Abstract: In one embodiment, a method for training a machine-learning model having multiple parameters includes instantiating trainers each associated with at least a worker thread, a synchronization thread, and a local version of the parameters, using the worker threads to perform training operations that comprise generating an updated local version of the parameters for each trainer using its associated worker thread, while the worker threads are performing training operations, using the synchronization threads to perform synchronization operations that comprise generating a global version of the parameters based on the updated local versions of the parameters and generating a synchronized local version of the parameters for each trainer based on the global version, continuing performing training operations based on the synchronized local versions of the parameters, and determining the parameters at the end of training based on at least a final local version of the parameters associated with one trainer.

Abstract: The present disclosure provides a new scalable coordinate descent (SCD) algorithm and associated system for generalized linear models whose convergence behavior is always the same, regardless of how much SCD is scaled out and regardless of the computing environment. This makes SCD highly robust and enables it to scale to massive datasets on low-cost commodity servers. According to one aspect, by using a natural partitioning of parameters into blocks, updates can be performed in parallel a block at a time without compromising convergence. Experimental results on a real advertising dataset are used to demonstrate SCD's cost effectiveness and scalability.

Abstract: Various aspects of the subject technology relate to systems, methods, and machine-readable media for automatically selecting an input language (e.g., English, Chinese, Japanese, Italian, Russian, French, etc.) for a resource. The resource may be an online resource such as a web page, a web service, or an application. The input language may be selected based on a resource identifier for an online resource, based on content analysis of the online resource, or based on a combination of these methods.

Abstract: The present disclosure provides a new scalable coordinate descent (SCD) algorithm and associated system for generalized linear models whose convergence behavior is always the same, regardless of how much SCD is scaled out and regardless of the computing environment. This makes SCD highly robust and enables it to scale to massive datasets on low-cost commodity servers. According to one aspect, by using a natural partitioning of parameters into blocks, updates can be performed in parallel a block at a time without compromising convergence. Experimental results on a real advertising dataset are used to demonstrate SCD's cost effectiveness and scalability.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for handling faults in a distributed key-value storage system. One of the methods includes receiving an indication that a machine storing a primary replica of a first replication chain is inactive, in response to receiving the indication, generating a concatenated replica comprising a first replica of the first replication chain and a second replica of a second replication chain, the second replication chain comprising replicas of a second key segment, the second key segment being adjacent to the first key segment in the multiple key segments of the plurality of keys, and providing, to another machine in the ordered sequence of machines, a notification of availability of the concatenated replica.

Abstract: An image represented by multiple nodes can be processed by determining whether information can be propagated to a node from another node (e.g., source node) of the image, thereby allowing significantly greater parallelism and scalability by taking advantage of multiprocessing or multi-core processors that are prevalent and widely available today. Conceptually, an image can be presented as a “structured grid” of multiple nodes (e.g., a structured grid of pixels of an image). In a “structured grid,” two or more of the nodes can determine whether to propagate information in parallel. In fact, each node of a “structured grid” can perform operations relating to propagation of information in parallel. This means that for an image of N pixels, it is possible to perform N operations in parallel. It is also possible to divide the processing of N operations for N pixels substantially equally between the number processors or processing cores available at a given time.

Abstract: An image represented by multiple nodes can be processed by determining whether labels can be propagated to a node from another node of the image. Conceptually, an image can be presented as a “structured grid” of multiple nodes (e.g., a structured grid of pixels of an image). In a “structured grid,” two or more nodes of the same level (e.g., nodes in the same gray level) can determine in parallel whether to propagate a label from one or more of its neighboring nodes that are labeled and propagate one or more labels accordingly. An image can be processed by iteratively repeating this process for nodes of successive levels. It will be appreciated that the disclosed techniques allow parallelism without requiring partitioning of an image or having to merge partitioned images. The disclosed techniques are especially suited for watershed algorithms.

Abstract: An image represented by multiple nodes can be processed by determining whether information can be propagated to a node from another node (e.g., source node) of the image, thereby allowing significantly greater parallelism and scalability by taking advantage of multiprocessing or multi-core processors that are prevalent and widely available today. Conceptually, an image can be presented as a “structured grid” of multiple nodes (e.g., a structured grid of pixels of an image). In a “structured grid,” two or more of the nodes can determine whether to propagate information in parallel. In fact, each node of a “structured grid” can perform operations relating to propagation of information in parallel. This means that for an image of N pixels, it is possible to perform N operations in parallel. It is also possible to divide the processing of N operations for N pixels substantially equally between the number processors or processing cores available at a given time.

Abstract: An image represented by multiple nodes can be processed by determining whether labels can be propagated to a node from another node of the image. Conceptually, an image can be presented as a “structured grid” of multiple nodes (e.g., a structured grid of pixels of an image). In a “structured grid,” two or more nodes of the same level (e.g., nodes in the same gray level) can determine in parallel whether to propagate a label from one or more of its neighboring nodes that are labeled and propagate one or more labels accordingly. An image can be processed by iteratively repeating this process for nodes of successive levels. It will be appreciated that the disclosed techniques allow parallelism without requiring partitioning of an image or having to merge partitioned images. The disclosed techniques are especially suited for watershed algorithms.