BestTechVideos: Tag computer-science Videos

Statistical Aspects of Data Mining (Stats 202) Day 2

scoundrel — Fri, 06 Jul 2007 13:42:24 GMT

Google Tech Talks
June 29, 2007

ABSTRACT

This is the Google campus version of Stats 202 which is being taught at Stanford this summer. I will follow the material from the Stanford class very closely. That material can be found at www.stats202.com. The main topics are exploring and visualizing data, association analysis, classification, and clustering. The textbook is Introduction to Data Mining by Tan, Steinbach and Kumar. Googlers are welcome to attend any classes which they think might be of interest to them.

Read more about this video…

Want to control this feed contents? Sign up here and create your own feed!

Want more on these topics?
Browse the archive of posts filed under Conferences, Companies, Broadcasting, Science, Techtalks, Google, Lectures, Statistics, Computer Science

Statistical Aspects of Data Mining (Stats 202) Day 1

scoundrel — Fri, 06 Jul 2007 13:41:43 GMT

Google Tech Talks
June 26, 2007

ABSTRACT

This is the Google campus version of Stats 202 which is being taught at Stanford this summer. I will follow the material from the Stanford class very closely. That material can be found at www.stats202.com. The main topics are exploring and visualizing data, association analysis, classification, and clustering. The textbook is Introduction to Data Mining by Tan, Steinbach and Kumar. Googlers are welcome to attend any classes which they think might be of interest to them.

Read more about this video…

Want to control this feed contents? Sign up here and create your own feed!

Want more on these topics?
Browse the archive of posts filed under Conferences, Companies, Broadcasting, Science, Techtalks, Google, Lectures, Statistics, Computer Science

Scaling the Internet routing table with Locator/ID Separation Protocol (LISP)

scoundrel — Fri, 06 Jul 2007 13:38:18 GMT

Google Tech Talks
June 12, 2007

ABSTRACT

This talk introduces a routing protocol approach (Locator/ID Separation Protocol) to solving the scaling problem of the global Internet routing table.

The Problem:
At the October '06 IAB meeting in Amsterdam the highest priority takeaway was the need to devise a scalable routing and addressing system for the Internet. The key points were:

Main driving forces behind the rapid growth of the Internet's DFZ (default-free zone) are multihoming, traffic engineering, non-aggregatable address allocations, and business events such as mergers and acquisitions.
Moore's Law does not apply to building high-end routers as far as cost is concerned. Low-volume, customized silicon used in core routing is well off Moore's Law's cost curve. In short, there are concerns regarding the industry's ability to scale core routing systems in the face of growing RIB (routing information base) sizes.
So-called "locator/identifier overload" of the IP address semantics is one of the causes of the routing scalability problem.

About LISP:
LISP is a simple, incremental, network-based protocol to implement separation of the Internet address into Endpoint IDs and Routing Locators. This mechanism requires no changes to host stacks and no major changes to existing database infrastructures.

Read more about this video…

Want to control this feed contents? Sign up here and create your own feed!

Want more on these topics?
Browse the archive of posts filed under Conferences, Companies, Broadcasting, Frameworks, Science, Techtalks, Google, Lectures, Symfony, XNA, Computer Science

Universal Cache Miss Equations for the Memory Hierarchy

scoundrel — Fri, 06 Jul 2007 13:31:58 GMT

Google Tech Talks
June 7, 2007

ABSTRACT

Computer systems in large enterprises and in the outsourcing industry are difficult to manage manually. This difficulty has led to a push for autonomic computing, where the goal is to have the system automatically configure its settings, tune its performance and recover from failures.

Such large systems have innumerable caches to enhance performance. One task in autonomic computing lies in sizing these caches and sharing them out among competing workloads. Cache miss equations can help to automate this task.

For the cache miss equations to be helpful, they must be universal: they must fit any real memory reference pattern and cache management policy, through parameters that can be calibrated automatically and dynamically.

This talk proposes a research program to derive such equations. This proposal is for a drastic departure from forty years of cache miss analysis. Three cache miss equations (for main memory, database buffers and processor caches) are presented as evidence for the feasibility of the research program.

Read more about this video…

Want to control this feed contents? Sign up here and create your own feed!

Want more on these topics?
Browse the archive of posts filed under Conferences, Companies, Broadcasting, Science, Techtalks, Google, Lectures, Math, Computer Science

The Asymptotic Performance of AdaBoost

scoundrel — Tue, 19 Jun 2007 02:39:48 GMT

Google Tech Talks
May 24, 2007

ABSTRACT

Many popular classification algorithms, including AdaBoost and the support vector machine, minimize a cost function that can be viewed as a convex surrogate of the 0-1 loss function. The convexity makes these algorithms computationally efficient. The use of a surrogate, however, has statistical consequences that must be balanced against the computational virtues of convexity. In this talk, we consider the universal consistency of such methods: does the risk, or expectation of the 0-1 loss, approach its optimal value, no matter what i.i.d. process generates.

Read more about this video…

Want to control this feed contents? Sign up here and create your own feed!

Want more on these topics?
Browse the archive of posts filed under Conferences, Companies, Broadcasting, Science, Techtalks, Google, Lectures, Computer Science

Multifractals: The scalability of random multiplicative processes

scoundrel — Fri, 15 Jun 2007 10:03:13 GMT

Google Tech Talks
May 15, 2007

ABSTRACT

Many interesting data are of sufficient size and complexity that they can only be studied statistically. Broad distributions, such as those produced by a multiplicative process, do not scale according to conventional laws. For these non-Gaussian distributions, a different exponent is needed to describe how each moment of the distribution scales. Multifractal analysis provides a means of determining these scaling exponents for a given data distribution. In this talk, we will introduce the concept of multifractals by first extending the more familiar concept of fractals, and then by deriving the theory of multifractals from statistical mechanics. We then provide an introduction to multiplicative random processes as a universal mechanism underlying the generation of many data. Then, after a brief review of an algorithm often used for calculating the multifractal spectrum (the Wavelet Transform Modulus Maxima), we will present a few applications of multifractals that we hope to be of interest: volatility of financial time series, characterization of (internet) traffic, and probabilistic measures on suffix trees.

Read more about this video…

Want to control this feed contents? Sign up here and create your own feed!

Want more on these topics?
Browse the archive of posts filed under Conferences, Companies, Broadcasting, Science, Techtalks, Google, Lectures, Math, Computer Science

A Multi-Feature Part-Based Object Detection System

scoundrel — Thu, 14 Jun 2007 03:53:43 GMT

Google Tech Talks
May 25, 2007

ABSTRACT

I will start with an overview on object recognition systems which use local features and analyze their strengths and weaknesses. I will then present a general purpose part-based object detection system which we evaluated on a benchmark pedestrian detection data set . In a first step, the system computes feature maps from the training images. It then randomly extracts a large number of rectangular parts from the feature maps and clusters the parts based on their feature similarity and their x-y-location in the feature maps. The cluster centers build an initial set of part templates from which the system selects a subset using the gentle-boost algorithm. The localization of the parts during classification is performed by normalized cross-correlation of the part templates with feature maps. Three different types of feature maps were used in our experiments: Original gray value images, the magnitudes of the gradient, and Gabor filtered images. In experiments on a benchmark pedestrian detection database, we investigate how the number of the components, the feature type and the training data affects the detection performance. The system is compared to state-of-the-art pedestrian detectors.

Speaker: Bernd Heisele, Honda

Read more about this video…

Want to control this feed contents? Sign up here and create your own feed!

Want more on these topics?
Browse the archive of posts filed under Conferences, Companies, Broadcasting, Science, Techtalks, Google, Lectures, Computer Science

Artificial General Intelligence: Now Is the Time

scoundrel — Wed, 13 Jun 2007 04:44:24 GMT

Google Tech Talks
May 30, 2007

ABSTRACT

Dr. Ben Goertzel - Artificial General Intelligence: Now Is the Time Essay
When the AI field was founded over 50 years ago, it was squarely focused on the grand dream of creating software displaying general intelligence at the human level or beyond. Since that time the field has drifted in a direction Ray Kurzweil has called "Narrow AI": the creation of intelligent software applications carrying out highly particular functions. The relationship between this sort of narrow AI and "artificial general intelligence" (AGI) as in the original dreams of the AI field, is an issue of dispute among experts.
Some researchers believe powerful AGI will result eventually from the development and combination of narrow AI products -- such as, for example, data mining software as is commonly used in the finance industry; auto navigation software like the kind used in the DARPA Grand Challenge; and last but not least, sophisticated search engines like Google. Other researchers believe that AGI will only come about via emulation of the human brain, once brain mapping technology has advanced further. On the other hand, an increasing minority of researchers believes that AGI is most likely to be achieved via computer science researchers explicitly attempting to create AGI software programs, divorced from any particular narrow application area.
In this talk I will briefly overview this emerging subdiscipline of "AGI", including the work of various researchers such as Stan Franklin, Pei Wang and Stuart Shapiro. I will then discuss my own work on the Novamente Cognition Engine, an AGI project based on combining a number of knowledge representations and reasoning and learning techniques into an integrative architecture motivated by complex systems theory, and initially oriented at the control of virtual agents in 3D simulation worlds such as Second Life.

Read more about this video…

Want to control this feed contents? Sign up here and create your own feed!

Want more on these topics?
Browse the archive of posts filed under Conferences, Companies, Broadcasting, Science, Techtalks, Google, Lectures, Computer Science

Advanced Topics in Programming Languages Series: Effective Static Race Detection

scoundrel — Tue, 12 Jun 2007 05:00:07 GMT

Google Tech Talks
June 4, 2007

ABSTRACT

Concurrent programs are notoriously difficult to write and debug, a problem that is becoming acute as concurrency becomes more common. A fundamental and particularly insidious concurrency bug is a race: a condition in which a pair of threads simultaneously access the same memory location and at least one of those accesses is a write. Today's concurrent programs are riddled with races.

This talk presents a novel approach to detecting races in concurrent Java programs. The approach is sound in that it detects all races and is precise in that it misidentifies few non-races as races. The approach is also effective, finding tens to hundreds of serious and previously unknown synchronization bugs in real-world programs.

Read more about this video…

Want to control this feed contents? Sign up here and create your own feed!

Want more on these topics?
Browse the archive of posts filed under Conferences, Companies, Development, Broadcasting, Science, Techtalks, Google, Debugging, Lectures, Computer Science

Is your RDBMS letting you down? Applications of TV Viewing Behavior

scoundrel — Tue, 12 Jun 2007 04:59:24 GMT

Google Tech Talks
May 23, 2007

ABSTRACT

As commodity hardware becomes cheaper and more powerful it becomes tempting to reflexively use relational databases to track and report on large datasets. Rentrak argues the case that a pure-RDBMS solution can be counterproductive, especially for complex analysis of those datasets. I'll describe an alternate approach we currently use to track and report on video-on-demand and linear television viewership.

Read more about this video…

Want to control this feed contents? Sign up here and create your own feed!

Want more on these topics?
Browse the archive of posts filed under Conferences, Companies, Databases, Broadcasting, Science, Techtalks, Google, Lectures, Computer Science