Elasticsearch B.v.

Deepmind Technologies Limited

Thomas Dullien

Sean Heelan

Oriol Vinyals

Yujia Li

Chenjie Gu

Pushmeet Kohli

In various embodiments, the system and method described herein provide functionality for selecting an appropriate compression algorithm and settings given a cost model. Specifically, in selecting a compression method and configuration, the described system and method use a cost model to take into account the financial cost of a number of aspects of a particular compression scenario, including, but not limited to, the cost of performing the compression/decompression and the cost of storing the data. In this manner, intelligent trade-offs can be made between CPU/computing cost and data storage/transmission cost in an environment where a dollar amount can be associated with CPU processing time and storage/transmission volume. The described system and method can make such decisions dynamically, so that compression and/or decompression operations can respond to changing conditions on the fly, thus leading to better and more cost-effective management of resources.

Selecting data compression parameters using a cost model

There is described a neural network system implemented by one or more computers for determining graph similarity. The neural network system comprises one or more neural networks configured to process an input graph to generate a node state representation vector for each node of the input graph and an edge representation vector for each edge of the input graph; and process the node state representation vectors and the edge representation vectors to generate a vector representation of the input graph. The neural network system further comprises one or more processors configured to: receive a first graph; receive a second graph; generate a vector representation of the first graph; generate a vector representation of the second graph; determine a similarity score for the first graph and the second graph based upon the vector representations of the first graph and the second graph.

Deep neural network system for similarity-based graph representations

Functionality is provided for unwinding program call stacks across native-to-interpreted code and native-to-JIT-compiled code boundaries, as well as across the kernel and user space boundaries, during performance profiling. The system thus enables profiling of code that crosses boundaries from native code to interpreted languages and native code to languages that run on a runtime supporting JIT compilation. Various embodiments provide cross-language profiling with a sufficiently low performance impact so as to enable such profiling to take place in a production environment.

Unwinding program call stacks for performance profiling

Functionality is provided for profiling code by unwinding stacks in frame-pointer omitted executables using C++ exception stack unwinding information. Information is extracted from executable files, and used to optimize stack unwinding operations. In at least one embodiment, the system uses information that has been included for exception handling. Storage of such information can be optimized by exploiting patterns in stack deltas.

Growing community of inventors

Zurich, Switzerland

Thomas Dullien