shouting

Have you even accidentally hit that caps lock key and not realized it until you've already written that whole email? Veronica shows you how to avoid the frustration on today's Tekzilla Daily.

Read more about this video…

Keep your bookmarks organized

You'll never have to spend another minute sorting your bookmarks into folders again! Veronica shows you how to make use of Firefox's smart folders on today's Tekzilla Daily.

Read more about this video…

Restore your favorite XP feature to Win7

Read more about this video…

Track your machine's resources

Read more about this video…

Terminator Salvation is just a week away, and to celebrate Ryan and QuintalDesigns.com have created a tutorial showing us how to use digital makeup to turn anyone into a Terminator. You'll soon be feeling like Arnold Swarchenegger telling people, Hasta La Vista, Baby when all is said and done.

This HD Video Tutorial in Photoshop covers masking techniques and layer effects, placing documents and blending things for realistic effects.

Read more about this video…

Verify your links

No one likes broken links! Find out how to avoid those deadends without having to check them manually on today's Tekzilla Daily.

Read more about this video…