公开(公告)号：US11803779B2

公开(公告)日：2023-10-31

申请号：US16800040

申请日：2020-02-25

Applicant: International Business Machines Corporation

Inventor： Thomas Parnell ,  Andreea Anghel ,  Nikolas Ioannou ,  Nikolaos Papandreou ,  Celestine Mendler-Duenner ,  Dimitrios Sarigiannis ,  Charalampos Pozidis

IPC: G06N20/20 ,  G06N5/01

CPC classification number: G06N20/20 ,  G06N5/01

Abstract: In an approach for constructing an ensemble model from a set of base learners, a processor performs a plurality of boosting iterations, where: at each boosting iteration of the plurality of boosting iterations, a base learner is selected at random from a set of base learners, according to a sampling probability distribution of the set of base learners, and trained according to a training dataset; and the sampling probability distribution is altered: (i) after selecting a first base learner at a first boosting iteration of the plurality of boosting iterations and (ii) prior to selecting a second base learner at a final boosting iteration of the plurality of boosting iterations. A processor constructs an ensemble model based on base learners selected and trained during the plurality of boosting iterations.

公开(公告)号：US20230177120A1

公开(公告)日：2023-06-08

申请号：US17457698

申请日：2021-12-06

Applicant: International Business Machines Corporation

Inventor： Nikolaos Papandreou ,  Charalampos Pozidis ,  Milos Stanisavljevic ,  Jan Van Lunteren ,  Thomas Parnell ,  Cedric Lichtenau ,  Andrew M. Sica

IPC: G06K9/62 ,  G06N5/00

CPC classification number: G06K9/6224 ,  G06K9/6269 ,  G06N5/003

Abstract: A tensor representation of a machine learning inferences to be performed is built by forming complementary tensor subsets that respectively correspond to complementary subsets of one or more leaf nodes of one or more decision trees based on statistics of the one or more leaf nodes of the one or more decision trees and data capturing attributes of one or more split nodes of the one or more decision trees and the one or more leaf nodes of the decision trees. The complementary tensor subsets are ranked such that a first tensor subset and a second tensor subset of the complementary tensor subsets correspond to a first leaf node subset and a second leaf node subset of the complementary subsets of the one or more leaf nodes.

公开(公告)号：US11573803B2

公开(公告)日：2023-02-07

申请号：US16405334

申请日：2019-05-07

Applicant: International Business Machines Corporation

Inventor： Nikolas Ioannou ,  Celestine Duenner ,  Thomas Parnell

IPC: G06F9/38 ,  G06N20/00 ,  G06F9/50

Abstract: Parallel training of a machine learning model on a computerized system is described. Computing tasks of a system can be assigned to multiple workers of the system. Training data can be accessed. The machine learning model is trained, whereby the training data accessed are dynamically partitioned across the workers of the system by shuffling subsets of the training data through the workers. As a result, different subsets of the training data are used by the workers over time as training proceeds. Related computerized systems and computer program products are also provided.

公开(公告)号：US11461694B2

公开(公告)日：2022-10-04

申请号：US16144550

申请日：2018-09-27

Applicant: International Business Machines Corporation

Inventor： Thomas Parnell ,  Celestine Duenner ,  Dimitrios Sarigiannis ,  Charalampos Pozidis

IPC: G06N20/00 ,  G06F17/16 ,  G06F17/13

Abstract: Methods are provided for implementing training of a machine learning model in a processing system, together with systems for performing such methods. A method includes providing a core module for effecting a generic optimization process in the processing system, and in response to a selective input, defining a set of derivative modules, for effecting computation of first and second derivatives of selected functions ƒ and g in the processing system, to be used with the core module in the training operation. The method further comprises performing, in the processing system, the generic optimization process effected by the core module using derivative computations effected by the derivative modules.

公开(公告)号：US11315035B2

公开(公告)日：2022-04-26

申请号：US16214918

申请日：2018-12-10

Applicant: International Business Machines Corporation

Inventor： Thomas Parnell ,  Celestine Duenner ,  Charalampos Pozidis ,  Dimitrios Sarigiannis

IPC: G06N20/00 ,  G06K9/62 ,  G06F17/11 ,  G06N7/08 ,  G06F7/58

Abstract: Computer-implemented methods are provided for implementing training of a machine learning model in a heterogeneous processing system comprising a host computer operatively interconnected with an accelerator unit. The training includes a stochastic optimization process for optimizing a function of a training data matrix X, having data elements Xi,j with row coordinates i=1 to n and column coordinates j=1 to m, and a model vector w having elements wj. For successive batches of the training data, defined by respective subsets of one of the row coordinates and column coordinates, random numbers associated with respective coordinates in a current batch b are generated in the host computer and sent to the accelerator unit. In parallel with generating the random numbers for batch b, batch b is copied from the host computer to the accelerator unit.

公开(公告)号：US20210334709A1

公开(公告)日：2021-10-28

申请号：US16858900

申请日：2020-04-27

Applicant: INTERNATIONAL BUSINESS MACHINES CORPORATION

Inventor： Nikolas Ioannou ,  Andreea Anghel ,  Thomas Parnell ,  Nikolaos Papandreou ,  Charalampos Pozidis

IPC: G06N20/20 ,  G06N3/00 ,  G06N5/00

Abstract: The present invention is notably directed to a computer-implemented method of training a cognitive model. The cognitive model includes decision trees as base learners. The method is performed using processing means to which a given cache memory is connected, so as to train the cognitive model based on training examples of a training dataset. The cognitive model is trained by running a hybrid tree building algorithm, so as to construct the decision trees and thereby associate the training examples to leaf nodes of the constructed decision trees, respectively. The hybrid tree building algorithm involves a first routine and a second routine. Each routine is designed to access the cache memory upon execution. The first routine involves a breadth-first search tree builder, while the second routine involves a depth-first search tree builder.

公开(公告)号：US20190146671A1

公开(公告)日：2019-05-16

申请号：US16243956

申请日：2019-01-09

Applicant: International Business Machines Corporation

Inventor： Charles J. Camp ,  Timothy J. Fisher ,  Aaron D. Fry ,  Nikolas Ioannou ,  Ioannis Koltsidas ,  Nikolaos Papandreou ,  Thomas Parnell ,  Roman A. Pletka ,  Charalampos Pozidis ,  Sasa Tomic

IPC: G06F3/06 ,  G11C29/02 ,  G11C16/10 ,  G11C16/14 ,  G11C16/26 ,  G11C16/34

CPC classification number: G06F3/061 ,  G06F3/0655 ,  G06F3/0688 ,  G11C16/10 ,  G11C16/14 ,  G11C16/26 ,  G11C16/3427 ,  G11C29/021 ,  G11C29/028 ,  G11C2029/0409

Abstract: A computer-implemented method according to one embodiment includes determining, after writing data to a non-volatile memory block, one or more delta threshold voltage shift (TVSΔ) values. One or more overall threshold voltage shift values for the data written to the non-volatile memory block are calculated, the values being a function of the one or more TVSΔ values to be used when writing data to the non-volatile memory block. The overall threshold voltage shift values are stored. A base threshold voltage shift (TVSBASE) value, the one or more TVSΔ values, or both the TVSBASE value and the one or more TVSΔ values are re-calibrated during a background health check after a predetermined number of background health checks without calibration are performed.

公开(公告)号：US10236067B2

公开(公告)日：2019-03-19

申请号：US15667473

申请日：2017-08-02

Applicant: INTERNATIONAL BUSINESS MACHINES CORPORATION

Inventor： Timothy J. Fisher ,  Thomas Mittelholzer ,  Nikolaos Papandreou ,  Thomas Parnell ,  Charalampos Pozidis

IPC: G11C11/34 ,  G11C16/26 ,  G11C16/04 ,  G11C16/22

Abstract: A controller adapts the read voltage thresholds of a memory unit in a non-volatile memory. In one embodiment, the controller determines, based on statistics for a memory unit of the non-volatile memory, an operating state of the memory unit from among a plurality of possible operating states and adapts at least one read voltage threshold for a memory cell in the memory unit based on the determined operating state.

公开(公告)号：US10147103B2

公开(公告)日：2018-12-04

申请号：US15468308

申请日：2017-03-24

Applicant: International Business Machines Corporation

Inventor： Celestine Duenner ,  Thomas Parnell ,  Charalampos Pozidis ,  Vasileios Vasileiadis ,  Michail Vlachos

IPC: G06F12/00 ,  G06Q30/02 ,  H04L29/08 ,  G06F13/16 ,  G06N3/04 ,  G06F9/30 ,  G06F9/52 ,  H04L12/801 ,  G06Q50/00

Abstract: Methods and apparatus are provided to determine entities and attributes dependencies for creating recommendations of items or entities using a highly scalable architecture. For example, a user may be recommended an item if a probability model of the method determines that the user relates to the item although the user has no contact to the item before the method is performed. The methods and apparatus provide a data structure representing a matrix having rows representing entities and columns representing attributes of the entities. Each entity of the entities of the data structure may include a user and each attribute of the attributes of the data structure may include an item. A cell of the matrix may be formed by a component pair including an entity and an attribute. In this manner, the methods and apparatus provide an efficient way for processing the probability model.

公开(公告)号：US09996420B2

公开(公告)日：2018-06-12

申请号：US14745515

申请日：2015-06-22

Applicant: International Business Machines Corporation

Inventor： Thomas Mittelholzer ,  Nikolaos Papandreou ,  Thomas Parnell ,  Charalampos Pozidis

IPC: G06F11/00 ,  G06F11/10 ,  H03M13/15 ,  H03M13/00 ,  H03M13/29 ,  G11C29/04 ,  H03M13/11

CPC classification number: G06F11/1076 ,  G06F11/1012 ,  G11C2029/0411 ,  H03M13/1102 ,  H03M13/152 ,  H03M13/2903 ,  H03M13/2909 ,  H03M13/616

Abstract: A data encoding method includes storing K input data symbols; assigning the symbols to respective symbol locations in a notional square array, having n rows and n columns of locations, to define a plurality of k-symbol words in respective rows; encoding the words by encoding rows and columns of the array dependent on a product code having identical row and column codes, each being a reversible error-correction code of dimension k and length n=2n′, thereby to define a codeword, having n2 code symbols corresponding to respective locations of the array, of a quarter product code defined by CQ={X−XT−(X−XT)F: X∈C} where X is an n by n-symbol matrix defining a codeword of the product code, XT is the transpose matrix of X, and (X−XT)F is a reflection of matrix (X−XT) in the anti-diagonal thereof.