9results about How to "Reduce bottlenecks" patented technology

The application discloses a data read-write method and device based on cooperation of DDR and SPI interfaces, and the method comprises the following steps: acquiring state data sets of the interfaces, data transmission priority requirements, real-time bus access states and system timing coordination requirements; performing load evaluation according to the state data sets to obtain a classification result, combining the data transmission priority requirements to calculate the emergency degree of bus access, and obtaining a priority sequence; performing time slice allocation according to the priority sequence and the real-time bus access states to obtain an access right allocation scheme; executing the access right allocation scheme to generate a resource conflict resolution strategy; performing task re-allocation according to the resource conflict resolution strategy and the system timing coordination requirements to obtain a read-write task re-allocation result; monitoring the read-write task re-allocation result to obtain interface state change information, and performing optimization according to the read-write task re-allocation result and the interface state change information to obtain a multi-interface cooperative working configuration.

This invention relates to the field of artificial intelligence technology and discloses a method for constructing a rewriting model based on a mixed-precision strategy. The method includes instantiating computation nodes in a computation flow graph into multiple intelligent computing units. Each intelligent computing unit maintains a local state vector containing its current computational precision and encapsulates a rule engine. The rule engine performs local inference and generates impact messages containing adjustment proposals. When impact messages are transmitted between intelligent computing units, distributed negotiation based on these messages is triggered, and the computational precision of the corresponding intelligent computing units is collaboratively adjusted according to the negotiation results. A central policy controller collects global state information from the computation flow graph and dynamically optimizes the inference rules used by the rule engine based on this global state information. Through a dual-layer driving approach of local inference and global layer optimization, the mixed-precision strategy and the computation flow graph are dynamically matched and optimized, balancing the allocation of computational precision and hardware resources.

This invention discloses a front-end hardware acceleration system, a program proof hardware acceleration system, and a method. Addressing the problem that existing zero-knowledge virtual machine front-ends still heavily rely on software implementation, this invention proposes a dedicated heterogeneous hardware architecture for front-end execution, main trajectory generation, and auxiliary trajectory generation. This significantly reduces front-end bottlenecks. Furthermore, the main trajectory unit, trajectory buffer unit, and auxiliary trajectory unit work collaboratively, forming a unified on-chip processing link for program execution, main trajectory generation, and auxiliary trajectory generation. On-chip collaborative processing of the main and auxiliary trajectories involves direct on-chip data transfer through the trajectory buffer unit, avoiding the need to read all main trajectories back from main memory and reducing the round-trip overhead of "main trajectory in main memory—software read back—regeneration of auxiliary trajectories." Based on the front-end hardware acceleration system, further through units such as trajectory shaping, polynomial processing, constraint combination, commitment generation, low-degree testing, query initiation, and proof encapsulation, the front-end output is directly extended to the back-end proof construction, reducing stage boundary overhead and forming an integrated end-to-end hardware processing architecture from "program execution - trajectory generation - commitment construction - low-degree testing - query initiation - proof encapsulation". Moreover, through the trajectory shaping unit, the front-end output can be directly connected to the back-end proof construction, reducing additional data format conversion overhead.

The invention provides a data reconstruction method, a distributed storage system and a computer readable storage medium, and is applied to the technical field of distributed storage. The data reconstruction method is applied to a proxy repair service, the proxy repair service is deployed at a persistence layer client of a distributed storage cluster, the distributed storage cluster comprises a plurality of nodes, and each node is configured with the persistence layer client and an object storage server. According to the method, the proxy repair service is applied in the distributed storage cluster, local loopback communication between the object storage server and the persistence layer client is realized, and cross-node network data transmission is remarkably reduced by reading local source fragmented data and executing erasure code reconstruction, so that the data reconstruction network bottleneck in a large-scale EC scene is effectively relieved, and the data reconstruction efficiency is improved. And the data reconstruction efficiency and the overall availability of the cluster are improved.

This invention discloses a method and system for reusing discarded convolutional information in small-sample fine-grained image classification, belonging to the field of computer vision. The method includes the following steps: acquiring images of a support set and a query set, and extracting the backbone feature descriptors of the images using a convolutional neural network; after at least one convolutional block in the convolutional neural network, retrieving the discarded feature information of the convolutional block through a feature retrieval path to generate a retrieved feature descriptor; fusing the backbone feature descriptor with the retrieved feature descriptor to obtain a high-dimensional feature descriptor in a unified format; calculating the similarity between the images of the query set and the support set based on the high-dimensional feature descriptor, and performing small-sample fine-grained image classification based on the similarity, thereby realizing the reuse of discarded convolutional information. This invention can effectively retrieve discarded feature information in convolutional neural networks, improving the performance of small-sample fine-grained image classification.

The application provides a UWB positioning method, system, electronic equipment and computer readable storage medium, aiming at improving the problems of channel congestion, limited capacity and insufficient scalability of the existing UWB system in large-scale deployment and multi-anchor cooperative ranging. The method part comprises: based on the local tag lookup table, a ranging request is sent to each to-be-ranged tag in the local tag lookup table in turn to obtain ranging information corresponding to each to-be-ranged tag; after the ranging request is sent to the local tag lookup table, a networking request is broadcasted, and the address of the tag sending the networking reply is added to the local tag lookup table based on the received networking reply to update the local tag lookup table; the obtained ranging information is reported to the positioning server, and the next target anchor point in the network is determined according to the local anchor lookup table, and the ranging right is handed over to the next target anchor point.

The utility model relates to the technical field of quartz sand drying, in particular to high-purity quartz sand drying equipment which comprises a drying tank, a stirring mechanism is arranged along the central axis, a heating layer covering the outer surface of the drying tank is arranged on the periphery of the drying tank, continuous heat is provided, the stability of the drying temperature is guaranteed, and heat transfer is even. And non-uniform drying of the quartz sand due to too high or too low temperature in the drying process is avoided. The feeding assembly comprises a plurality of feeding hoppers arranged side by side, and feeding is conducted along each feeding hopper, so that the flow of materials entering the drying tank is effectively controlled. Quartz sand is gradually and uniformly added into the drying tank, it is ensured that the quartz sand can be stirred and uniformly distributed once entering the drying tank, uneven drying caused by adding excessive materials at a time is avoided, and material accumulation or caking is avoided. And the dried quartz sand is rapidly discharged in a vacuum mode, and the drying efficiency of the quartz sand is improved through timely discharging. Partial or full automatic control is achieved, the heat transfer efficiency is high, and energy consumption is reduced.

The invention discloses a man-machine cooperation calibration method, system and device and a storage medium, and aims to solve the problem that noise data generated by manual recheck pollutes an artificial intelligence model. The method comprises the following steps: establishing a dynamic credibility model for a worker and calculating a current credibility score; when man-machine judgment generates difference, carrying out shunting processing on difference cases based on credibility scores, and distributing the difference cases to a direct adoption or arbitration path; carrying out weighted training on the artificial intelligence model according to the training weight mapped by the credibility score on the corrected data which are confirmed to be correct; and establishing a personal error semantic vector library by the correction cases of which the errors are confirmed, and performing similarity matching when a new task is processed so as to push an early warning in real time. According to the invention, noise data can be suppressed, the expert resource bottleneck can be relieved, and accurate real-time tutoring of personnel can be realized.

The invention discloses a contract recursion method for distributed tensor data updating, and relates to the technical field of computer data processing, and the method comprises the following steps: initializing a dynamic adaptive contract cluster, sensing and aggregating the dynamic state of the cluster in real time, executing a contract recursion update cycle, carrying out fault tolerance and state restoration in a recursion process, and carrying out iteration termination judgment. According to the invention, through combination of the dynamic adaptive contract and real-time cluster state perception, intelligent recursion of distributed tensor update is realized, the efficiency and stability of large-scale training are significantly improved, and a communication strategy can be dynamically optimized according to a network and computing power state; state consistency in a complex environment is ensured, conflicts are accurately resolved through version vectors and intelligent arbitration, and elegant fault tolerance is realized by utilizing prediction compensation; and meanwhile, the method has high adaptability, parameters can be optimized on line to reduce manual tuning cost, and robust and efficient support is provided for distributed machine learning.