16 results about "Discrete set" patented technology

Methods and systems for one or more computers, in which a method includes obtaining encoding sequences of an input data item, in which each encoding sequence includes a respective encoding vector at each position of multiple positions. The method includes generating a combined encoding sequence by, at each position, combining the respective encoding vectors at the position in the multiple of encoding sequences. The method includes processing the combined encoding sequence using a deduplicator neural network to generate a deduplicated encoding sequence that includes a respective deduplicated encoding vector for each of the positions and applying a tokenizer to the deduplicated encoding sequence to identify, for each deduplicated encoding vector, a discrete representation of the deduplicated encoding vector generated from respective codebook vectors from each of a set of one or more codebooks, in which each codebook is a respective discrete set of codebook vectors.

Imitation learning, or artificial intelligence-based learning from demonstration, aims to acquire an agent policy by observing and mimicking the behavior demonstrated in expert demonstrations. Imitation learning can be used to generate reliable and robust learned policies in a variety of tasks involving sequential decision-making, such as autonomous driving and robotics tasks. However, existing methods that use next-token-prediction (NTP) models, where the policy reduces to a classifier over a discrete set of trajectory tokens, suffer from covariate shift due to their open-loop training a closed-loop execution. The present disclosure provides closed-loop fine tuning of autonomous agent policies in a manner that can mitigate covariate shift.

Methods and apparatus for compressing and decompressing weight values associated with a neural network. The input weight values are divided into a plurality of groups, wherein each group is processed using a scaling factor. For each set of scaled input weight values, a codebook is identified from a plurality of codebooks, where each codebook represents a discrete set of centroid values. The input weight values within each group are encoded using centroid values from the identified codebook, thereby producing encoded weight values including a codebook index and a centroid index. During decompression, the encoded weight values are processed using corresponding codebooks and scaling factors to reconstruct output weight values. A codebook is generated by identifying similar distributions of scaled input weight values across different groups and clustering these values to determine centroid values. The processing element performs a decompression operation to reconstruct the weight values for neural network operations.

Imitation learning, or AI-based demonstration-based learning, aims to derive a policy for an agent by observing and imitating the behavior shown in expert demonstrations. Imitation learning can be used to generate reliable and robust learned policies for a variety of tasks requiring sequential decision-making, such as autonomous driving and robotics. However, existing methods that use next-token prediction (NTP) models, where the policy is reduced to a classifier over a discrete set of trajectory tokens, suffer from covariate shifting due to their open-loop training and closed-loop execution.The present disclosure provides a fine-tuning of the policies of autonomous closed-loop agents in a manner that can mitigate covariate shifting.

Database query processing techniques are disclosed. In various embodiments, a query associated with a database is received. A byte code representation of the query is generated, including by decomposing the query into a discrete set of streaming operators defined over associated data frames, wherein the byte code includes code defining for each operator in the discrete set of streaming operators the processing to be performed by that operator and further embodies a data flow graph that defines a flow of data to and through the discrete set of streaming operators. The byte code is executed by a query processing engine to generate and return a query result.

This invention belongs to the field of electric vehicle charging load prediction technology, and discloses an electric vehicle charging load prediction method based on GCN-Rollout, including the following steps: real-time acquisition of multi-source data including road network topology data, traffic flow status data, and individual electric vehicle status data; based on graph convolutional network (GCN), extracting charging load impact features reflecting spatial dependencies from the multi-source data, and preprocessing them to obtain the current state vector representing the traffic flow and charging status at the road network nodes at the current moment; defining the charging demand ratio as a decision variable, and generating a discrete set of decision variables based on the current state vector; performing simulation evaluation for each candidate decision variable in the decision variable set to construct the corresponding long-term cost evaluation function; effectively solving the problem in the prior art where the prediction model has insufficient ability to fuse multi-source dynamic information, resulting in poor prediction accuracy and disconnection from scheduling decisions.

This invention discloses a method for assessing the multi-energy flexibility of industrial loads considering uncertainties. The method includes: constructing a model of the energy consumption and output relationship of a production process within an industrial load considering output adjustment constraints; using graph theory to associate the production process and product storage warehouses to construct storage and output requirement constraints; constructing a discrete set of states for the industrial load and performing a feasible interval search to construct a multi-energy feasible interval; searching the feasible interval to obtain the multi-energy flexibility of the industrial load, thus achieving the assessment of multi-energy flexibility. This invention assesses the multi-energy flexible adjustment capability of industrial loads considering production uncertainties. It can be used to adjust the demand of industrial loads for various energy forms at the operational level, and can be applied to multiple fields such as electricity and multi-energy demand response. This assists energy systems in utilizing the adjustment capability of industrial loads, enhancing system flexibility, and supporting the flexible operation and optimized scheduling of energy systems.

This application discloses a method and related products for weld seam extraction and robot trajectory generation based on digital model prior guidance, relating to the field of automated welding technology. The method includes: acquiring downsampled point clouds and undownsampled point clouds, where the downsampled point clouds are obtained by downsampling a coarsely registered point cloud, and the undownsampled point clouds are the coarsely registered point clouds without downsampling; finely registering the downsampled point clouds with the digital model point cloud to obtain a fine registration matrix; projecting the undownsampled point clouds onto the spatial coordinate system of the digital model point cloud using this matrix to obtain a projected point cloud; spatially clipping the projected point cloud using a predefined region of interest (ROI) for the weld seam in the digital model point cloud to obtain a weld seam region point cloud; performing local geometric feature analysis on the weld seam region point cloud, and extracting the weld seam trajectory based on a discrete set of weld seam feature points, and generating a smooth robot welding trajectory. This application can achieve high-precision, low-computing-cost weld seam trajectory generation.

This invention relates to the field of intelligent operation and maintenance technology for new energy power generation and power electronic equipment, specifically a fault diagnosis system for photovoltaic inverter operation anomalies. The system includes: an edge data acquisition and dynamic reconstruction unit that maps discrete time-series data into high-dimensional phase space trajectory data; a topological cross-section feature extraction unit that generates a discrete set of two-dimensional intersection coordinates; a topological fingerprint generation unit that generates a topological fingerprint matrix; a cloud-based topological inference unit that generates diagnostic status results; and an anomaly tracing closed-loop control unit that monitors the diagnostic status results in real time and generates a backtracking lock command when the diagnostic status result is abnormal or suspected of being faulty, triggering the edge controller to stop overwriting the circular buffer and lock the corresponding high-frequency original waveform data. This invention enables resource-constrained microcontrollers to complete feature extraction within microseconds, ensuring the real-time performance of edge-side processing.

Methods and apparatus for compressing and decompressing weight values associated with neural networks. Input weight values are divided into groups, with each group being processed using a scale factor. For each group of scaled input weight values, a codebook is identified from multiple codebooks, where each codebook represents a discrete set of centroid values. Input weight values within each group are encoded using centroid values from the identified codebook, resulting in encoded weight values comprising codebook indices and centroid indices. During decompression, the encoded weight values are processed to reconstruct output weight values using the corresponding codebooks and scale factors. The codebooks are generated by identifying similar distributions of scaled input weight values across different groups and clustering these values to determine centroid values. A processing element performs the decompression operations to reconstruct the weight values for use in neural network operations.

Database query processing techniques are disclosed. In various embodiments, a query associated with a database is received. A byte code representation of the query is generated, including by decomposing the query into a discrete set of streaming operators defined over associated data frames, wherein the byte code includes code defining for each operator in the discrete set of streaming operators the processing to be performed by that operator and further embodies a data flow graph that defines a flow of data to and through the discrete set of streaming operators. The byte code is executed by a query processing engine to generate and return a query result.

A combination deterministic and stochastic waveform generator having a waveform generating algorithm for a waveform s(t)=exp{j2π[g∫0tfDET(τ)dτ+h∫0tfSTO(τ; x)dτ]}, wherein fDET(t) is a deterministic frequency function of time; wherein fSTO(t; x) is a stochastic frequency function of time; wherein 0≤g≤1 and 0≤h≤1 are relative weighting parameters; wherein fSTO(t; x) is parameterized on vector x comprised of a discrete set of random / pseudo-random values to yield a waveform; and wherein, for h and g values in between 0 and 1, the waveform is a combination deterministic and stochastic waveform and is partially Doppler tolerant and resistant to electronic attack. The combination deterministic and stochastic waveform is unique from pulse to pulse, is independent of velocity, can be generated by a standard waveform generator, and can be implemented with a standard modern radar system.

The invention relates to an optimized binary convolution unit (ESBCU) based on a channel expansion and contraction technology, and belongs to the technical field of binary neural network optimization. The optimized binary convolution unit aims to solve the problem that an existing basic binary convolution unit (BBCU) is insufficient in information expression ability due to the fact that an output value is limited to a limited discrete set. According to the core scheme, through channel expansion, contraction and copy operation, firstly, the number of input channels is expanded by t times to increase a discrete set output by binary convolution, then the number of convolution kernel output channels is reduced to 1 / t of the original number to maintain the calculation complexity unchanged, and finally, through processing of a channel copy splicing layer and a batch normalization (BN) layer, the convolution kernel output channel number is reduced to 1 / t of the original number to maintain the calculation complexity unchanged. And recovering the final output channels to the number compatible with the next layer. According to the method, on the premise of not increasing the calculation complexity, the output dynamic range of the binary convolution is effectively expanded, and the feature representation capability of the binary convolution is enhanced. Experiments show that the unit can effectively improve model performance (such as a PSNR index) in low-level visual tasks such as image super-resolution and the like, and can be flexibly integrated into various deep neural networks as a plug-and-play standard module.

The invention discloses a method for efficiently constructing a three-dimensional entity model topological relation based on spatial hash. The method comprises the following steps: firstly, reading and preprocessing model data, and extracting boundary curves and endpoint coordinates of all industrial parts; establishing spatial position indexes for boundary curve end points of all industrial parts by utilizing a spatial hash data structure; in the matching stage, a small number of candidate curves (wide stage) adjacent to the boundary curve of each industrial part in space are quickly screened out by querying a spatial hash table, and then only the candidate curves are subjected to accurate geometric consistency verification (narrow stage), so that global pairwise comparison is avoided; and finally, according to the successfully verified matching pairs, generating shared topological entities such as edges and vertexes, and suturing discrete curved surface sets into a B-Rep entity model with a complete topological relation. According to the method, the complexity of a matching algorithm is reduced, and the speed and efficiency of importing the complex model are greatly improved.

This application provides a data processing method and electronic device for geometric components, relating to the field of data processing technology. The method includes: dividing a discrete set of points according to the normal direction corresponding to the geometric component to obtain multiple original layered point sets; sorting the points in each original layered point set according to the multidimensional features of each point to obtain layered ordered point sets corresponding to each original layered point set; constructing a two-dimensional mapping mesh corresponding to each layered ordered point set; mapping each point in each layered ordered point set to the two-dimensional mapping mesh and performing a transformation to obtain the three-dimensional mapping coordinates of each point in each layered ordered point set; and obtaining the mapping result based on the three-dimensional mapping coordinates of each point in each layered ordered point set. This application improves mapping accuracy by adapting complex geometric components to normal layering and combining multidimensional feature sorting, making it applicable to complex component point set processing in multiple fields.

The present disclosure provides "tire wear estimation". For a corresponding example, a noise factor of a vehicle tire is detected. The noise factors comprise tire pressure, load and tire rotating speed. For the respective instance, a respective radius of the vehicle tire is determined. The respective radii are grouped based on respective discrete sets of noise factors. A transition instance of a health condition of the vehicle tire is estimated based on extrapolating a radius of the respective packet.