Christopher David Meek

US Patent:

6345265, Feb 5, 2002

Filed:

Dec 23, 1998

Appl. No.:

09/220192

Inventors:

Bo Thiesson - Kirkland WA 98033
Christopher A. Meek - Kirkland WA 98033
David Maxwell Chickering - Redmond WA 98052
David Earl Heckerman - Bellevue WA 98008

International Classification:

G06N 302

US Classification:

706 52, 706 45

Abstract:

The invention employs mixtures of Bayesian networks to perform clustering. A mixture of Bayesian networks (MBN) consists of plural hypothesis-specific Bayesian networks (HSBNs) having possibly hidden and observed variables. A common external hidden variable is associated with the MBN, but is not included in any of the HSBNs. The number of HSBNs in the MBN corresponds to the number of states of the common external hidden variable, and each HSBN is based upon the hypothesis that the common external hidden variable is in a corresponding one of those states. In one mode of the invention, the MBN having the highest MBN score is selected for use in performing inferencing. The invention determines membership of an individual case in a cluster based upon a set of data of plural individual cases by first learning the structure and parameters of an MBN given that data and then using the MBN to compute the probability of each HSBN generating the data of the individual case.

US Patent:

6408290, Jun 18, 2002

Filed:

Dec 23, 1998

Appl. No.:

09/220200

Inventors:

Bo Thiesson - Kirkland WA
Christopher A. Meek - Kirkland WA
David Maxwell Chickering - Redmond WA
David Earl Heckerman - Bellevue WA

Assignee:

Microsoft Corporation - Redmond WA

International Classification:

G06N 302

US Classification:

706 52, 706 45

Abstract:

One aspect of the invention is the construction of mixtures of Bayesian networks. Another aspect of the invention is the use of such mixtures of Bayesian networks to perform inferencing. A mixture of Bayesian networks (MBN) consists of plural hypothesis-specific Bayesian networks (HSBNs) having possibly hidden and observed variables. A common external hidden variable is associated with the MBN, but is not included in any of the HSBNs. The number of HSBNs in the MBN corresponds to the number of states of the common external hidden variable, and each HSBN is based upon the hypothesis that the common external hidden variable is in a corresponding one of those states. In one mode of the invention, the MBN having the highest MBN score is selected for use in performing inferencing. In another mode of the invention, some or all of the MBNs are retained as a collection of MBNs which perform inferencing in parallel, their outputs being weighted in accordance with the corresponding MBN scores and the MBN collection output being the weighted sum of all the MBN outputs. In one application of the invention, collaborative filtering may be performed by defining the observed variables to be choices made among a sample of users and the hidden variables to be the preferences of those users.

US Patent:

6496816, Dec 17, 2002

Filed:

Dec 23, 1998

Appl. No.:

09/220199

Inventors:

Bo Thiesson - Kirkland WA
Christopher A. Meek - Kirkland WA
David Maxwell Chickering - Redmond WA
David Earl Heckerman - Bellevue WA

Assignee:

Microsoft Corporation - Redmond WA

International Classification:

G06N 302

US Classification:

706 52, 706 45

Abstract:

One aspect of the invention is the construction of mixtures of Bayesian networks. Another aspect of the invention is the use of such mixtures of Bayesian networks to perform inferencing. A mixture of Bayesian networks (MBN) consists of plural hypothesis-specific Bayesian networks (HSBNs) having possibly hidden and observed variables. A common external hidden variable is associated with the MBN, but is not included in any of the HSBNs. The number of HSBNs in the MBN corresponds to the number of states of the common external hidden variable, and each HSBN is based upon the hypothesis that the common external hidden variable is in a corresponding one of those states. In one mode of the invention, the MBN having the highest MBN score is selected for use in performing inferencing. In another mode of the invention, some or all of the MBNs are retained as a collection of MBNs which perform inferencing in parallel, their outputs being weighted in accordance with the corresponding MBN scores and the MBN collection output being the weighted sum of all the MBN outputs. In one application of the invention, collaborative filtering may be performed by defining the observed variables to be choices made among a sample of users and the hidden variables to be the preferences of those users.

US Patent:

6519599, Feb 11, 2003

Filed:

Mar 2, 2000

Appl. No.:

09/517138

Inventors:

D. Maxwell Chickering - Redmond WA
David E. Heckerman - Bellevue WA
Christopher A. Meek - Kirkland WA
Robert L. Rounthwaite - Fall City WA
Amir Netz - Bellevue WA
Thierry DHers - Issaquah WA

Assignee:

Microsoft Corporation - Redmond WA

International Classification:

G06F 1730

US Classification:

707 10, 7071041

Abstract:

Visualization of high-dimensional data sets is disclosed, particularly the display of a network model for a data set. The network, such as a dependency or a Bayesian network, has a number of nodes having dependencies thereamong. The network can be displayed items and connections, corresponding to nodes and dependencies, respectively. Selection of a particular item in one embodiment results in the display of the local distribution associated with the node for the item. In one embodiment, only a predetermined number of the items are shown, such as only the items representing the most popular nodes. Furthermore, in one embodiment, in response to receiving a user input, a sub-set of the connections is displayed, proportional to the user input. In another embodiment, a particular item is displayed in an emphasized manner, and the particular connections representing dependencies including the node represented by the particular item, as well as the items representing nodes also in these dependencies, are also displayed in the emphasized manner. Furthermore, in one embodiment, only an indicated sub-set of the items is displayed.

US Patent:

6633852, Oct 14, 2003

Filed:

May 21, 1999

Appl. No.:

09/316704

Inventors:

David E. Heckerman - Bellevue WA
Christopher A. Meek - Kirkland WA
Usama M. Fayad - Mercer Island WA

Assignee:

Microsoft Corporation - Redmond WA

International Classification:

G06F 1760

US Classification:

705 27, 706 12

Abstract:

An electronic shopping aid is provided that assists a user in selecting a product from an electronic catalog of products based on their preferences for various features of the products. Since the electronic shopping aid helps a user select a product based on the users preferences, it is referred to as a preference-based product browser. In using the browser, the user initially inputs an indication of their like or dislike for various features of the products as well as an indication of how strongly they feel about the like or dislike. The browser then utilizes this information to determine a list of products in which the user is most likely interested. As part of this determination, the browser performs collaborative filtering and bases the determination on what other users with similar characteristics (e. g. , age and income) have liked. After creating this list, the browser displays the list and also displays a list of features which the user has not indicated either a like or a dislike for and which the browser has identified as being most relevant to the determination of the products that the user may like.

US Patent:

6694301, Feb 17, 2004

Filed:

Mar 31, 2000

Appl. No.:

09/540255

Inventors:

David E. Heckerman - Bellevue WA
D. Maxwell Chickering - Bellevue WA
John C. Platt - Bellevue WA
Christopher A. Meek - Kirkland WA
Bo Thiesson - Woodinville WA

Assignee:

Microsoft Corporation - Redmond WA

International Classification:

G06N 502

US Classification:

706 45, 706 26, 706 20, 706 25

Abstract:

Clustering for purposes of data visualization and making predictions is disclosed. Embodiments of the invention are operable on a number of variables that have a predetermined representation. The variables include input-only variables, output-only variables, and both input-and-output variables. Embodiments of the invention generate a model that has a bottleneck architecture. The model includes a top layer of nodes of at least the input-only variables, one or more middle layer of hidden nodes, and a bottom layer of nodes of the output-only and the input-and-output variables. At least one cluster is determined from this model. The model can be a probabilistic neural network and/or a Bayesian network.

US Patent:

6718315, Apr 6, 2004

Filed:

Dec 18, 2000

Appl. No.:

09/740067

Inventors:

Christopher A. Meek - Kirkland WA
David M. Chickering - Bellevue WA
Jeffrey R. Bernhardt - Woodinville WA
Robert L. Rounthwaite - Fall City WA

Assignee:

Microsoft Corporation - Redmond WA

International Classification:

G06F 1518

US Classification:

706 12, 706 14, 706 46

Abstract:

Disclosed is a system for approximating conditional probabilities using an annotated decision tree where predictor values that did not exist in training data for the system are tracked, stored, and referenced to determine if statistical aggregation should be invoked. Further disclosed is a system for storing statistics for deriving a non-leaf probability corresponding to predictor values, and a system for aggregating such statistics to approximate conditional probabilities.

US Patent:

6728690, Apr 27, 2004

Filed:

Nov 23, 1999

Appl. No.:

09/448408

Inventors:

Christopher A. Meek - Kirkland WA
John C. Platt - Bellevue WA

Assignee:

Microsoft Corporation - Redmond WA

International Classification:

G06N 302

US Classification:

706 25, 706 12

Abstract:

A training system for a classifier utilizes both a back-propagation system to iteratively modify parameters of functions which provide raw output indications of desired categories, wherein the parameters are modified based on a weighted decay, and a probability determining system with further parameters that are determined during iterative training. A margin error metric may be combined with weight decay, and a sigmoid is used to calibrate the raw outputs to probability percentages for each category. A method of training such a system involves gathering a training set of inputs and desired corresponding outputs. Classifier parameters are then initialized and an error margin is calculated with respect to the classifier parameters. A weight decay is then used to adjust the parameters. After a selected number of times through the training set, the parameters are deemed in final form, and an optimization routine is used to derive a set of probability transducer parameters for use in calculating the probable classification for each input.