International Business Machines Corporation

Globalfoundries Inc.

Liana Liyow Fong

Wei Tan

Seetharami R Seelam

Norman Bobroff

David Carl Frank

Linh Hue Lam

Arun Kwangil Iyengar

Michael H Kalantar

John Lewars

Yanbin Liu

Yuan-Chi Chang

Michael H Dawson

Evelyn Duesterwald

Peter Hans Westerink

Lingfei Wu

Shiyu Chang

Liangliang Cao

Ru Fang

Xiaolong Xie

Rhonda L Childress

John Arthur Pershing, Jr

Keun Soo Yim

Mark Steven Squillante

Jason L Crawford

Brian Frank Veale

Zhi Le Zou

John Divirgilio

Srirama Mandyam Krishnakumar

Andrew Julius Greff

David George King

Li Feng Zhang

Parijat Dube

Xiaoqiao Meng

Vijay Krishnarao Naik

Yuqing Gao

Michel H T Hack

Tushar Deepak Chandra

Xavier Rene Guerin

Graeme Johnson

Ajei Sarat Gopal

Donald Philip Pazel

Megumi Ito

Christopher Paul Vignola

William Francis Jerome

Patrick Joseph Brooks

Nayeem Islam

Michael John Witbrock

Toby Burton

Pramod B Nagaraja

Nai K Ling

Andreas Leonidas Prodromidis

Valentina Salapura

Ying Li

Jeffrey Owen Kephart

Yang Zhao

Tamar Eilam

Guerney D H Hunt

Roger E Hough

Fabio Benedetti

Maria Rene Ebling

Stewart Jason Hyman

German Sergio Goldszmidt

Lily Barkovic Mummert

Manoj Khangaonkar

John Joseph Edward Turek

Sameh Afif Fakhouri

Karen Appleby

Prabhat Maddikunta Reddy

Punleuk Oum

Debashish Saha

Steve Weinberger

Ganesh Venkataraman

Archit Verma

Yuan-chi Chang

Peter Hans Westernick

Michael Hilton Dawson

Walter Joseph Grom, Jr

Seelam R Seetharami

Ahmed-Sameh Afif Fakhouri

Husheng Zhou

Techniques regarding code retrieval tasks are provided. For example, one or more embodiments described herein can comprise a system, which can comprise a memory that can store computer executable components. The system can also comprise a processor, operably coupled to the memory, and that can execute the computer executable components stored in the memory. The computer executable components can comprise a code retrieval component that can execute a code retrieval machine learning task by computing an amount of similarity between neural network embeddings of graph representations of a query text and at least a portion of a computer program code.

Semantic code retrieval using graph matching

Techniques that facilitate matrix factorization associated with graphics processing units are provided. In one example, a system includes a first graphics processing unit, a second graphics processing unit and a central processing unit. The first graphics processing unit processes a first data block of a data matrix associated with a matrix factorization system to generate first information for the matrix factorization system. The second graphics processing unit processes a first portion of a second data block of the data matrix separate from a second portion of the second data block to generate second information for the matrix factorization system. The central processing unit processes a machine learning model for the matrix factorization system based on at least the first information provided by the first graphics processing unit and the second information provided by the second graphics processing unit.

Matrix factorization with two-stage data block dispatch associated with graphics processing units

Embodiments relate to a system, program product, and method for supporting elastic execution of a machine learning (ML) workload using application based profiling. A joint profile comprised of both ML application execution and resource usage data is generated. One or more feature(s) and signature(s) from the joint profile are identified, and a ML execution model for ML application execution and resource usage is built. The ML execution model leverages the feature(s) and signature(s) and is applied to provide one or more directives to subsequent application execution. The application of the ML execution model supports and enables the ML execution to elastically allocate and request one or more resources from a resource management component, with the elastic allocation supporting application execution.

Elastic execution of machine learning workloads using application based profiling

Techniques for generating a grounded video description for a video input are provided. Hierarchical Attention based Spatial-Temporal Graph-to-Sequence Learning framework for producing a GVD is provided by generating an initial graph representing a plurality of object features in a plurality of frames of a received video input and generating an implicit graph for the plurality of object features in the plurality of frames using a similarity function. The initial graph and the implicit graph are combined to form a refined graph and the refined graph is processed using attention processes, to generate an attended hierarchical graph of the plurality of object features for the plurality of frames. The grounded video description is generated for the received video input using at least the hierarchical graph of the plurality of features.

Spatial-temporal graph-to-sequence learning based grounded video descriptions

A method, system and computer program product for index maintenance in a computer system comprising a plurality of nodes, a database, and an index to the database. In one embodiment, the method comprises, for a defined period of time, building a snapshot of selected change requests received by the nodes to change the database. After this defined period of time, a selection is made, based on specified criteria, whether to rebuild a new index to the database, or to add entries to a current index. When the selection is to rebuild a new index, the new index is rebuilt based on data in the database and in the change requests in the snapshot. When the selection is to add entries to a current index, entries are added to the current index based on data in the database and in the change requests in the snapshot.

Index maintenance based on a comparison of rebuild vs. update

Techniques that facilitate matrix factorization associated with graphics processing units are provided. In one example, a computer-implemented method is provided. The computer-implemented method can comprise loading, by a graphics processing unit operatively coupled to a processor, item features from a data matrix into a shared memory. The data matrix can be a matrix based on one or more user features and item features. The computer-implemented method can further comprise tiling and aggregating, by the graphics processing unit, outer products of the data matrix tiles to generate an aggregate value and approximating, by the graphics processing unit, an update to a user feature of the data matrix based on the aggregate value and the loaded item features.

Matrix factorization with approximate computing

A random binning featurization process method, system, and computer program product for a distributed random binning featurization process on one or more multicore systems with a hybrid two-level parallelism, the method including in a training phase, receiving a first data matrix dividing the random binning featurization process into two orthogonal levels, in a high-level generating a randomized number of high-dimension grids and evenly partitioning the grids into nodes in a parallel system, and in a low-level, evenly partitioning dimensions in each grid to construct look-up tables of index vectors and compute a local feature matrix for each node.

Distributed random binning featurization with hybrid two-level parallelism

A workload scheduling method, system, and computer program product include analyzing a resource scheduling requirement for processes of a workload including the communication patterns among CPUs and accelerators, creating feasible resources based on static resource information of the resources for the processes of the workload, and selecting an available resource of the feasible resources to assign the workload based on the resource scheduling requirement, such that the CPU and GPU connection topology of the selection matches the communication patterns of the workload.

Coordinated, topology-aware CPU-GPU-memory scheduling for containerized workloads

A block coordinate descent method, system, and computer program product for partitioning a global feature matrix into blocks, each node of the nodes of the blocks having a block size of a number of the blocks over a number of the nodes, selecting, at each node, a subset of the blocks from the blocks, and in one of the nodes, launching a thread to simultaneously update a closed-form solution by minimizing a single coordinate in one of the blocks.

High-performance block coordinate based on L1 regularized random binning

A method, information processing system, and computer program product are provided for managing operating system interference on applications in a parallel processing system. A mapping of hardware multi-threading threads to at least one processing core is determined, and first and second sets of logical processors of the at least one processing core are determined. The first set includes at least one of the logical processors of the at least one processing core, and the second set includes at least one of a remainder of the logical processors of the at least one processing core. A processor schedules application tasks only on the logical processors of the first set of logical processors of the at least one processing core. Operating system interference events are scheduled only on the logical processors of the second set of logical processors of the at least one processing core.

Migrating operating system interference events to a second set of logical processors along with a set of timers that are synchronized using a global clock

An alternating least square recommendation method, system, and non-transitory computer readable medium, include partially updating a user's feature when an update ratio is less than a pre-defined threshold ratio, the pre-defined threshold ratio being configurable based on a preference of an existing sparse matrix factorization, where the pre-defined threshold ratio is pre-defined by setting the pre-defined threshold ratio to variable values according to a past acceptable update.

System, method, and recording medium for differentiated and partial feature update in alternating least square

A topology-aware parallel reduction method, system, and recording medium including obtaining the GPU connection topology of each of the plurality of GPUs as a connection tree, transforming the connection tree into a three layer tree comprising an intra-root tree, an intra-node tree, and an inter-node tree, evenly partitioning each entry on each of the GPUS, and selectively transferring data either in either direction or in each direction, simultaneously, along the evenly partitioned three layer tree using a full-duplex configuration of a PCIe bandwidth.

Topology-aware parallel reduction in an accelerator

Growing community of inventors

Yorktown Heights, NY, United States of America

Liana Liyow Fong