Task (computing)

About: Task (computing) is a research topic. Over the lifetime, 9718 publications have been published within this topic receiving 129364 citations.

CrossTransformers: spatially-aware few-shot transfer

Abstract: Given new tasks with very little data$-$such as new classes in a classification problem or a domain shift in the input$-$performance of modern vision systems degrades remarkably quickly. In this work, we illustrate how the neural network representations which underpin modern vision systems are subject to supervision collapse, whereby they lose any information that is not necessary for performing the training task, including information that may be necessary for transfer to new tasks or domains. We then propose two methods to mitigate this problem. First, we employ self-supervised learning to encourage general-purpose features that transfer better. Second, we propose a novel Transformer based neural network architecture called CrossTransformers, which can take a small number of labeled images and an unlabeled query, find coarse spatial correspondence between the query and the labeled images, and then infer class membership by computing distances between spatially-corresponding features. The result is a classifier that is more robust to task and domain shift, which we demonstrate via state-of-the-art performance on Meta-Dataset, a recent dataset for evaluating transfer from ImageNet to many other vision datasets.

Computing platform security methods and apparatus

Abstract: Computing platform security methods and apparatus are disclosed. An example apparatus includes a security application to configure a security task, the security task to detect a malicious element on a computing platform, the computing platform including a central processing unit and a graphics processing unit; and an offloader to determine whether the central processing unit or the graphics processing unit is to execute the security task; and when the graphics processing unit is to execute the security task, offload the security task to the graphics processing unit for execution.

Managing write requests in cache directed to different storage groups

Abstract: Provided are a method, system, and article of manufacture for managing write requests in cache directed to different storage groups. A determination is made of a high and low thresholds for a plurality of storage groups configured in a storage, wherein the high and low thresholds for one storage group indicate a high and low percentage of a cache that may be used to store write requests to the storage group. A determination is made of a number of tasks to assign to the storage groups based on the determined high and low thresholds for the storage groups, wherein each task assigned to one storage group destages write requests from the cache to the storage group.

Single-keyed indexed file for TP queue repository

Abstract: A task queue is structured as a single-keyed indexed file in which the key has a most significant portion indicating a priority level and a less significant portion that is ordered with the loading of the tasks into the queue. For any given task record in the queue, the less significant portion of the key is determinable from a respective task identifier. Preferably the less significant portion of the key is a "time stamp" including the current date when the task was created and a representation of the data processor's internal 24 hour time clock, and a task identification number is formed by appending a node number to the time stamp in the event that the system has multiple processors capable of creating different tasks at the same time. This format of the single key causes an internal ordering of the records in the queue that is sequential with respect to the less significant portion of the key within blocks of records having the same priority. Therefore, due to the relationship between the key and the task identification number for each task record in the queue, it is possible to quickly search for the record having a requested task identification number. Conventional memory management facilities for accessing key-indexed files can be used for searching the queue. In this case the queue is searched by random or "key next" access to repeatedly step through the possible priority levels until a record having a matching key is found or the end of file is reached. When a record having a matching key is found, the requested task identification number is compared to the identification field of the record. If there is a match, the desired record has been found. If not, then searching continues by sequential access until there is a match of the identification numbers or until the keys no longer match.

Task Backpressure and Deletion in a Multi-Flow Network Processor Architecture

Abstract: Described embodiments generate tasks corresponding to packets received by a network processor. A source processing module sends task messages including a task identifier and a task size to a destination processing module. The destination module receives the task message and determines a queue in which to store the task. Based on a used cache counter of the queue and a number of cache lines for the received task, the destination module determines whether the queue has reached a usage threshold. If the queue has reached the threshold, the destination module sends a backpressure message to the source module. Otherwise, if the queue has not reached the threshold, the destination module accepts the received task, stores data of the received task in the queue, increments the used cache counter for the queue corresponding to the number of cache lines for the received task, and processes the received task.