Microsoft Technology Licensing, LLC

Yuxiong He

Sameh Elnikety

Kathryn S Mckinley

Shaolei Ren

Minjia Zhang

Wenhan Wang

Jianfeng Gao

Johnson T Apacible

James R Larus

Trishul A Chilimbi

Xiaodong Liu

Karthik Kalyanaraman

Yutaka Suzue

Olatunji Ruwase

Feng Yan

Chenyu Yan

Sherif Sakr

Samyam Rajbhandari

A processing system is described which assigns jobs to heterogeneous processing modules. The processing system assigns jobs to the processing modules in a manner that attempts to accommodate the service demands of the jobs, but without advance knowledge of the service demands. In one case, the processing system implements the processing modules as computing units that have different physical characteristics. Alternatively, or in addition, the processing system may implement the processing modules as threads that are executed by computing units. Each thread which runs on a computing unit offers a level of performance that depends on a number of other threads that are simultaneously being executed by the same computing unit.

Assigning jobs to heterogeneous processing modules

Systems, methods, and computer-executable instructions for determining a computation schedule for a recurrent neural network (RNN). A matrix multiplication (MM) directed-acyclic graph (DAG) is received for the RNN. Valid phased computation schedules for the RNN are generated. Each of the valid phase computation schedule includes an ordering of MM operations. For each of the plurality of valid phased computation schedules, each of the MM operations is partitioned to processor cores based on L3 cache to L2 cache data movement. The RNN is executed based on the valid phased computation schedules. A final computation schedule is stored. The final computation schedule is used for future executions of the RNN.

Deep learning model scheduling

A method for semantic search includes receiving a query vector including a semantic feature value for each of a plurality of semantic feature dimensions. A cluster is selected from a plurality of different candidate clusters held in a relatively fast memory, each candidate cluster including a plurality of compressed answer vectors. A subset of the plurality of compressed answer vectors are promoted as candidate answers. For each of the candidate answers, a corresponding uncompressed answer vector is retrieved from a relatively slower memory. A selected answer is promoted from among the candidate answers.

Multi-layer semantic search

A performance investigation tool (PIT) is described herein for investigating the performance of a distributed processing system (DPS). The PIT operates by first receiving input information that describes a graph processing task to be executed using a plurality of computing units. The PIT then determines, based on the input information, at least one time-based performance measure that describes the performance of a DPS that is capable of performing the graphical task. More specifically, the PIT can operate in a manual mode to explore the behavior of a specified DPS, or in an automatic mode to find an optimal DPS from within a search space of candidate DPSs. A configuration system may then be used to construct a selected DPS, using the plurality of computing units. In one case, the graph processing task involves training a deep neural network model having a plurality of layers.

Tool for investigating the performance of a distributed processing system

Systems, methods, and computer-executable instructions for approximating a softmax layer are disclosed. A small world graph that includes a plurality of nodes is constructed for a vocabulary of a natural language model. A context vector is transformed. The small world graph is searched using the transformed context vector to identify a top-K hypothesis. A distance from the context vector for each of the top-K hypothesis is determined. The distance is transformed to an original inner product space. A softmax distribution is computed for the softmax layer over the inner product space of the top-K hypothesis. The softmax layer is useful for determining a next word in a speech recognition or machine translation.

Graph representations for identifying a next word

The subject disclosure is directed towards scheduling requests using quality values that are defined for partial responses to the requests. For each request in a queue, an associated processing time is determined using a system load and/or the quality values. The associated processing time is less than or equal to a service demand, which represents an amount of time to produce a complete response.

Scheduling execution requests to allow partial results

Graph queries are processed using a plurality of independent query execution engines. A graph query submitted to a graph database which is modeled by an attributed graph is received. The graph query is decomposed into a plurality of query components. For each of the query components, a one of the query execution engines that is available to process the query component is identified, a sub-query representing the query component is generated, the sub-query is sent to the identified query execution engine for processing, and results for the sub-query are received from the identified query execution engine. The results received are then combined to generate a response to the graph query.

Growing community of inventors

Bellevue, WA, United States of America

Yuxiong He