Eric T Bax, 56314 N Sunnyside Ave, Sierra Madre, CA 91024

6732083, May 4, 2004

Feb 21, 2001

09/790124

Eric T Bax - Pasadena CA, 91116-6543

G06F 1518

706 12, 706 20, 711173

A computer-based system computes a probabilistic bound on the error probability of a nearest neighbor classifier as follows. A subset of the examples in the classifier is used to form a reduced classifier. The error frequency of the reduced classifier on the remaining examples is computed as a baseline estimate of the error probability for the original classifier. Additionally, subsets of the examples outside the reduced classifier are combined with the reduced classifier and applied to the remaining examples in order to estimate the difference in error probability for the reduced classifier and error probability for the original classifier.

6850873, Feb 1, 2005

Dec 18, 2000

09/740188

Eric T Bax - Pasadena CA, US

G06F 1710

703 2, 700 51, 700 91

Given a set of basis functions, a set of example inputs, and a set of uniform error bounds for the basis functions over the example inputs, a quadratic program is formed. The quadratic program is solved, producing a solution vector and a solution value. A hypothesis function is formed through fusion by using the solution vector to weight the outputs of the basis function. The hypothesis function is a function with minimum error bound among the functions formed by convex combination of basis function outputs. The solution value is an error bound for the hypothesis function. The error bound is logically implied by the uniform error bounds over the basis functions rather than uniform error bounds over the entire class of functions formed by convex combination of basis function outputs.

7590635, Sep 15, 2009

Dec 14, 2005

11/300950

W. Daniel Hillis - Encino CA, US
Eric Bax - Altadena CA, US
Mathias L. Kolehmainen - Los Angeles CA, US

Applied Minds, Inc. - Glendale CA

G06F 7/00
G06F 17/30

707 10, 707 4

A distributed data store labels operations with globally unique identifiers that contain approximate timestamps. The labels are used to address causes of inconsistency in the distributed data store while ensuring progress. A first mode is provided that stores the latest label for each entry is useful if re-inserts and deletes are rare. Another mode is provided that stores a history of labels for each entry can be used if there are many re-inserts and deletes. A further mode is provided that stores a history of labels for queries can report updates to query answers as inserts and deletes settle across the distributed data store.

7599913, Oct 6, 2009

Feb 9, 2004

10/775744

Joseph Greg Billock - Altadena CA, US
Ian Douglas Swett - Pasadena CA, US
Eric Theodore Bax - Altadena CA, US

Avaya, Inc. - Basking Ridge NJ

G06F 17/30

707 3, 707 5, 707101, 7071041

A time series monitoring system, implemented in software, executes persistent queries on multiple input time series, handling high data throughput with low response time. The system supports dynamic management of time series, of windows in time series, and of persistent queries. Also, the system can use historical values in present windows to help populate inserted windows.

7672965, Mar 2, 2010

Feb 9, 2004

10/775745

Eric Theodore Bax - Altadena CA, US

Avaya, Inc. - Basking Ridge NJ

G06F 17/00

707102, 707 3, 716 1, 326121, 381 61

A process performs multiple evaluations of text simultaneously. There are multiple counters, each with pattern-amount pairs. The pattern-amount pairs are accumulated into a single finite-state machine, with each state having a list of (counter, value) pairs instead of a single value. While the finite-state machine is applied to text, a score for each counter is accumulated by summing values for the counter from value lists of visited states. With one state transition per character, evaluating text using one finite-state machine for multiple counters is more efficient than using separate finite-state machines for counters or patterns.

7689906, Mar 30, 2010

Dec 1, 2000

09/728689

Eric T. Bax - Pasadena CA, US
Charless C. Fowlkes - Bozeman MT, US
Louis Cisnero, Jr. - Jourdanton TX, US

AVAYA, Inc. - Basking Ridge NJ

G06F 17/21

715237, 715234

The present invention discloses a technique for extracting data from a file. In accordance with the present invention, a request to extract one or more data records from the file is received. The data records within the file are identified, without using prior knowledge of a structure of the file. The data records are then extracted.

7778803, Aug 17, 2010

Sep 28, 2000

09/677334

Eric T Bax - Pasadena CA, US

G06F 17/10

703 2, 703 6, 706 12, 706 20, 707100

An error bound that indicates how well a hypothesis function approximates an unknown target function over a set of out-of-sample examples is computed from known error bounds for basis functions, as follows. An optimization problem is formed in which basis function error bounds imply constraints on feasible outputs of the target function over out-of-sample inputs. The optimization problem is solved to find an upper bound on the differences between the hypothesis function outputs and feasible target function outputs. This upper bound is a an error bound for the hypothesis function.

7899766, Mar 1, 2011

Feb 7, 2008

12/069129

Eric Theodore Bax - Altadena CA, US
Augusto Daniel Callejas - Pasadena CA, US

G06F 15/18
G06F 11/07

706 20, 706 25

Given a set of training examples—with known inputs and outputs—and a set of working examples—with known inputs but unknown outputs—train a classifier on the training examples. For each possible assignment of outputs to the working examples, determine whether assigning the outputs to the working examples results in a training and working set that are likely to have resulted from the same distribution. If so, then add the assignment to a likely set of assignments. For each assignment in the likely set, compute the error of the trained classifier on the assignment. Use the maximum of these errors as a probably approximately correct error bound for the classifier.