Abstract: A computer-implemented method for aligning a sensor to a vehicle includes receiving a first frame of measurement from the sensor which includes a first point cloud. One or more clusters Ci representing one or more objects or the ground are segmented. A first set of feature vectors fi is computed for each cluster Ci. Based on the first set of feature vectors fi a second set of feature vectors fi? is predicted respectively using an initial transformation. A third set of feature vectors fj is computed for a second frame with a second point cloud with clusters Cj. A pair of matching clusters is identified from Ci and Cj. A feature distance between the matching clusters is computed. An alignment transformation is computed by updating the initial transformation based on the feature distance. The method further includes aligning the sensor and the vehicle based on the alignment transformation.

Abstract: A vehicle includes a system performing a method of operating a camera for the vehicle. The camera is configured to capture an image. The processor is configured to query the camera to obtain an intrinsic parameter of the camera, estimate a unified parameter based on the intrinsic parameter of the camera, and perform at least one application on the image using the unified parameter.

Abstract: A computer-implemented method for aligning a sensor to reference coordinate system includes initiating a plurality of threads, each thread executes simultaneously and independent of each other. A first thread parses data received from the sensor and stores the parsed data in a data buffer. A second thread computes an alignment transformation using the parsed data to determine alignment between the sensor and the reference coordinate system. The computing includes checking that the data buffer contains at least predetermined amount of data. If at least the predetermined amount of data exists, an intermediate result is computed using the parsed data in the data buffer; otherwise, the second thread waits for the first thread to add more data to the data buffer. The second thread outputs the intermediate result into the data buffer. A third thread outputs the alignment transformation, in response to completion of alignment computations.

Abstract: A data stream classification method includes obtaining, based on a packet feature of a current data stream and a behavior classification model, at least one first confidence that is of the current data stream and that corresponds to at least one data stream class, where the behavior classification model is based on a plurality of data stream samples; obtaining, based on feature information of the current data stream and a content classification model, at least one second confidence that is of the current data stream and that corresponds to the at least one data stream class, where the content classification model is based on one or more historical data streams; and determining a data stream class of the current data stream based on the at least one first confidence and the at least one second confidence.

Abstract: An Internet Protocol Security (IPsec) acceleration method includes generating, by an Internet Key Exchange (IKE) device, an IKE link establishment session packet according to an IPSec configuration parameter and a security policy in a security policy database (SPD), sending, by the IKE device, the IKE link establishment session packet to a peer device, establishing a security association (SA) with the peer device, and sending, by the IKE module, the SA to a data forwarding device, where the IKE device and the data forwarding device are discrete devices.

Abstract: An estimated time of arrival (ETA) of a vehicle is predicted by receiving a request for the vehicle to conduct a trip that includes a first location. A predicted ETA for the vehicle to travel from a particular location to the first location is computed. The predicted ETA is refined to compute a refined ETA using a machine-learned model that takes as input a plurality of features associated with the trip. The plurality of features including at least geospatial features transformed using a locality-sensitive hashing function. An action is performed based on the refined ETA. The action may include one or more of estimating a pickup time or drop-off time for the trip, matching a driver to the trip, and planning a delivery.

Abstract: A sample data annotation system includes an edge node and a central node. The edge node obtains a key feature of sample data, determines, based on the key feature, whether the sample data is unknown sample data, when the sample data is unknown sample data, performs annotation processing on the sample data to obtain a first annotation result, and sends the first annotation result to the central node. The central node receives the first annotation result, and determines whether the first annotation result indicates successful annotation; and when the first annotation result indicates that the unknown sample data is successfully annotated, performs consistency processing on the first annotation result to obtain a second annotation result, or when the annotation result indicates that the unknown sample data fails to be annotated, performs annotation processing on the unknown sample data to obtain a third annotation result.

Abstract: Laser additive manufacturing and a method for preparing thin-walled preforms by laser metal deposition and follow-up rolling. This can solve the problems that when the existing laser metal deposition technology prepares the thin-walled preforms, the limit width size of a molten pool at high power affects the forming wall thickness of the preforms so that it is difficult to prepare preforms with wall thickness less than 2 mm, and the problems of poor surface quality and low accuracy of preforms due to convex and concave peaks caused by the interlayer overlapping, but also can solve the problems that a laser beam with a preset trajectory cannot act on the end surfaces of the preforms due to preform deformation caused by residual stress in a printing process so that the preforms cannot be continuously formed.

Abstract: Embodiments of this application disclose a traffic identification method and a traffic identification device. The method in embodiments of this application includes: The traffic identification device obtains to-be-analyzed traffic of a target data flow. The traffic identification device obtains arrival time intervals of packets of the to-be-analyzed traffic. The traffic identification device determines a type of the to-be-analyzed traffic based on a distribution feature of probabilities of a part or all of the arrival time intervals.

Abstract: A virtual machine live migration method, apparatus, and system relating to the field of communications technologies are disclosed. The method includes: in a process of live migration of a target virtual machine, obtaining pre-recorded storage location information corresponding to target service configuration information, where the target service configuration information is used by a target virtual acceleration component corresponding to the target virtual machine to perform service processing; reading the target service configuration information in a storage location corresponding to the storage location information; and sending the target service configuration information to a destination server, so that the destination server stores the target service configuration information. In the present disclosure, a service processing failure can be avoided.

Abstract: Laser additive manufacturing and a method for preparing thin-walled preforms by laser metal deposition and follow-up rolling. This can solve the problems that when the existing laser metal deposition technology prepares the thin-walled preforms, the limit width size of a molten pool at high power affects the forming wall thickness of the preforms so that it is difficult to prepare preforms with wall thickness less than 2 mm, and the problems of poor surface quality and low accuracy of preforms due to convex and concave peaks caused by the interlayer overlapping, but also can solve the problems that a laser beam with a preset trajectory cannot act on the end surfaces of the preforms due to preform deformation caused by residual stress in a printing process so that the preforms cannot be continuously formed.

Abstract: Embodiments of this application describe a retransmission timeout (RTO) determining method and a related apparatus. The method includes a transmission device sending, to a network analyzer, a network throughput at which communication is performed through a first communication connection, where the network throughput is used by the network analyzer to determine an RTO corresponding to the first communication connection. The network analyzer obtains a first network throughput at which the transmission device performs communication through the first communication connection. The network analyzer calculates, based on the first network throughput, a first RTO corresponding to the first communication connection. The network analyzer sends the first RTO to the transmission device. The transmission device receives the first RTO sent by the network analyzer. The embodiments of this application help improve efficiency of retransmitting a lost data packet.

Abstract: A method, an apparatus, and a device for obtaining an artificial intelligence model, and a storage medium are provided. A client receives a first artificial intelligence AI model sent by a service end (303). The first AI model includes a plurality of neurons. The client determines, from the plurality of neurons, a target neuron participating in a current round of training, where the current round of training is a non-first round of training, and a quantity of target neurons is less than a total quantity of the plurality of neurons (304). The client trains the target neuron based on local data (305). The client returns parameter data corresponding to the target neuron to the service end (306). The parameter data corresponding to the target neuron is used by the service end to obtain a converged target AI model.

Abstract: A method and an apparatus for updating an application identification model, and a storage medium are provided. A client device may determine a plurality of training samples based on identification results of a plurality of pieces of data traffic, and train an application identification model using the training samples. Then, the client device may upload model data of the trained application identification model to a server, and the server performs joint update based on the model data uploaded by a plurality of client devices. Then, the client device may obtain a jointly updated application identification model based on jointly updated model data delivered by the server.

Abstract: A data stream classification model updating method is disclosed. Determining, based on packet information of a current data stream and a behavior classification model, a first data stream class corresponding to the current data stream; determining, based on a target correspondence and a common feature of the current data stream, a second data stream class corresponding to the current data stream, where the target correspondence is a correspondence between a plurality of common features and a plurality of data stream classes; and if the first data stream class is different from the second data stream class, obtaining correction data corresponding to the current data stream, where the correction data includes the packet information of the current data stream and the second data stream class, and the correction data is used as a training sample to update the behavior classification model.

Abstract: A method for implementing model update is provided and used in a federated update framework. The method includes: a service device receives difference information uploaded by at least two clients. The service device performs calculation based on the difference information uploaded by the at least two clients, to obtain first difference consistency information, where the first difference consistency information indicates a consistency degree of the difference information uploaded by the at least two clients. The service device performs calculation based on the first difference consistency information, to obtain first training information, where the first training information is used to train a third model, and the third model is obtained by updating the first model by the service device based on the difference information uploaded by the at least two clients. The service device sends the first training information to the at least two clients.

Abstract: A method and device for manufacturing a large-sized thin-walled tubular part by gas-liquid internal high pressure forming (IHPF). A gas and a liquid are filled at a certain volume ratio into a thin-walled blank. The pressure of the gas-liquid mixed fluid is mainly determined by the gas pressure. During the deformation of the thin-walled blank, due to a large compression ratio of the gas, the gas-liquid pressure will not basically change with the change of the volume of a blank cavity. A support pressure on the cavity of the thin-walled blank is stable in the entire forming process. In addition, even if there is a slight leakage of the liquid or gas during the forming process, the medium pressure inside the blank will not fluctuate largely. In this way, embodiments lower the requirements for the sealing effect during the tubular part forming process.

Abstract: An accelerator loading apparatus obtains an acceleration requirement, where the acceleration requirement includes an acceleration function of a to-be-created virtual machine and acceleration performance of the to-be-created virtual machine. The accelerator loading apparatus determines a target accelerator that meets the acceleration function of the to-be-created virtual machine and the acceleration performance of the to-be-created virtual machine. The accelerator loading apparatus determines an image corresponding to the target accelerator, and sends an image loading command to a target host in which the target accelerator is located, where the image loading command is used to enable the target host to load the image for the target accelerator based on the image loading command.

Abstract: A data stream classification method includes obtaining, based on a packet feature of a current data stream and a behavior classification model, at least one first confidence that is of the current data stream and that corresponds to at least one data stream class, where the behavior classification model is based on a plurality of data stream samples; obtaining, based on feature information of the current data stream and a content classification model, at least one second confidence that is of the current data stream and that corresponds to the at least one data stream class, where the content classification model is based on one or more historical data streams; and determining a data stream class of the current data stream based on the at least one first confidence and the at least one second confidence.

Abstract: An accelerator loading apparatus obtains an acceleration requirement, where the acceleration requirement includes an acceleration function and acceleration performance of a to-be-created virtual machine, determines an image that meets the acceleration function and the acceleration performance, and determines a target host in which an available accelerator that can load the image is located, and then sends an image loading command to the target host. The image loading command includes a descriptor of the image, and is used to enable the target host to load the image for the available accelerator. In the method, a target host that can create the virtual machine may be determined based on the acceleration function and the acceleration performance of the to-be-created virtual machine, and an image used for acceleration is loaded to an available accelerator of the target host, to implement dynamic accelerator loading and deployment.