International Business Machines Corporation

Keith W Campbell

Berthold Reinwald

Shirish Tatikonda

Alexandre Valentinovich Evfimievski

Matthias Boehm

John D Keenleyside

Arash Ashari

Rohit Bahl

Alan Yeung

A method for optimization of machine learning (ML) workloads on a graphics processor unit (GPU). The method includes identifying a computation having a generic pattern commonly observed in ML processes. Hierarchical aggregation spanning a memory hierarchy of the GPU for processing is performed for the identified computation including maintaining partial output vector results in shared memory of the GPU. Hierarchical aggregation for vectors is performed including performing intra-block aggregation for multiple thread blocks of a partial output vector results on GPU global memory.

Pipelined approach to fused kernels for optimization of machine learning workloads on graphical processing units

A method for optimization of machine learning (ML) workloads on a graphics processor unit (GPU). The method includes identifying a computation having a generic pattern commonly observed in ML processes. An optimized fused GPU kernel is employed to exploit temporal locality for inherent data-flow dependencies in the identified computation. Hierarchical aggregation spanning a memory hierarchy of the GPU for processing for the identified computation is performed. GPU kernel launch parameters are estimated following an analytical model that maximizes thread occupancy and minimizes atomic writes to GPU global memory.

Systems and methods for conversion and importation of models that describe system behavior into a UML meta model-based representation, include parsing through the textual model for the plurality of elements, searching for an element semantic definition, element view definitions corresponding to a semantic definition, or an element view containing diagram definition within the textual model for each of the plurality of elements, generating element reference nodes for placement on an internally constructed custom tree, attaching a listener to each of the element reference nodes, wherein the listener is configured to await population of the element reference node with an equivalent unified modeling language semantic element, wherein a listener awaiting population is an awaiting sequenced listener, completing an inheritance hierarchy between the element reference nodes up to a parent node inferred from the diagramming definitions and resolving awaiting sequenced listeners that are made aware of an awaited unified modeling language value.

Single parse, diagram-assisted import into a unified modeling language based meta-model

Growing community of inventors

Ottawa, Canada

Keith W Campbell