International Business Machines Corporation

Augusto Javier Vega

Alper Buyuktosunoglu

Pradip Bose

Hans Mikael Jacobson

Timothy Joseph Chainer

Hubertus Franke

Pritish Ranjan Parida

John-David Wellman

Vaidyanathan Srinivasan

Karthik V Swaminathan

Ranjal Gautham Shenoy

Aporva Amarnath

Chen-Yong Cher

Robert Matthew Senger

Ramon Bertran Monfort

William J Song

Liang Wang

Karthick Rajamani

Michael Stephen Floyd

Ravi Nair

Bruce Martin Fleischer

Thomas Winters Fox

Srinivasan Ramani

Todd Jon Rosedahl

Swagath Venkataramani

Heather Lynn Hanson

Schuyler Eldridge

Parth Sanjaybhai Shah

Saketh V Rama

In a first device, a local classification and a local classification confidence score corresponding to an event input are computed. At the first device in response to a broadcast request, a remote classification and a remote classification confidence score corresponding to the event input are received, the remote classification and the remote classification confidence score being computed at a second device. At the first device, a consensus classification including the most frequent classification from a set of all received remote classifications and the local classification is formed, provided the number of classifications including the most frequent classification exceeds a threshold. In response to a consensus classification confidence score corresponding to the consensus classification exceeding a confidence threshold, a local classification model is updated. Based on the local classification and the consensus classification, the event input is assigned to a classification.

Adaptive multi-agent cooperative computation and inference

Tasks of directed acyclic graphs (DAGs) may be dynamically scheduled based on a plurality of constraints and conditions, task prioritization policies, task execution estimates, and configurations of a heterogenous system. A machine learning component may be initialized to dynamically schedule the tasks of the DAGs.

Learning agent based application scheduling

Described are techniques for scheduling tasks on a heterogeneous system on a chip (SoC). The techniques including receiving a directed acyclic graph at a meta pre-processor associated with a heterogeneous system-on-chip and communicatively coupled to a scheduler, where the directed acyclic graph corresponds to a control flow graph of respective tasks associated with an application executed by the heterogeneous system-on-chip. The techniques further including determining, using a learning agent implementing machine learning algorithms, a rank for a respective task in the directed acyclic graph, wherein the learning agent receives as input the directed acyclic graph, constraints associated with the directed acyclic graph, and heuristics regarding previously completed tasks. The techniques further including providing the respective task to the scheduler for execution on the heterogeneous system-on-chip according to the rank.

Heterogeneous system on a chip scheduler with learning agent

Systems, computer-implemented methods and/or computer program products that facilitate management of resources are provided. In one embodiment, a computer-implemented method comprises: employing, by a system operatively coupled to a processor, at least one model to predict respective token needs by a set of processing elements during execution of a workload; and exchanging, by the system, one or more tokens between a subset of the processing elements as a function of the predicted token needs.

Swarm-based resource management

Described are techniques for scheduling tasks on a heterogeneous system on a chip (SoC). The techniques including receiving a directed acyclic graph at a meta pre-processor associated with a heterogeneous SoC and communicatively coupled to a scheduler, wherein the directed acyclic graph corresponds to a control flow graph of tasks associated with an application executed by the heterogeneous SoC. The techniques further including determining a rank for a respective task in the directed acyclic graph, wherein the rank is based on a priority of the respective task and a slack in the directed acyclic graph. The techniques further including providing the respective task to the scheduler for execution on the heterogeneous SoC according to the rank.

Heterogeneous system on a chip scheduler

Various embodiments are provided for load balancing of machine learning operations in a computing environment by a processor. One or more machine learning operations performing inference or training operations may by dynamically balanced between one or more edge computing devices in a wireless communication network and a cloud computing system for increasing performance of a selected metric.

Power efficient machine learning in cloud-backed mobile systems

An N modular redundancy method, system, and computer program product include a computer-implemented N modular redundancy method for neural networks, the method including selectively replicating the neural network by employing one of checker neural networks and selective N modular redundancy (N-MR) applied only to critical computations.

Reducing the cost of n modular redundancy for neural networks

A computer-implemented method optimizes a neural network. One or more processors define layers in a neural network based on neuron locations relative to incoming initial inputs and original outgoing final outputs of the neural network, where a first defined layer is closer to the incoming initial inputs than a second defined layer, and where the second defined layer is closer to the original outgoing final outputs than the first defined layer. The processor(s) define parameter criticalities for parameter weights stored in a memory used by the neural network, and associate defined layers in the neural network with different memory banks based on the parameter criticalities for the parameter weights. The processor(s) store parameter weights used by neurons in the first defined layer in the first memory bank and parameter weights used by neurons in the second defined layer in the second memory bank.

Parameter criticality-aware resilience

A method, computer program product, and/or system associates a number of tokens with a plurality of frequency domains (for example, cores) of a central processing unit (CPU) computer chip. The number of tokens allotted to the CPU is based on the CPU power budget. Cores are organized as a ring topology. A token pool traverses the ring, picks up excess tokens from cores having excess tokens, and gives the tokens to cores that need additional tokens. Tokens acquired by a core allows the core to increase operating frequency by an increment represented by the tokens. Consequently, power usage is weighted toward heavily loaded cores and away from lightly loaded cores. Overall power usage of the CPU remains within a power budget. The method budgets power optimally to sustain turbo frequencies for longer durations by not allowing control units to increase frequency in absence of any useful high frequency benefiting workload.

Governing power budget with token passing

Embodiments of the present invention include a system, computer-implemented method, and a computer program product. A non-limiting example of the method includes a processor utilizing a model having a plurality of parameters. The processor compares a current value of a model parameter to a prior value of the model parameter. Based at least in part on comparing the current value of the model parameter to the prior value of the model parameter, a determination is made that the model being utilized by the processor has changed. The current value of the model parameter is transmitted by the processor.

Model training by discarding relatively less relevant parameters

A computer system that has two or more processing engines (PE), each capable of performing one or more operations on one or more operands but one or more of the PEs performs the operations unreliably. Initial results of each operation are debiased to create a debiased result used by the system instead of the initial result. The debiased result has an expected value equal to a correct output where the correct output is the initial result the respective operation would have produced if the respective operation performed was reliable.

Computing with unreliable processor cores

Techniques for decoupling cognitive model training from execution of the cognitive model are provided. In one example, a computer program product is provided that determines cognitive data based on context data and a model of interpreting the context data. The cognitive data can comprise prediction data that represents a prediction relating to a state of an environment. The context data and the cognitive data can be transmitted to a server, and an updated model can be received in response.

Cognitive computing for servers and mobile devices

An aspect includes optimizing an application workflow. The optimizing includes characterizing the application workflow by determining at least one baseline metric related to an operational control knob of an embedded system processor. The application workflow performs a real-time computational task encountered by at least one mobile embedded system of a wirelessly connected cluster of systems supported by a server system. The optimizing of the application workflow further includes performing an optimization operation on the at least one baseline metric of the application workflow while satisfying at least one runtime constraint. An annotated workflow that is the result of performing the optimization operation is output.

Optimization of application workflow in mobile embedded devices

Techniques for inducing heterogeneous microprocessor behavior using non-uniform cooling are described. According to an embodiment, a device is provided that comprises an IC chip comprising a plurality of cores and a cooling apparatus coupled to the integrated chip that cools the integrated chip in association with electrical operation of the plurality of cores. The cooling apparatus cools a first core of the plurality of cores to a lower temperature than a second core of the plurality of cores. In various embodiments, the cooling apparatus comprises a plurality of channels embedded within the integrated chip and the cooling apparatus cools the integrated chip via flow of liquid coolant through the plurality of channels.

Growing community of inventors

Astoria, NY, United States of America

Augusto Javier Vega