32 results about "IMG" patented technology

Embodiments of the present application provide a system, a device, and a method for approximate homomorphic ring encryption. A message z, which is an (n / 4)×(n / 2) matrix may be encoded to a polynomial M(X, Y) in a polynomial ring =[X, Y] / Xn / 2+1, Yn / 2−(Xn / 8−X3n / 8). The polynomial M(X, Y) may then be encrypted to a ciphertext and sent to a target device. Encoding in the polynomial ring allows at least approximate homomorphic manipulation of the subsequent ciphertext for improved security.

The invention relates to methods and apparatuses for reconstructing phase information in a time series, the time series being derived from a time evolution of an input state acted on by a Hamiltonian, H. First, an input state, |ψ, a total time evolution time, T, and the Hamiltonian, H are received. Next a time evolution of the input state is enacted, using H to generate an absolute value of a first time series, |f1|. In addition at least one additional family of absolute values of time series is generated, each time series being a discrete or continuous function of time t, for t∈[0,T]. Finally, phase information is extracted for the time series of the input state from the absolute values of the first time series and the at least one additional family of absolute values of time series.

Described are various embodiments of a system and method for minimizing a storage mode dephasing error caused by a reset of the auxiliary qubit. In one embodiment, the method comprises the steps of performing on the auxiliary qubit a first ηge pulse causing a |g↔|e transition in said auxiliary qubit, waiting for a designated echo time duration; and performing on the auxiliary qubit a second ηge pulse. The method may further comprise performing a f0g1 reset on the auxiliary qubit.

The application discloses a kind of based on the safe processing and transmission method of DICOM file, client and server, comprising: obtaining DICOM file, the DICOM file includes metadata info and image data img;The metadata info is de-sensitized and is handled, and sensitive data is obtained;The sensitive data is encrypted, and the sensitive data after encryption is written in image data img;The metadata info after de-sensitization processing and the image data img of writing encrypted sensitive data are generated new DICOM file according to preset format, and the new DICOM file is sent to server.The application retains the IMG data and META data of original file, and is stored in somewhere of file itself by safe encryption means, and processed DICOM cannot view METADATA and other private information by ordinary method, but using corresponding authorized reverse method can still analyze and view relevant information.Therefore, the method retains META information under encryption mode, while patient privacy is maximally protected.

This invention discloses a low-dose CT image denoising method, comprising the following steps: S1: establishing a dataset, defining a set of noisy LDCT images x, and a set of high-quality NDCT images y corresponding to the LDCT image set x; S2: constructing a denoising model with parameter θ. Use templates to build Mapping; S3: Training the denoising model using hybrid loss Set the learning rate and momentum parameters, and use error backpropagation to train the denoising model. S4: Image denoising; After preprocessing the LDCT images in the test set, the images are input into the trained denoising model to obtain the corresponding denoised images. The denoising method proposed in this application takes into account the denoising difficulty of different regions of the image, and adaptively balances them by adding weights. It also promotes the generation of image texture details by losing high-frequency information, thus avoiding the denoised image from becoming overly smooth.

A quantum circuit is configured to implement a quantum gradient algorithm when executed on qubits of a quantum computing system. The quantum gradient algorithm includes a phase oracleOSfmdefined by a finite difference approximation with an order greater than zero, and a complexity of the quantum gradient algorithm scales as (√{square root over (k)} / ϵ). The quantum circuit is repeatedly executed on qubits of a quantum computing system to determine a k-dimensional gradient of a function ƒ(x) within an error ϵ at point x0, where ƒ(x) is not a Gevrey class G1 / 2 function.

This disclosure relates to a spoken language understanding (SLU) system and method for training the same. The method comprises: generating a vector ei to represent an ith word token wi based on n word tokens; generating an intent-specific matrix EI and a slot-specific matrix ES based on a sequence of vectors e1:n; generating an intent label-specific matrix VI and slot label-specific matrices VS,K based on the intent-specific matrix EI and the slot-specific matrix ES, respectively; generating a multiple intent representing matrix 1 and a slot representing matrix +2 based on the slot-specific matrix ES, the intent label-specific matrix VI and the slot label-specific matrices VS,k; generating final intent outputs based on the multiple intent representing matrix 1 and the intent label-specific matrix VI; generating final slot output based on the slot representing matrix +2 and the slot-specific matrix ES; adjusting the SLU system based on an objective loss .

The invention discloses a time synchronization information security identification method and system based on a self-attention mechanism, and relates to a time synchronization and high-precision clock technology, and the system comprises a data collection and processing module which is used for receiving, caching and processing original time information data, converting the original time information data into variable input of a self-attention matrix calculation module, and outputting the variable input of the self-attention matrix calculation module; the variable input is a preprocessed clock error sequence; the self-attention matrix calculation module is used for receiving the preprocessed clock error sequence, calculating context correlation of clock error data and outputting a probability matrix; the security identification decision module is used for averaging all time points represented by the sequence so as to aggregate and obtain global features; global features are mapped into abnormal probabilities through weights and offsets, and sequence judgment results are divided into normal results and abnormal results through binary decision. According to the invention, illegal information embedded in a complex modulation mode can be effectively identified, and the perception and defense capability of the system to hidden attacks is significantly improved.

A system for matrix-based public key homomorphic encryption, including a processor of a computing node configured to host a homomorphic encryption module and connected to at least one cloud server and a memory on which are stored machine-readable instructions that when executed by the processor, cause the processor to: acquire plaintext x required to be encrypted; select a size of a matrix and modulus n; select invertible × matrix S over n, wherein n is a residue ring modulo n; compute an invertible × matrix S−1 over n; select two random secret keys (S, S−1) and (K, K−1) of a homomorphic encryption scheme Σ(m), where S and K belong to a ring of ×-matrix Ml(m) over residue ring m with modulus m=pq, where p, q are primes; select (K, K−1) as a master key; use the master key to encrypt all elements of matrices S and S−1; obtain two ciphertexts P1=EncK(S) and P2=EncK(S−1), which both form public key of the Σ(m) including pk=(P1, P2), wherein a secret key is sk=((S, S−1), (K, K−1)); and apply the pk and the sk for encryption and decryption of the plaintext x.

A storage unit stores evaluation function information of an evaluation function of a combinatorial optimization problem including a sum of an objective function and one or more constraint functions including a first constraint function weighted by a first coefficient (t) representing a weight of a first constraint condition. When repeatedly executing a local search for searching for a solution candidate (x*) of the combinatorial optimization problem using the evaluation function information while changing a value of t, a processing unit changes the value of t in a direction in which a value (V(x*)) of the first constraint function corresponding to x* obtained by the local search using each value of t approaches a target value (Vtarget) that is a positive value.

A method for processing homomorphic ciphertexts and a computing device, the method comprising: obtaining a sequence of a plurality of first RLWE ciphertexts, a plaintext corresponding to each first RLWE ciphertext comprising a first monomial expression using each coefficient of a first polynomial as a first index; group key ciphertexts are obtained, each group key ciphertext comprises a plurality of RGSW ciphertexts which are binary bits of a second index of a second monomial expression, the binary bits are larger than or equal to 2, and the ith group key ciphertext corresponds to the value of the ith bit in the second index; based on the group key ciphertext, secondary rotation processing is carried out on the sequence of the multiple first RLWE ciphertexts to obtain a sequence of multiple second RLWE ciphertexts, the rotation mode is selected based on the ith group key ciphertext in the ith rotation processing, and the plaintext corresponding to each second RLWE ciphertext comprises a third monomial expression using each coefficient of a second polynomial as a third index; the second polynomial is obtained by multiplying the first polynomial by the second monomial.

The invention discloses an OTA upgrading method, equipment and device for an application program and a medium, and relates to the technical field of vehicles. A part of the disk partitions in the whole disk partitions adopt the IMG upgrading mode, and compared with the mode that all the disk partitions adopt the IMG upgrading mode, the overall quantity of upgrading packages is reduced. And the other part of disk partitions adopt a file upgrading mode, so that multiple versions of redundant files are totally reduced, and the disk space is saved. And according to the partition address information of the second disk partition, constructing a virtual mirror image environment corresponding to the file upgrading mode, so that the mirror image environment under the file upgrading mode is the same as the mirror image environment under the IMG upgrading mode, and collaborative upgrading of different upgrading modes is realized. The stored application program is upgraded according to the upgrading modes of the first disk partition and the second disk partition, so that OTA upgrading is completed, the advantages and disadvantages of the two upgrading modes are considered, upgrading balance is achieved, it is guaranteed that the occupied disk space is small, and meanwhile the upgrading speed is increased.

The invention provides a method for improving AutoFocus performance, which comprises the following steps of: S1, extracting an edge value by a sobel operator: extracting image edge information through a cv2. Sobel () instruction, and representing as sobelx = cv2. Sobel (img, cv2. CV32F, 1, 0) and sobely = cv2. Sobel (img, cv2. CV32F, 0, 1); s2, solving a variance: af = (np.var (sobelx) + np.var (sobely)) * N, np.var is a function for solving the variance, the formula is that the sum of the variances of sobelx and sobely is multiplied by a coefficient N, and the obtained af is the definition value of the picture; pixel values obtained after Sobel operator processing are distributed more dispersedly, so that the variance can be used as an evaluation criterion because the variance is the measure of the dispersion degree when a random variable or a group of data is measured. Compared with the prior art, the method has the advantages that the performance is improved while the effect is better than that of the prior art, the difference value between the peak value and the peak valley of the optimized method is larger in the aspect of the effect, and therefore the peak value can be found more easily, the calculation amount is small in the aspect of the performance, and the CPU utilization rate and the memory use can be reduced.

This invention relates to fields such as AI adversarial techniques, and discloses a data-free black-box adversarial example generation method based on attention enhancement, comprising: randomly sampling a noise space, inputting it into a generator to generate synthetic data. ; Utilizing synthetic data Query black-box target model The system calculates the decision output and caches the input-output pairs; it strengthens the alternative model using an attention enhancement module and trains the enhanced alternative model; it trains the generator using a joint structural information learning strategy; it replays the cached input-output pairs to further train the enhanced alternative model; it employs a white-box attack strategy to construct adversarial examples on the enhanced alternative model and transfers them to the black-box target model to carry out the attack; it overcomes the problems of low learning efficiency of existing alternative models for different synthetic data, poor diversity of synthetic data generated by the generator, and low training efficiency of alternative models, and by optimizing the learning strategies of the alternative model and the generator, the optimized alternative model can accurately simulate the decision output of the black-box target model.

A method and system for enhancing large language model auxiliary network operation and maintenance capability, the method comprising: collecting a training sample, the sample collection process comprising: inputting analog network data into a network simulator, generating a text attribute graph of an analog network and a corresponding natural language query, and collecting network behavior data according to the natural language query; a large language model is constructed and trained, and the working process is as follows: natural language query and equipment configuration texts are generated through a configuration encoder, text attribute graphs are obtained through a topology encoder, then the text attribute graphs are spliced into network representation, and then the network representation is mapped into a Q layer and a V layer of an attention mechanism of the large language model in a low-rank fine tuning mode; and the large language model receives natural language query input by the user and the text attribute graph of the current network for reasoning output. The invention relates to the field of information communication networks, can efficiently model multi-modal network data, enhances the understanding ability of a large model for the current network state, and effectively assists network operation and maintenance.

The invention provides an efficient fault-tolerant broadcasting method for a dragonfly network. The method comprises the following steps: selecting any node as a root node r; dividing the construction process of the vertex independent spanning tree into two situations according to whether a first sub-node connected with a root node r in the spanning tree is an intra-group node or an out-group node; respectively generating n + h-1 vertex independent spanning trees which take r as a root and cover all vertexes in the network according to a preset construction process of the corresponding situation; and broadcasting data by using the spanning trees. According to the method, based on the topological characteristics of'in-group full interconnection and inter-group regular connection 'of the dragonfly network, unified VISTs construction logic is designed in combination with the types (in-group nodes and out-group nodes) of first-hop connection nodes of root nodes, a construction algorithm does not need to be designed for different values of the dragonfly network, and efficient broadcasting and safe distribution of data can be achieved.

This invention discloses a digital image encoder, decoder, processor, and electronic device, belonging to the field of image transmission technology. The encoder includes: a primary color calculation module, used to receive RGB888 image patches and divide them into two equal rectangular sub-blocks, and calculate the primary color calculation value β of channel k in sub-block i. ik The difference value calculation module is used to iterate through the original color value γ of channel k of pixel j in sub-block i. ijk And calculate the corresponding primary color value β ik By subtracting, we obtain the 9-bit color difference value Δ. ijk Brightness modulation calculation module, used to calculate brightness modulation value. The exponent calculation module is used to determine the shared exponent for all luminance modulation values, obtaining the shared index shift; the luminance modulation quantization module is used to determine each luminance modulation value. 4-bit mantissa δ ij The compressed encoding carries a 5-bit primary color storage value α. ik 4-bit mantissa δ ij And shared index shift. This invention improves the compression ratio of compressed coding with minimal image loss through a simple pipeline design.

The invention relates to a method and an optical device for processing tiled images, wherein a plurality of overlapping tile images of a sample are captured using the optical device. The method is characterized in that a model function is specified for a specified relation, in particular for a shadow function, correction factors at support positions in the overlapping region of adjacent tile images are determined, and a shadow function and a shadow-free sample image are at least approximately determined in each case.

In a first stage, a respective identity of each individual is determined by matching a respective code (C [1]) obtained from a radio-based readout of a respective tag carried by an individual in a subset of animals with a set of reference codes (DR) in a source database (115). In a second stage, respective animal-related data (Dimg) in the form of 2D and / or 3D image data is registered for each individual in the subset. The registered image data (Dimg) is matched with a subset of reference data records (dr) in the temporary database (125) containing only reference data records with reference image data relating to individuals in the subset of animals. In response to a match between the registered image data and the reference image data, a confirmed identity is assigned to one of the individuals in the subset, the confirmed identity being an animal identity associated with the matching reference data record.

The invention provides a Gaussian splashing human body three-dimensional reconstruction method based on SDF constraint and a storage medium. The method comprises the following steps: constructing a voxel grid with the resolution of N3 in a normalized space omega based on an SMPL-X human body template, and for each discrete sampling center in the voxel grid, calculating the minimum Euclidean distance from the discrete sampling center to the surface of a human body to construct a symbolic distance field SDF; approximately calculating a gradient vector of each voxel point on the discrete grid by adopting a central finite difference method; mapping the center coordinate of the 3D Gaussian ball to the voxel grid, and obtaining the SDF value and gradient of the position of the 3D Gaussian ball through trilinear interpolation; based on the SDF value and the gradient, executing an adaptive density control operation guided by the SDF; according to the method, image reconstruction loss is adopted as an optimization target, and the color, opacity and three-dimensional covariance parameters of a Gaussian ball are driven to perform iterative updating by minimizing the difference between a rendered image and a real reference image, so that dynamic human body reconstruction is realized.

The invention relates to a computer-implemented method for solving a combinatorial optimization problem encoded in the ground state of an Ising spin-glass model.In particular, the invention introduces a method called bias field digitized counterdiabatic quantum optimization (bf-DCQO) for solving combinatorial optimization problems on digital quantum computers. This method integrates auxiliary counterdiabatic (CD) terms into the adiabatic Hamiltonian and introduces bias terms derived from classical methods, quantum annealers, or iterative DCQO applications. This approach addresses the limitations of current quantum computers, particularly their limited coherence times, and improves the time to solution for large-scale problems compared to traditional DCQO and finite-time adiabatic quantum optimization.The bf-DCQO method is entirely quantum, avoiding the trainability issues of variational quantum optimization algorithms. It significantly outperforms the Quantum Approximate Optimization Algorithm (QAOA) in success probability and approximation ratio. Experimental implementation on a trapped-ion quantum computer and a superconducting processor demonstrated its effectiveness on a maximum weighted independent set problem with 36 qubits and a spin-glass problem on a heavy-hex lattice with 100 qubits. This represents the largest such problem solved on a gate-based quantum computer using a purely quantum algorithm.

A computer implemented method, for mitigating errors in a quantum circuit comprising at least one occurrence of a quantum logic operation G. The method includes computing a set of coefficients {cp}, associated with a set of basis operations ={Bp}, to obtain a quasi-probability decomposition G0≈ΣpcpBp on the set of basis operations . The quasi-probability decomposition is of a target version of the quantum logic operation G, denoted G0. The set of basis operations {Bp} forms a multi-type basis, constructed from the quantum logic operation G and elements of a set of mitigation operations . The decomposition is computed so as to reach a decomposition target, being based on at least one of a decomposition accuracy target, and a decomposition sampling overhead target. The method includes implementing the quasi-probability decomposition on the quantum processor.

Disclosed are a neural network processor, a system-on-a-chip, a data processing method, and a storage medium, relating to the technical field of systems-on-a-chip. The neural network processor includes a first processor core, where the processor core includes: a first buffer, configured to buffer a first input tensor corresponding to a first neural network layer in the neural network model; a first direct memory access controller, configured to read a second input tensor corresponding to the first neural network layer from a second buffer, and write the second input tensor into an operational array; and the operational array, configured to read the first input tensor from the first buffer, and perform a first operation based on the first input tensor and the second input tensor, to obtain a first output tensor.

The invention relates to a GPU progressive optimization method for neural operator backbone network reasoning, and belongs to the technical field of electronic information. Comprising the steps of S1, initializing constraints and tolerances; s2, baseline measurement is carried out to obtain baseline execution time and a reference result; s3, obtaining a pre-estimated segmentation value based on the working set model and expanding the pre-estimated segmentation value; s4, carrying out KSweep evaluation and consistency check, and carrying out optimization; the process of scanning and measuring the candidate set one by one is called KSweep; s5, fixing, newly adding fusion reasoning optimization, carrying out consistency verification, if the consistency verification is passed, outputting and storing optimal configuration, and directly loading and executing in an operation stage; otherwise, returning or adjusting and recalibrating, and returning to the step S3. According to the method, the multiplexing distance between adjacent layers is shortened, and the round-trip overhead of intermediate activation between the video memory and the cache is reduced, so that the reasoning efficiency is improved.

A quantum non-demolition photon counter includes a quantum system, a processor, and a memory. The quantum system includes photons. The memory includes instructions stored thereon, which, when executed by the processor, cause the system to: couple an initial photonic state of the quantum system to an array of atoms in a |g-|e transition; store the photon(s) in the array using a classical control field acting on the |s-|e transition; oscillate the array between the states |s and |r, where |r is a Rydberg state; and perform a projective measurement of a presence of a Rydberg excitation by indirectly and progressively measuring a photon number n by directly and repeatedly measuring the presence of the Rydberg excitation after under Ĥ for a predetermined period of time. The array of atoms is initially in a ground state |g and |e is an excited state.

The embodiment of the invention provides a method for determining a damage identification model, and a damage identification method and device, and relates to the technical field of artificial intelligence, and the method comprises the steps: obtaining a plurality of sample groups; for the Tth sample in the sample group, domain invariant damage features are obtained through adversarial training of a feature sub-network and a domain discriminator; determining a reference decision action based on the splicing result of the domain invariant damage feature and the historical action vector by using the strategy sub-network; determining a state value based on the splicing result and the reference decision action by using a commentator network; calculating a loss value based on a preset loss function; and adjusting parameters of a feature sub-network, a domain discriminator, a strategy sub-network and a commentator network in a preset damage identification model based on the loss values, updating T = T + 1 and historical action vectors, stopping training of the sample groups until TM, and outputting the damage identification model when training of the plurality of sample groups is finished. Therefore, the damage identification model is accurately obtained, and the structural damage can be accurately identified.

A compound represented by the following general formula (1) is provided as a hole transport material useful as a photoelectric conversion element, and a photoelectric conversion element using the compound in a hole transport layer is provided.

The invention relates to an adaptive load balancing hybrid expert model training method and system. The method comprises the following steps: calculating the affinity of each token to each expert; in the first stage, an expert with the highest affinity degree is allocated to each token, discarding of the over-limit tokens is limited according to the capacity of the experts, and the token discarding rate and the affinity degree index of the current batch are calculated; in the second stage, the remaining experts are distributed to each token under the condition that capacity limitation is met through multi-round iterative distribution; and dynamically adjusting the numerical value of the next training batch based on the discarding rate of the current batch and the affinity index. By adopting the method, self-adaptive load balancing can be realized in the pre-training process of the hybrid expert model, and the training efficiency and the resource utilization rate are effectively improved.