Xiao Hui Li

7634406, Dec 15, 2009

Dec 10, 2004

11/009630

Xiao Li - Seattle WA, US
Asela J. Gunawardana - Seattle WA, US
Alejandro Acero - Bellevue WA, US
Milind Mahajan - Redmond WA, US
Dong Yu - Kirkland WA, US

Microsoft Corporation - Redmond WA

G10L 15/06

704244, 704243, 704257

In accordance with one embodiment of the present invention, unanticipated semantic intents are discovered in audio data in an unsupervised manner. For instance, the audio acoustics are clustered based on semantic intent and representative acoustics are chosen for each cluster. The human then need only listen to a small number of representative acoustics for each cluster (and possibly only one per cluster) in order to identify the unforeseen semantic intents.

7881820, Feb 1, 2011

Aug 30, 2006

11/512631

Felix F. Antony - Issaquah WA, US
Xiao Yu Li - Seattle WA, US

Amazon Technologies, Inc. - Reno NV

G06F 7/00

700214

A method and system for inventory placement according to expected item picking rates. In one embodiment, a method may include determining a respective expected picking rate for each of a number of inventory items, and dependent upon the expected picking rate, selecting a corresponding one of a number of zones of an inventory storage area for each of the items. The zones may be physically arranged within the inventory storage area such that a first, innermost zone is successively and at least partially surrounded by one or more other zones. The method may further include storing each of the items within the corresponding zones, such that the expected picking rates of members of a given group of items stored in a given zone are less than the expected picking rates of members of another group of items stored in a successive zone that at least partially surrounds the given zone.

7991615, Aug 2, 2011

Dec 7, 2007

11/952267

Xiao Li - Bellevue WA, US
Asela J. R. Gunawardana - Seattle WA, US
Alejandro Acero - Bellevue WA, US

Microsoft Corporation - Redmond WA

G10L 15/04

704254, 704244, 704255

Described is the use of acoustic data to improve grapheme-to-phoneme conversion for speech recognition, such as to more accurately recognize spoken names in a voice-dialing system. A joint model of acoustics and graphonemes (acoustic data, phonemes sequences, grapheme sequences and an alignment between phoneme sequences and grapheme sequences) is described, as is retraining by maximum likelihood training and discriminative training in adapting graphoneme model parameters using acoustic data. Also described is the unsupervised collection of grapheme labels for received acoustic data, thereby automatically obtaining a substantial number of actual samples that may be used in retraining. Speech input that does not meet a confidence threshold may be filtered out so as to not be used by the retrained model.

8024064, Sep 20, 2011

Jun 30, 2008

12/164636

Kaushal A. Sanghavi - Seattle WA, US
Kalyanaraman Prasad - Seattle WA, US
Xiao Yu Li - Seattle WA, US
Pradeep Desai - Bellevue WA, US
Han S. Lee - Issaquah WA, US
Nadia Shouraboura - Newcastle WA, US

Amazon Technologies, Inc. - Reno NV

G06F 12/00
G06F 1/14
G06F 7/00

700226, 711165, 711114, 705 28, 700213

In various embodiments, approaches for the placement of inbound inventory in a materials handling facility are described. A product identifier is input from inbound inventory into a computer system, where the inbound inventory is to be stocked in a materials handling facility. At least one available inventory location is identified in the computer system that presents a lowest cost for storage of the inbound inventory in the materials handling facility. The inbound inventory is stocked in the at least one available inventory location contemporaneously with the determination of the at least one available inventory location in the computer system that presents the lowest cost for storage of the inbound inventory in the materials handling facility.

8180640, May 15, 2012

Jun 20, 2011

13/164683

Xiao Li - Bellevue WA, US
Asela J. R. Gunawardana - Seattle WA, US
Alejandro Acero, Jr. - Bellevue WA, US

Microsoft Corporation - Redmond WA

G10L 15/04

704254, 704255, 704257

Described is the use of acoustic data to improve grapheme-to-phoneme conversion for speech recognition, such as to more accurately recognize spoken names in a voice-dialing system. A joint model of acoustics and graphonemes (acoustic data, phonemes sequences, grapheme sequences and an alignment between phoneme sequences and grapheme sequences) is described, as is retraining by maximum likelihood training and discriminative training in adapting graphoneme model parameters using acoustic data. Also described is the unsupervised collection of grapheme labels for received acoustic data, thereby automatically obtaining a substantial number of actual samples that may be used in retraining. Speech input that does not meet a confidence threshold may be filtered out so as to not be used by the retrained model.

8250015, Aug 21, 2012

Apr 7, 2009

12/419336

Xiao Li - Bellevue WA, US

Microsoft Corporation - Redmond WA

G06F 17/00
G06N 5/00

706 45, 700 47, 707723

A training module is described for training a conditional random field (CRF) tagging model. The training module trains the tagging model based on an explicitly-labeled training set and an implicitly-labeled training set. The explicitly-labeled training set includes explicit labels that are manually selected via human annotation, while the implicitly-labeled training set includes implicit labels that are generated in an unsupervised manner. In one approach, the training module can train the tagging model by treating the implicit labels as non-binding evidence that has a bearing on values of hidden state sequence variables. In another approach, the training module can treat the implicit labels as binding or hard evidence. A labeling system is also described for providing the implicit labels.

8396585, Mar 12, 2013

Dec 29, 2010

12/981388

Felix F. Antony - Issaquah WA, US
Xiao Yu Li - Seattle WA, US

Amazon Technologies, Inc. - Reno NV

G06F 7/00

700216

A method and system for inventory placement according to expected item picking rates. In one embodiment, a method may include determining a respective expected picking rate for each of a number of inventory items, and dependent upon the expected picking rate, selecting a corresponding one of a number of zones of an inventory storage area for each of the items. The zones may be physically arranged within the inventory storage area such that a first, innermost zone is successively and at least partially surrounded by one or more other zones. The method may further include storing each of the items within the corresponding zones, such that the expected picking rates of members of a given group of items stored in a given zone are less than the expected picking rates of members of another group of items stored in a successive zone that at least partially surrounds the given zone.

8407214, Mar 26, 2013

Jun 25, 2008

12/145508

Xiao Li - Bellevue WA, US

Microsoft Corp. - Redmond WA

G06F 17/30

707726, 707728, 707731, 707737, 707749, 707798

To construct a classifier, a data structure correlating queries to items identified by the queries is received, where the data structure contains initial labeled queries that have been labeled with respect to predetermined classes, and unlabeled queries that have not been labeled with respect to the predetermined classes. The data structure is used to label at least some of the unlabeled queries with respect to the predetermined classes. Queries in the data structure that have been labeled with respect to the predetermined classes are used as training data to train the classifier.