Google Inc.

Other

Pagebites, Inc.

Georges R Harik

Jeffrey Adgate Dean

Noam M Shazeer

Simon Tong

Paul Thomas Buchheit

Joshua L Levenberg

Jeremy Bem

John Piscitello

Benedict A Gomes

Narayanan Shivakumar

Darrell Anderson

Krishna Asur Bharat

Deepak Jindal

Yingwei Cui

Eric Russell Fredricksen

Alexander Paul Carobus

Monika Hildegard Henzinger

Obeka Tallis Brown Bakin

David R Cheng

Sanjay Ghemawat

Fritz Schneider

Arturo Crespo

Niels Provos

Hanping Feng

Louis Perrochon

Thomas Alan Nielsen

Naga Sridhar Kataru

Benjamin Chan-Bin Ling

Ralph Harik

Marissa Ann Mayer

Keith Jonathan Coleman

Ross Koningstein

Monika Henzinger

Valentin Spitkovsky

Praveen Krishnamurthy

Paul Bucheit

Kin Lun Law

Siwei Shen

Kai Huang

Carl Laurence Gonsalves

Eric Russell Fredrickson

Paul Buchhcit

The relevance of advertisements to a user's interests is improved. In one implementation, the content of a web page is analyzed to determine a list of one or more topics associated with that web page. An advertisement is considered to be relevant to that web page if it is associated with keywords belonging to the list of one or more topics. One or more of these relevant advertisements may be provided for rendering in conjunction with the web page or related web pages.

Methods and apparatus for serving relevant advertisements

Targeting information (also referred to as ad 'serving constraints') or candidate targeting information for an advertisement is identified. Targeting information may be identified by extracting topics or concepts from, and/or generating topics or concepts based on, ad information, such as information from a Web page to which an ad is linked (or some other Web page of interest to the ad or advertiser). The topics or concepts may be relevant queries associated with the Web page of interest, clusters, etc.

Suggesting and/or providing ad serving constraint information

A data compression system includes: (a) a data compression module that receives a sequence of input vectors and that provides a sequence of compressed vectors; (b) a data decompression module that receives the compressed vectors to provide a sequence of output vectors; and (c) a parameter update module that receives the sequence of input vectors and the sequence of output vectors, and which learns the data compression module and data decompression module based on evaluating a loss function of the input vectors, the output vectors, and the parameters controlling the compression module and the decompression module. Each input vector and its corresponding output vector may represent digitized time-domain signals (e.g., speech, audio or video signals) over a predetermined time period. The loss function may be evaluated for each of a sequence of predetermined time periods.

Method for data compression

Keyword suggestions that are category-aware (and field-proven) may be used to help advertisers better target the serving of their ads, and may reduce unused ad spot inventory. The advertiser can enter ad information, such as a creative, a landing Webpage, other keywords, etc. for example. A keyword facility may use this entered ad information as seed information to infer one or more categories. It may then request that the advertiser confirm or deny some basic feedback information (e.g., categories, Webpage information, etc.). For example, an advertiser may be provided with candidate categories and may be asked to confirm (e.g., using checkboxes) which of the categories are relevant to their ad. Keywords may be determined using at least the categories. The determined keywords may be provided to the advertiser as suggested keywords, or may automatically populate ad serving constraint information as targeting keywords. The ad server system can run a trial on the determined keywords to qualify or disqualify them as targeting keyword.

Suggesting and/or providing targeting criteria for advertisements

A method which does not rely on explicit inverted indices is provided to search for documents in a corpus of documents responsive to a textual search query. The method includes (a) selecting a program that is customized by setting values for a plurality of parameters, the program structured to receive the textual search query as input and to provide as output values indicating the relevance of the documents in the corpus to the search query; (b) training the program using a machine learning technique; and (c) applying the trained program to the textual search query. The program may be based on a structure that is developed based on a genetic programming technique.

Searching text via function learning

A system ranks documents based, at least in part, on a ranking model. The ranking model may be generated to predict the likelihood that a document will be selected. The system may receive a search query and identify documents relating to the search query. The system may then rank the documents based, at least in part, on the ranking model and form search results for the search query from the ranked documents.

Ranking documents based on large data sets

Different ad selection techniques may be evaluated and compared by (i) combining ads generated using at least two different techniques and (ii) determining performance statistics of these combined (e.g., interleaved) advertising serves over time. The relative performance of the different techniques can then be determined. These principles can also be used to gauge different ad scoring techniques. These principles can also be used to gauge different ad rendering techniques.

Determining better ad selection, scoring, and/or presentation techniques

A method optimizes a predictive computation model efficiently. The method includes (i) selecting model parameters that are expected to take real values within a one-sided predetermined range; and (ii) iteratively: (a) receiving a set of input values; (b) executing the computation model based on the input values; (c) updating the values of the model parameters to minimize a loss function; and (d) examining each of the model parameters, such that, when the examined model parameter attains or moves past a value that is idempotent to the computation model, removing the model parameter from the computation model. In one embodiment, the predetermined range is either the range between a predetermined positive real value and positive infinity or the range between a predetermined negative real value and negative infinity. The predetermined positive real value or the predetermined negative real value may be an idempotent value to the model computation.

Method for improving efficiency in an optimizing predictive model using stochastic gradient descent

A method provides program structures for constructing a program that is learned over training data. In one example, two specific program structures are provided in which the first program structure transforms each vector in an input tuple of vectors to provide an output tuple of vectors, and the second program structure operates on an input tuple of vectors to provide an output tuple of vectors by applying one or more transformations that each involves two or more vectors in the input tuple. The transformations of the first and second program structures may be linear transformations. The program may alternatively execute the first program structure and the second program structure in any suitable order a number of times, beginning with operating one of the program structures on an initial tuple of vectors. The vectors may each consist of an ordered set of real numbers.

Configuring and optimizing computational structure for a machine learning application using a tuple of vectors

An apparatus and a method are provided for learning a program with a large number of parameters. In one embodiment, a method not only distorts the input values, but also distorts some of the parameters in the program model. Such an approach not only forces the learned program to acquire parameter values to predict missing or desired data, but also to correct errors in the input data and the program parameters themselves, thereby rendering the learned program more resilient to overfitting and falling into local optima.

Method and apparatus for improving resilience in customized program learning network computational environments

Communications between entities who may share common interests. For entities determined to be sharing common interests (e.g., searching using the same terms or topics, browsing a page, a site or a groups of topically related sites), options for communication among the entities are provided. For example, a chat room may be dynamically created for persons who are currently searching or browsing the same or related information. As another example, a 'homepage' may be created for each query and contain various types of information related to the query. A permission module controls which entities may participate, what types of information (and from what sources) an entity can (or desires to) receive, what types of information the entity may (or desires to) share.

System and method for impromptu shared communication spaces

Growing community of inventors

Mountain View, CA, United States of America

Georges R Harik