Intel Corporation

Todd A Anderson

David M Putzolu

Hongbo Rong

Justin E Gottschlich

Cheng C Wang

Youfeng Wu

Ignacio Javier Alvarez

Tatiana Shpeisman

Victor W Lee

Javier Felip Leon

Javier Perez-Ramirez

Sridhar G Sharma

Erik J Johnson

Paul M Petersen

Arch D Robison

Aaron R Kunze

Sharad K Garg

Mikael Bourges-Sevenier

Adam Herr

Vy Vo

Derek Gerstmann

Peng Tu

Jongsoo Park

Hai Liu

P G Lowney

Methods, apparatus, systems and articles of manufacture to optimize execution of a machine learning model are disclosed. An example apparatus includes a quantizer to quantize a layer of a model based on an execution constraint, the layer of the model represented by a matrix. A packer is to pack the quantized layer of the matrix to create a packed layer represented by a packed matrix, the packed matrix having non-zero values of the matrix grouped together along at least one of a row or a column of the matrix. A blocker is to block the packed layer into a blocked layer by dividing the non-zero values in the packed matrix into blocks. A fuser is to fuse the blocked layer into a pipeline. A packager is to package the pipeline into a binary.

Methods and apparatus to optimize execution of a machine learning model

Systems, apparatuses and methods may provide for technology that obtains categorization information and corresponding uncertainty information from a perception subsystem, wherein the categorization information and the corresponding uncertainty information are to be associated with an object in an environment. The technology may also determine whether the corresponding uncertainty information satisfies one or more relevance criteria, and automatically control the perception subsystem to increase an accuracy in one or more subsequent categorizations of the object if the corresponding uncertainty information satisfies the one or more relevance criteria. In one example, determining whether the corresponding uncertainty information satisfies the relevance criteria includes taking a plurality of samples from the categorization information and the corresponding uncertainty information, generating a plurality of actuation plans based on the plurality of samples, and determining a safety deviation across the plurality of actuation plans, wherein the relevance criteria are satisfied if the safety deviation exceeds a threshold.

Technology to handle ambiguity in automated control systems

Technologies for automatic reordering of sparse matrices include a computing device to determine a distributivity of an expression defined in a code region of a program code. The expression is determined to be distributive if semantics of the expression are unaffected by a reordering of an input/output of the expression. The computing device performs inter-dependent array analysis on the expression to determine one or more clusters of inter-dependent arrays of the expression, wherein each array of a cluster of the one or more clusters is inter-dependent on each other array of the cluster, and performs bi-directional data flow analysis on the code region by iterative backward and forward propagation of reorderable arrays through expressions in the code region based on the one or more clusters of the inter-dependent arrays. The backward propagation is based on a backward transfer function and the forward propagation is based on a forward transfer function.

Technologies for automatic reordering of sparse matrices

Electronic garbage collection moves objects within memory to consolidate the objects thereby reducing access time and improving memory performance. Transactions occurring in an atomic transactional memory appear to occur instantaneously&#x2014;such that the transaction completes in its entirety or is aborted. A garbage collection circuit attempts to move a memory object from a first memory location to a second memory location using a transactional fast-path move operation. If the transactional fast-path move operation is unsuccessful after a number of attempts, the garbage collection circuit attempts to move the object using a non-transactional slow-path move.

Concurrent, moving, garbage collector

Technologies for generating composable library functions include a first computing device that includes a library compiler configured to compile a composable library and second computing device that includes an application compiler configured to compose library functions of the composable library based on a plurality of abstractions written at different levels of abstractions. For example, the abstractions may include an algorithm abstraction at a high level, a blocked-algorithm abstraction at medium level, and a region-based code abstraction at a low level. Other embodiments are described and claimed herein.

Technologies for low-level composable high performance computing libraries

Network bundles may be processed in a distributed network having a decentralized serving structure. The message bundles may be modified to include a client address. Additionally, each message bundle comprises a plurality of sub-messages, and each sub-message may contain either a link to the output of another sub-message, or a network address. A network device may be implemented to gather responses to the sub-messages from various servers and to organize the responses into a final response to send to the client.

Method, apparatus and system for processing message bundles on a network

A system and method of transmitting network packets between network processing elements through links are disclosed. One or more configuration entities may allocate one or more fixed length slots to be appended to network packets forwarded on a link between network processing elements in-band of the link.

System and method of annotating network packets

Variable write back delay apparatus, methods and articles of manufacture are disclosed. The disclosed variable write back delay apparatus, methods and articles of manufacture identify a probability associated with a loss of a file and determine the write back delay based on the probability associated with the loss of the file. The disclosed apparatus, methods and articles of manufacture associate a write back delay with each file that results in a minimal expected total time cost of recreating and storing that file.

Variable write back delay apparatus and methods

In a distributed system, a server may distribute a primary object copy having one or more chunks of data associated with a primary per-chunk metadata copy to a client over a network. The client may selectively fetch and update the one or more chunks of data of the primary object copy based on the primary per-chunk metadata copy. For example, by determining at a client whether a portion of the data in a cached data object, such as a file, is changed relative to a copy of a data object for the cached data object at a server, less than all of the data of the cached data object may be replaced with a corresponding data in the copy of the data object.

Reducing communication for reads and updates in distributed object systems

An apparatus and method for secure, automated response to distributed denial of service (DDoS) attacks are described. The method includes notification of a DDoS attack received by an Internet host. Once received by an Internet host, the Internet host establishes security authentication from an upstream router from which the attack traffic, transmitted by one or more host computers, is received. The Internet host then transmits filter(s) to the upstream router generated based upon characteristics of the attack traffic. Once installed by the upstream router, the attack traffic is dropped to terminate a DDoS attack. In addition, the router may determine upstream router(s) coupled to ports from which attack traffic is received, and securely forward the filter(s) to the upstream routers as a routing protocol updated in order to drop the attack traffic at a point closer to a source of the DDoS attack.

Growing community of inventors

Beaverton, OR, United States of America

Todd A Anderson