AT&T Intellectual Property I, LP

At+t Corp.

Daniel Leon Stern

Rittwik Jana

Yih-Farn Robin Chen

David Dolan Lewis

Amitabh Kumar Singhal

Ping-Fai Yang

Hailong Sun

Dilek Zeynep Hakkani-Tur

Gokhan Tur

Mazin G Rahim

Giuseppe Riccardi

Barbara B Hollister

James M Wilson

Tirso M Alonso

Ilana Bromberg

Lawrence Lyon Rose

A video-on-demand (VOD) server in an Internet protocol television (IPTV) network with a network of set-top boxes includes IPTV operational data and an analysis module that selects a content item. The VOD server divides the content item into sequential chunks, divides the chunks into stripes, pre-populates set-top boxes in a peering network with at least one stripe from each chunk, receives a request for the content item, and directs each set-top box to sequentially stream its stripes. A method includes selecting a content item based upon IPTV operational data, dividing the content item into sequential chunks, dividing the chunks into stripes, pre-populating set-top boxes in a peering network with at least one stripe from each chunk, receiving a request for the content item, and directing each set-top box to sequentially stream its stripes.

Analyzing internet protocol television data to support peer-assisted video-on-demand content delivery

Systems and methods for annotating speech data. The present invention reduces the time required to annotate speech data by selecting utterances for annotation that will be of greatest benefit. A selection module uses speech models, including speech recognition models and spoken language understanding models, to identify utterances that should be annotated based on criteria such as confidence scores generated by the models. These utterances are placed in an annotation list along with a type of annotation to be performed for the utterances and an order in which the annotation should proceed. The utterances in the annotation list can be annotated for speech recognition purposes, spoken language understanding purposes, labeling purposes, etc. The selection module can also select utterances for annotation based on previously annotated speech data and deficiencies in the various models.

Reducing time for annotating speech data to develop a dialog application

Apparatus for adding new learning tasks to an incremental supervised learner provides a flexible incremental representation of all encountered training examples, thereby permitting state representations for new learning tasks to take advantage of incremental training already completed by encoding all past training examples as negative examples for a hypothetical learning task. The state representation of the hypothetical learning task is copied as the initial state representation for a new learning task to be initiated, and is initialized with negative training examples of all previously presented training examples, thereby permitting the learning task to efficiently incorporate the previous examples.

Methods and apparatus for communicating information in a supervised learning system

A method and apparatus for adding new learning tasks to an incremental supervised learner provide a flexible incremental representation of all training examples encountered, thereby permitting state representations for new learning tasks to take advantage of incremental training already completed by encoding all past training examples as negative examples for a hypothetical learning task. The state representation of the hypothetical learning task is copied as the initial state representation for a new learning task to be initiated, is initialized with negative training examples of all previously presented training examples, thereby permitting the learning task to incorporate the previous examples efficiently.

A method and apparatus for communicating accumulated state information between internal and external tasks in a supervised learning system. A supervised learning system encodes state information for a hypothetical learning task on initialization. This hypothetical learning task state information indicates that no training instances have been received. During the supervised learning, training instances are presented to the supervised learner. The training instances are encoded with feature vector and target value information. For each task name paired with a non-default target value, the learner initializes a new learning task by copying the hypothetical learning task state representation for use as the state representation for the new learning task. Predictors are then produced for all learning tasks, except the hypothetical learning task. The new training instance is used to update all learning tasks as specified in the target vector. The new training instance is then used.to update the hypothetical learning task state representation as a negative example. Further training instances are handled similarly, new learning tasks are started based on the examination of the sparse target vector for task name, target value pairs which match received training instance target values and for which tasks have not yet been started.

Growing community of inventors

Princeton, NJ, United States of America

Daniel Leon Stern