437 results about "Log likelihood" patented technology

The invention discloses an LDPC (Low Density Parity Check) coding and decoding system and method for high-density tape storage, and the system comprises a main controller which is used for scheduling coding and decoding mode switching and module cooperative control; the storage and calculation integrated array is used for executing zero-carrying storage calculation; the ping-pong buffer framework is composed of an input buffer module Buffer A and an output buffer module Buffer B; the dynamic code length extension module is used for reconstructing a cascade link; the preprocessing module is used for converting the noisy code word into a log likelihood ratio (LLR) value; the hard decision module is used for converting the VN message into user data; the main controller is connected with all the modules through a control bus, and the dynamic code length extension module controls the power state and interconnection topology of the sub-arrays. According to the LDPC coding and decoding system and method based on the storage and calculation integrated array, zero-carrying calculation is achieved, iterative decoding is accelerated through a ping-pong buffer architecture, and the code length is flexibly configured through the dynamic code length extension module, the coding and decoding efficiency of high-density tape storage is improved, delay and energy consumption are reduced, meanwhile, multi-code-length self-adaption is supported, and high performance and flexibility are considered.

The invention provides an unsupervised industrial anomaly detection method based on an improved full-convolution cross-scale flow network. The method comprises the steps of performing data preprocessing on an industrial image data set; industrial image data is used as input, a pre-trained visual backbone network is used for extracting multi-scale features, a resolution-perceived channel attention module is used for enhancing expression of the multi-scale features, and an enhanced multi-scale feature map is obtained; performing multilayer reversible transformation by taking the enhanced multi-scale feature map as input and taking an ICSF-Net model based on hierarchical attention and expansion convolution as a cross-scale normalized flow network, and modeling multi-scale feature distribution; in the training stage, optimizing and updating ICSF-Net model parameters by maximizing the log likelihood of a normal sample under potential Gaussian distribution based on multi-scale feature distribution and combining an LSGR mechanism; in the reasoning stage, probability density estimation is carried out based on multi-scale feature distribution, an abnormal score graph is generated, and defect detection and positioning are achieved. According to the invention, the accuracy, robustness and pixel-level positioning precision of industrial product defect detection are improved.

The invention discloses a distribution line early-stage fault time sequence hidden Markov modeling and identification method and system based on multistage evolution characteristics, and belongs to the field of distribution line early-stage fault identification. Comprising the steps of obtaining a current waveform sample sequence of an early fault in a distribution line; for each current waveform sample in the sequence, extracting a multi-dimensional time-frequency feature, and constructing a feature vector of each sample; performing fault stage identification on each sample by using the first-level hidden Markov model group, and outputting a fault stage tag sequence corresponding to each sample; combining the fault stage label sequences of the current sample and a plurality of previous historical samples to form a stage label sequence window; and respectively inputting the stage label sequence window into a second-level tree line fault hidden Markov model and a second-level non-tree line fault hidden Markov model, calculating a corresponding first average log-likelihood value and a corresponding second average log-likelihood value, comparing the two average log-likelihood values, and judging whether a current sample belongs to a tree line early fault or not.

The invention provides a resistive random access memory ReRAM data judgment method based on sneak path correlation. The method comprises the implementation steps that parameters are initialized; calculating a reading resistance value of the ReRAM storage unit; classifying the ReRAM storage units based on the sneak path correlation; calculating a log-likelihood ratio and a judgment threshold of each type of storage unit; and obtaining a ReRAM data judgment result. The storage units are classified according to the sneak path correlation values of the row and the column where each storage unit is located, and the likelihood ratio function of each category of storage units is calculated according to the conditional probability distribution of the resistance value read by the storage data of each storage unit under different value conditions. According to the method for calculating the likelihood ratio function by using the sneak path correlation, the influence of the difference of interference suffered by different storage units on the bit error rate is fully considered, and the judgment precision is effectively improved.

A receiver configured to receive a signal may include a symbol, the receiver may include: at least one memory storing instructions; and at least one processor configured to execute the instructions to: generate a first posteriori log likelihood ratio corresponding to a bit in the symbol based on a channel log likelihood ratio corresponding to the bit; apply a first scaled value to a first extrinsic log likelihood ratio in the first posteriori log likelihood ratio; generate a first priori log likelihood ratio by selectively using a first reference value based on a comparison result; generate a second posteriori log likelihood ratio corresponding to the bit based on the first priori log likelihood ratio and the channel log likelihood ratio; and decode the signal based on the second posteriori log likelihood ratio.

The invention discloses a deformable electron beam lithography machine beam spot size calculation method and system, and the method comprises the steps: S1, employing a hyperbolic cosine function as a primary function according to a relation curve between the current intensity and the beam spot position, and constructing a composite function model through translation and superposition; s2, assuming that the observation value is a noise version of the composite function model, and establishing a likelihood function; a logarithm likelihood function is maximized and converted into a minimum residual sum of squares, partial derivatives of the minimum residual sum of squares about all parameters are solved, and a gradient equation set is obtained; s3, calculating the measurement signal, and extracting key feature points in the curve; calculating an initial parameter value according to the position of the feature point: taking the initial parameter as input, and iteratively solving a gradient equation set to obtain an exact solution of each parameter; and S4, calculating the actual size, the central position and the sharpness of the left and right edges of the electron beam spot according to the exact solutions of the parameters. The method has the advantages of high calculation precision and the like.

The invention discloses a dynamic self-adaptive NAND Flash reading method and system based on artificial intelligence, a medium and equipment, and relates to the technical field of solid-state storage, the method comprises the following steps: S1, initiating a reading request, reading page data by using default voltage, decoding the page data through an LDPC algorithm, constructing a feature vector and performing preprocessing, inputting the voltage into a pre-trained neural network model, and carrying out voltage prediction; s2, carrying out fine sampling on the predicted voltage, reconstructing accurate threshold voltage distribution, and calculating a log-likelihood ratio; s3, designing a shared bottom layer backbone network, enabling a neural network model to have independent output branches, predicting the voltage, and evaluating the long-term health state of the block; and S4, designing an online transfer learning mechanism, and enabling the neural network model to quickly adapt to a new environment by using a small amount of new data. According to the method, the delay is greatly reduced, the power consumption is remarkably saved, data loss is prevented through early warning of bad blocks, and the write amplification factor is reduced to 1.26 through dynamic read interference suppression.

The invention belongs to the technical field of intelligent detection and identification, and particularly relates to a contact network suspension state defect intelligent identification system, which comprises an abnormal index calculation unit, a defect category probability analysis unit, a data enhancement unit and a spatial proximity probability field inference unit, the abnormal index calculation unit is used for obtaining a normalized value and obtaining a composite abnormal index based on the normalized value and in combination with the position parameter; the defect category probability analysis unit is used for forming defect category probability distribution through logarithm likelihood function mapping; the data enhancement unit is used for obtaining a defect triple consisting of a defect category, a defect barycentric coordinate and a defect severity quantized value; and the spatial proximity probability field inference unit is used for automatically giving an alarm when the probability of any grid node exceeds an alarm threshold value. According to the invention, the defect detection rate, the positioning precision and the intelligent maintenance efficiency are obviously improved.

The invention discloses a diversity combining method and system for soft demodulation output, and relates to the technical field of wireless communication. The method comprises the following steps: receiving a plurality of antenna signals, and acquiring a signal-to-noise ratio of each signal receiving path as an input parameter; establishing a Taylor expansion model based on the relationship between the signal-to-noise ratio and the log-likelihood ratio, and searching an optimal polynomial fitting formula coefficient; calculating a weighting coefficient of each signal receiving path according to an optimization formula, wherein the weighting coefficient is generated by a polynomial function of a signal-to-noise ratio; and carrying out weighted summation on the log-likelihood ratios of all the signal receiving paths according to the calculated weighting coefficient. According to the method, the weighting coefficient of bit-level signal combination is optimized, and the optimal weighting coefficient is determined in a manner of Taylor expansion at a zero point, so that the defect that the diversity gain is reduced when the signal-to-noise ratio difference is relatively large in a traditional method is overcome, the bit error rate performance under a complex channel condition is remarkably improved, and the reliability and the anti-interference capability of a system are enhanced.

The invention discloses a soft decoding method and device, a medium and a product, and relates to the technical field of error correction coding, and the method is applied to a low-density parity check code, and comprises the steps: obtaining a target low-density parity check code with an information node and a check node containing an erasure bit; and allocating a preset non-zero log-likelihood ratio to the erasure bit on the check node, and executing first soft decoding. And if the first soft decoding fails, judging whether to carry out second soft decoding or not based on the weight of the syndrome generated by the first soft decoding and the number of the erasure bits on the information node. If it is judged that the second soft decoding needs to be carried out, a preset non-zero log likelihood ratio is distributed to the erasure bit located on the information node, and the second soft decoding is carried out. According to the method and the device, the erasing bits on different nodes are conditionally processed in stages, so that the success rate and the efficiency of decoding are effectively improved.

The invention relates to the technical field of intelligent traffic, in particular to a traffic abnormal event dynamic detection method and system based on trajectory analysis, and the method comprises the steps: receiving original space-time trajectory data; preprocessing the original spatio-temporal trajectory data, and mapping the original spatio-temporal trajectory data to a road section sequence of a road network through a map matching algorithm; constructing a behavior model based on the track sequence after map matching, learning the track sequence of the normal behavior mode on the road section, and establishing an observation emission model and a state transition matrix; in the online stage, the log-likelihood value of an observation sequence and / or the similarity between the observation sequence and a normal behavior pattern cluster are / is calculated for a real-time track in a set sliding window, and when the log-likelihood value and / or the similarity exceed a set threshold value, the track is marked as abnormal; performing time-space aggregation on abnormal trajectories of the same road section or intersection in unit time according to single vehicle abnormality judgment, and triggering group abnormal event alarm when aggregation data exceed a preset value.

The invention discloses a large model fine tuning method and system based on a suffix prompt learning method and ensemble learning, and the method comprises the steps: carrying out the preprocessing of an existing data set, obtaining K subsets through the replacement sampling, enabling each subset to account for a specified percentage of the total, and constructing an artificial prompt; according to the teacher model and the student model, a predicted target word classification score is obtained through suffix prompt learning, through probability normalization, teacher knowledge is distilled to the student model by using a KL divergence loss function, and a prompt vector is initialized; an input task vector is obtained through context learning, and knowledge of the input task vector is distilled to a student model prediction vector by using a mean square error loss function; constructing a mixed loss function, and optimizing knowledge distillation and task feature learning; performing end-to-end optimization on the initialized prompt vector by using suffix prompt in combination with a negative log-likelihood loss function; during reasoning, after the optimized prompt vectors are spliced to the input vectors, suffix prompts share the input vector key value cache to generate output. According to the method, the adaptability and reasoning performance of the model are remarkably improved.

The embodiment of the invention provides a data error correction method and device, electronic equipment and a computer readable storage medium, and relates to the technical field of storage, the bias reading voltage value of a storage unit of a memory is obtained, hard decoding is carried out according to the bias reading voltage value and a reference reading voltage value, and the data error correction efficiency is improved. The method comprises the following steps: acquiring a read voltage characteristic diagram of a memory, acquiring a reference log-likelihood ratio according to the read voltage characteristic diagram, a bias read voltage value and a reference read voltage value of the memory, if hard decoding fails, offsetting the reference read voltage value by the bias read voltage value in the read voltage characteristic diagram, and determining the CELL change quantity between the offset reference read voltage value and the reference read voltage value according to the read voltage characteristic diagram, and adjusting the reference log-likelihood ratio according to the CELL change number, and performing soft decoding according to the adjusted reference log-likelihood ratio. The log-likelihood ratio is adjusted according to the actual CELL change number, the more accurate log-likelihood ratio is obtained, the error correction performance of the LDPC is improved, the reliability of a memory is improved, and the service life of the memory is prolonged.

The invention discloses an error detection method and system in a programmed task video, relates to a video time sequence action segmentation and anomaly detection technology, and realizes robust error detection aiming at the problem of accuracy of error recognition in the programmed task video. According to the method, firstly, a time sequence action segmentation model is trained, a Gaussian mixture model is trained for each action category by utilizing each frame feature of an intermediate layer of the time sequence action segmentation model, the frame features are sent to the Gaussian mixture model of the corresponding action category to be processed, the log-likelihood score of the frame is obtained, and the detection of the frame level anomaly is realized. According to the method, the problem of error detection in the first-view programmed task video can be effectively solved by combining action segmentation and anomaly detection methods, and the method can be used for scenes requiring accurate recognition of sequence action errors.

The invention discloses a multi-time slot MAMP receiver and coding method for an MIMO multicarrier modulation system, comprising a time domain linear estimation module gamma l (.) and a symbol domain nonlinear estimation module phi l (.), where l represents the number of iterations; the time domain linear estimation module adopts a matched filter to recover the received time domain signal to obtain an estimated value of the time domain signal; transforming the estimated value into a symbol domain through unitary transformation; and the symbol domain nonlinear estimation module calculates a log-likelihood ratio of each symbol and calculates an estimated value of each symbol through estimation of the symbol domain returned by the time domain linear estimation module. According to the method, the optimal MSE detection performance can be realized with extremely low complexity in a multi-time-slot transmission scene.

The invention discloses an anti-interference underwater acoustic channel estimation method and device based on model data dual drive, and the method comprises the steps: obtaining an interference data set, and training a neural network; preprocessing the underwater acoustic signal to obtain a pilot frequency part signal, and initializing a channel, an interference sample, a sampling moment and a channel precedence variance gamma; respectively obtaining interference and channel prior scores at the t'moment by using a neural network and an analytical model; the logarithmic likelihood function obtains a gradient about interference and channel conjugation to obtain interference and channel condition scores, and the prior and condition scores are weighted and summed to obtain interference and channel posterior scores; the channel samples and the interference samples are corrected through annealing sampling, gamma is updated, and the posterior score is calculated again; a channel sample and an interference sample at the t'moment are obtained through the inverse process of the diffusion model; and if t'reaches tmin, outputting a final estimated value, otherwise, updating gamma according to the channel sample at the t 'moment, and circulating. According to the invention, accurate estimation of the single carrier communication channel in the structured interference environment is realized.

The invention provides a knowledge retrieval-oriented legal element recognition method and system, and the method comprises the steps: inputting to-be-processed target information and retrieved legal knowledge into a trained open source large language model, and carrying out the legal element recognition and knowledge reasoning through the open source large language model; in an open source large language model training process, estimating model parameters by using a maximum likelihood method, determining model parameters corresponding to maximization of a log-likelihood function, and updating the model parameters by using a gradient descent method; an open source large language model is utilized to deeply analyze and understand legal cases and laws and regulations, and comprehensive retrieval requirements for complex legal problems are met; and the model parameter accuracy is improved through a maximum likelihood method and gradient descent.

The invention discloses a subcarrier-level diversity combining method and subcarrier-level diversity combining equipment for multi-frequency diversity, and relates to the technical field of wireless communication. The method comprises the following steps: receiving a plurality of frequency points bearing the same service, and respectively carrying out orthogonal frequency division multiplexing demodulation on each frequency point to obtain a symbol estimation value; calculating a soft decision log-likelihood ratio and a reliability vector according to the symbol estimation value; aligning the time deviation, the carrier frequency deviation and the sampling rate deviation generated by frequency crossing of each frequency point based on each symbol estimation value; performing decorrelation processing on the aligned soft decision log-likelihood ratio to obtain a decorrelated soft decision log-likelihood ratio, and performing merging operation on the decorrelated soft decision log-likelihood ratio according to the aligned reliability vector to obtain a merged soft decision log-likelihood ratio; and decoding the merged soft decision log-likelihood ratio to obtain a decoding result, and adjusting the merging operation according to the decoding result. By implementing the technical scheme provided by the invention, the diversity combining accuracy can be improved.

The invention relates to the technical field of communication and computer instruction execution and processing, in particular to a polar code encoding and decoding algorithm-oriented special instruction set processor and an implementation method, and a vector processor has efficient parallel computing capability. A coding and decoding algorithm of a polar code relates to parallel computing of a large number of log-likelihood ratios, a vector processor of a special instruction set can process a plurality of LLR values at the same time through a parallel processing unit, and the throughput rate is remarkably improved; special instructions such as polarization shuffling and vector G operation instructions are designed for polarization code encoding and decoding and are used for accelerating algorithm operation. The method can achieve the balance among the performance, the energy efficiency and the flexibility, can adapt to the continuous iteration upgrading of the algorithm due to the programmable and reusable characteristics, and has a good application prospect.

The invention relates to a database anomaly detection method and system based on a GMM (Gaussian Mixture Model) algorithm and a medium, and the database anomaly detection method based on the GMM algorithm is applied to a database data auditing tracking platform and comprises the following steps: inputting basic information; auditing information is added on the SQL statement through a Hint label, and the SQL statement is executed in a proxy mode; analyzing the log and extracting data on the Hint tag; audit and data track information is constructed, stored and recorded; collecting user behavior feature data, and training for each user to obtain a behavior GMM model of the user; and calculating a log-likelihood value of the user behavior data point under the GMM model, and judging whether the log-likelihood value is smaller than a set threshold value or not. According to the method, various business data in the database can be audited and tracked effectively in real time, and abnormal operation of the user database can be detected.

At least one example embodiment provides a radio access network element comprising at least one processor and at least one memory. The at least one memory stores instructions that, when executed by the at least one processor, cause the radio access network element to: extract sign information from a plurality of log-likelihood ratio (LLR) sample values included in LLR data in a resource block group (RBG); convert negative LLR sample values, from among the plurality of LLR sample values, into positive integer values to obtain converted LLR data; generate compressed LLR data based on the converted LLR data; and output the compressed LLR data.

The invention belongs to the technical field of intelligent task-based dialogue, and particularly relates to a task-based dialogue strategy learning method and system based on reinforcement learning, which adopts a soft actor commentator algorithm to be combined with behavior cloning and maximum entropy inverse reinforcement learning to relieve a cold start problem. According to the method, maximum entropy inverse reinforcement learning is utilized, a reward value is calculated, and a user target in a dialogue is accurately deduced according to maximum log likelihood estimation of a human dialogue and a simulated dialogue; a potential reward function is extracted from a successful track through inverse reinforcement learning, manual reward design is replaced, accumulated reward maximization is pursued during strategy optimization, diversity and exploratory performance of the strategy are encouraged, and the dialogue strategy is prevented from being converged to a single mode too early.

A modem chip may include a turbo decoding circuit configured to receive a plurality of channel log likelihood ratios respectively corresponding to a plurality of bits included in a symbol, generate a plurality of first posteriori log likelihood ratios respectively corresponding to the plurality of bits in an Nth iterative loop based on the plurality of channel log likelihood ratios, wherein N is greater than or equals to 1, and generate a first input / output counting value by counting a number of first posteriori log likelihood ratios, among the plurality of first posteriori log likelihood ratios, that differ in sign from the corresponding channel log-likelihood ratios, and a decoding control circuit configured to stop decoding for the plurality of bits in the turbo decoding circuit based on the first input / output counting value being greater than or equal to a first threshold.

State of art techniques using moderate sized Language Models (LMs) for text classification need fine-tuning or in-context learning. A method and system providing a two-step classification using moderate-sized (#params≤2.7B) causal LM (Gen AI) is disclosed. Firstly, for a text instance to be classified, a set of perplexity and log-likelihood based features are obtained from an LM. Further, a light-weight classifier is trained in the second step to predict the final label. The system enables a new way of exploiting the available labelled instances, in addition to the existing ways like fine-tuning LMs or in-context learning. It neither needs any parameter updates in LMs like fine-tuning nor it is restricted by the number of training examples to be provided in the prompt like in-context learning. The key advantages of the disclosed system are explainability through most suitable key phrases and its applicability in resource poor environment.

The invention provides a polar code decoding method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining a plurality of log likelihood ratio LLR values corresponding to a to-be-decoded bit sequence; for a current decoding layer in a preset multi-layer decoding architecture, circularly executing the following steps until an intermediate LLR value output by the first decoding layer is determined: taking the acquired LLR values as LLR values input by the last current decoding layer, and taking the intermediate LLR value output by the next decoding layer after the non-last current decoding layer as the LLR value input by the non-last current decoding layer, inputting the LLR value input by the current decoding layer into a judgment unit set by the current decoding layer, performing nonlinear processing based on a middle LLR value output by the judgment unit, and determining the middle LLR value obtained by processing as the middle LLR value output by the current decoding layer; based on the middle LLR value output by the first decoding layer, the bit sequence after decoding is determined, and consumption of software and hardware resources in the decoding process is smaller.

The invention discloses a single-carrier signal detection method and device based on an expansion factor graph and Gaussian interference, relates to the field of digital signal processing, solves the problem that a traditional single-carrier frequency domain equalization technology and a message passing algorithm are invalid in an SCMA system, and is characterized in that a signal model is established according to a received signal; modeling the interference term as a Gaussian random variable distribution model to obtain Gaussian information; constructing an expansion factor graph; performing function node updating to obtain transmission information; the transmission message is transmitted between the function node and the variable node according to the expansion factor graph; performing variable node updating on the transmitted information to obtain a normalized variable node message; returning the variable node message to the function node for iterative updating until a preset number of times is reached; after the last iteration is completed, calculating and outputting the log-likelihood ratio of the bit stream according to the variable node information obtained after the iteration update; and the effects of remarkably reducing the calculation burden and remarkably improving the detection performance are achieved.

In some example embodiments, there may be provided a method that includes receiving, by a machine learning encoder as part of a training phase, channel state information as data samples; generating, by the machine learning encoder, a latent variable comprising a log likelihood ratio value representation for the channel state information, wherein the latent variable provides a lower dimension binary representation when compared to the received channel state information to enable compression of the received channel state information; generating, by the binary sampler, a binary coding value representation of the latent variable, wherein the binary coding value converts the latent variable to a binary form; and generating, by the machine learning decoder, a reconstructed channel state information, wherein the generating is based in part on the binary coding value representation of the latent variable generated by the binary sampler. Related systems, methods, and articles of manufacture are also disclosed.

The invention discloses an unmanned aerial vehicle identification method and device based on category specific modeling, equipment and a storage medium, and relates to the technical field of mode identification based on digital data processing, and the method comprises the steps: obtaining a radio frequency signal of an unmanned aerial vehicle, and generating a time-frequency graph through short-time Fourier transform; inputting the time-frequency graph into a target residual network, and extracting feature vectors of the unmanned aerial vehicle; calculating a log-likelihood value of the feature vector under each model by using a Gaussian mixture model constructed for the known class of each unmanned aerial vehicle in the training stage, and obtaining a preliminary judgment result according to the maximum log-likelihood value and a preset adaptive threshold value; and for the samples which are preliminarily judged as the unknown class, calculating the minimum feature distance between the samples and all samples in the known class unmanned aerial vehicle sample feature matrix, and if the distance is smaller than a preset recovery threshold value, re-classifying the samples as the known class unmanned aerial vehicle to which the corresponding nearest neighbor sample belongs. According to the invention, known unmanned aerial vehicles can be accurately identified in an open environment, and unknown unmanned aerial vehicles can be effectively distinguished.

The invention relates to a method for predicting intraoperative hypertension, and belongs to the field of digital medical treatment. Comprising the following steps: preprocessing, standardizing and marking collected patient data to obtain first data; selecting characteristic variables from the first data, and selecting influence characteristics from the characteristic variables according to correlation coefficient calculation and expert judgment; establishing a sample set based on the influence characteristics, and training through the sample set to obtain a trained intraoperative hypertension prediction model; preprocessing and standardizing data of a patient to be tested to obtain second data, and inputting the second data into the trained intraoperative hypertension prediction model to realize prediction of intraoperative hypertension of the patient. According to the method, the influence factors of intraoperative hypertension are searched from a large amount of data, and the intraoperative hypertension prediction model is trained based on the log-likelihood function, so that the accuracy of intraoperative hypertension prediction is improved, and the problem that current intraoperative hypertension prediction depends on clinical experience of doctors is solved.

The present disclosure provides a system and a method for implementing rate recovery in the PUSCH and PDCH bit rate processing chain of network. The system packs log likelihood ratios (LLRs) data in such a way that for each equalized in phase and quadrature (IQ) symbols, a predetermined number of LLRs equal to the modulation order are packed. The system de-interleaves the packed LLRs by reading the most significant bit (MSB) LLRs row wise for the number of columns equal to the modulation order. The system uses a single buffer for de-interleaving, bit-deselection, and filler bit addition stages, thereby reducing memory requirement of the system. The system processes only a predetermined number of LLRs to a HARQ combining stage to optimize memory and reduce latency.