11results about How to "Shorten communication time" patented technology

The invention discloses a large model reasoning hardware accelerator based on data flow execution, and belongs to the technical field of large model reasoning hardware acceleration. Each node comprises a linear calculation kernel used for sequentially completing query / key / value projection and attention output projection in multi-head attention calculation, dimension raising projection and dimension reduction projection calculation in a feedforward neural network layer, and inverse quantization operation after each projection is output; the multi-head attention calculation kernel is used for completing loading and quantization of query / key / value vectors, multi-head division and cache management, attention score calculation, Softmax normalization, context vector generation and KV cache updating; the RLA kernel is used for completing residual addition, layer normalization, quantization / inverse quantization and nonlinear activation calculation of each layer; the three cores in the same node are located in the same SLR area, and data transmission is carried out through a hardware queue and an HLS blocking mechanism. According to the invention, high hardware utilization rate and low-delay reasoning can be realized.

This application provides a method, apparatus, device, and storage medium for recommending test environments. The method includes: receiving an input identifier of a business system to be tested; determining, based on the identifier, whether there is a business logic relationship between the business system to be tested and other business systems, indicating whether the testing of the business system to be tested depends on the test results of other business systems; if a business logic relationship exists, identifying the other business systems as related business systems; determining a target environment chain from multiple preset environment chains based on the identifier of the business system to be tested and the identifiers of the related business systems; and displaying test environment recommendation information for the business system to be tested, including information about the target environment chain. In this application, by matching the business system to be tested to determine related business systems with which it has a business logic relationship, and automatically recommending the target environment chain, communication time is saved.

The application provides a kind of dynamic selection method and system of ultra-wideband positioning base station, comprising: by mobile terminal broadcast query signal, collect the base station response message and corresponding RSSI reaching preset quantity;RSSI validity screening is carried out to the collected base station, and weak signal base station is eliminated;Effective base station is divided into odd-numbered base station group and even-numbered base station group according to the parity of its number;The strongest RSSI of the preset number of base stations is selected from the two groups respectively, and the final positioning base station combination is formed by merging, also includes smooth switching mechanism and GDOP check step.Through the staggered layout mode that odd-numbered base station is deployed on one side of scene and even-numbered base station is deployed on the other side, the application ensures that the selected base station is distributed on both sides of the tag from the mechanism, forms the positioning configuration of geometric distribution balance, effectively reduces GDOP value.At the same time, the cut-off mechanism of limiting the number of acquisition significantly shortens the positioning waiting time, improves the real-time performance of the system.

The invention discloses a gradient communication compression method and system for a shared memory heterogeneous processor, and the method comprises the steps: carrying out the Top-k precise sorting of a gradient obtained by carrying out the forward propagation and back propagation of a deep learning model in a specified round through all nodes based on a universal CPU (Central Processing Unit), and combining with a compression ratio, so as to obtain a latest threshold value through compression; gradient compression is carried out on the basis of cooperation of the heterogeneous acceleration core and the universal CPU in combination with the threshold value of the last round; each node performs gradient communication, sparse decoding and parameter updating based on the general CPU; carrying out step self-increasing on the training round; and according to the training round step, judging whether the training round ends or not, and deciding to continue iteration or exit. The invention aims to realize optimization of a communication algorithm for distributed deep learning training of a shared memory heterogeneous processor, solve the problem of low efficiency of a traditional gradient communication compression method, reduce the communication time, reduce the compression cost and improve the communication efficiency.

The invention discloses a diabetic retinopathy lesion visualization method and device and a readable medium, and the method comprises the steps: extracting a region of interest from an original fundus image, and generating a region of interest image; determining a feature stroke code of a feature segmentation region in the region-of-interest image in the original fundus image, and determining a focus stroke code of a diabetic retinopathy focus segmentation region in the region-of-interest image in the original fundus image; based on the feature run-length codes, a reference area used for assisting in judging the diabetic retinopathy focus is obtained, and corresponding reference run-length codes are generated; and displaying the feature stroke code, the focus stroke code and the reference stroke code in a blank image differently by using different colors to generate a fundus focus visual image. Therefore, the original fundus image is simplified, and the simplified image is visually displayed, so that a doctor and a patient can check the image conveniently, the doctor-patient communication time is saved, and the utilization rate of medical resources is improved.

The application discloses a longitudinal federated EBM commodity recommendation method with differential privacy protection, and comprises the following steps: 1, a master server uses a consumption feature data set to train an EBM recommendation model; 2, a slave server number is initialized; 3, the slave server uses the consumption feature data set to train the EBM recommendation model; and 4, whether the EBM recommendation model training is ended is determined according to the number of iterations. The application can effectively protect the user consumption feature information, thereby preventing the leakage of the user consumption feature information and providing effective commodity recommendation.

This invention discloses a low-cost fusion positioning system and method for personnel location management in smart construction sites, including fusion positioning tags, an iBeacon beacon module, and a location management backend module. The tags incorporate GPS, Bluetooth, and G-Sensor sensors, achieving seamless indoor and outdoor positioning by dynamically switching positioning frequency and power consumption modes: GPS is used outdoors, while indoors, the ID corresponding to the strongest signal is obtained by scanning iBeacon beacons and uploaded to the backend. The backend indexes preset coordinates to complete location calculation and coordinate unification. This invention supports data transmission via smartphone cellular networks, featuring breakpoint resume and local caching functions to ensure the integrity of location information. The system is low-cost, adaptable to environmental changes, and suitable for real-time location management of construction site personnel, vehicles, and materials, as well as hazardous area control.

This invention discloses a dual dialing method, communication method, and corresponding device for UID codes in a digital electronic detonator initiation network, relating to the field of digital electronic detonator technology. The invention involves splitting the UIDs of all digital electronic detonators in the initiation network system and performing dual dialing according to the split UID groups. It sequentially confirms the first and second dialing states of each digital electronic detonator after dual dialing, and finally determines whether the corresponding digital electronic detonator has entered the selected dialing state based on the first and second dialing states. The digital electronic detonator that ultimately enters the selected dialing state can communicate one-to-one with the detonator, achieving fast and efficient UID code dialing, reducing the time consumed by UID code dialing, thereby reducing the communication time of the digital electronic detonator initiation network system and improving network communication efficiency.

The embodiment of the invention provides a vehicle positioning method, electronic equipment and a vehicle. The method comprises the steps that vehicle positioning data, vehicle environment information and road-level map data are acquired; determining the initial position of the vehicle according to the vehicle positioning data; determining real-time road data according to the vehicle environment information; matching the real-time road data with the road-level map data, and determining a position correction parameter; and correcting the initial position of the vehicle based on the position correction parameter, and determining the target position of the vehicle. The method gets rid of dependence on an external base station, solves the problem that a sensing device cannot obtain effective lane information, reduces the dependence on map precision, and can achieve the accurate positioning of a vehicle at high efficiency and low cost.

This application discloses a data sharding processing method, apparatus, computer device, medium, and program product. The method is applied to a target node in a cluster comprising multiple nodes. Specifically, it retrieves ownership information corresponding to multiple data shards from a preset storage location. If, based on the ownership information of a first data shard, it is determined that none of the nodes in the cluster owns the first data shard, then ownership of the first data shard is acquired. If ownership of the first data shard is successfully acquired, the first data shard is loaded. In this process, the node indicated by the ownership information of the first data shard in the preset storage location is updated to the target node. The ownership information of each data shard is uniformly stored in the preset storage location, eliminating the need for frequent communication between multiple nodes in the cluster. Instead, the information is retrieved directly from the preset storage location, which is convenient, fast, reduces network overhead, shortens communication time, and improves the processing efficiency of data shards.

The utility model provides a command system for intelligent automobile loading, and relates to the technical field of automobile transportation. The utility model provides a command system for intelligent loading of an automobile. The command system comprises an industrial personal computer, positioning equipment for positioning the position of the automobile, an industrial camera and an interactive prompt device, the positioning equipment is arranged in a loading area and is connected with the industrial personal computer through a signal line; the industrial camera is arranged above the blanking port, a lens of the industrial camera faces the blanking port of the loading machine and a carriage of the loading vehicle, the industrial camera is connected with the industrial personal computer through a data line, and the industrial personal computer is configured to generate a command instruction according to image data shot by the industrial camera; the interactive prompt device is arranged in the loading area and is in communication connection with the industrial personal computer. The utility model provides a command system for intelligent automobile loading, which can improve the loading quality and the loading efficiency.