62 results about "Early stopping" patented technology

A video encoder and corresponding methods are provided for selecting the mode of a current macroblock of an inter-coded frame, including one or more of checking first modes for a subset of macroblock modes, selectively checking other modes in response to motion vector information of the checked first modes, and selecting the mode for the current macroblock in response to the checked modes; checking the macroblock mode of at least one neighboring macroblock, and selecting the mode for the current macroblock in response to the macroblock mode of the at least one checked neighboring macroblock; checking the cost of a subset of macroblock modes, further checking only intra-coded modes if the checked cost meets a preset criteria, and selecting the mode for the current macroblock in response to the checked modes; and adjusting an early-stopping threshold in response to checked macroblock modes, and selecting the mode for the current macroblock in response to the checked macroblock modes if the adjusted early-stopping threshold is met.

An efficient motion estimation with an accurate starting point prediction and a scalable searching range is disclosed. A storage device saving MVs and SADs of an entire frame of the nearest neighboring frame and surrounding blocks is implemented. The majority or an average of MVs of the surrounding blocks and the corresponding position of at least one nearest neighboring frame is selected to be the starting point of the best match block full search. A threshold value is determined to early stop the calculation of the best match block search. Depending on the MV values of the surrounding blocks and a corresponding block in a nearest neighboring frame, pixels within a calculated scalable searching range are moved into a smaller on-chip searching range buffer from a larger reference frame buffer.

The invention discloses a dynamic coordination and control method of traffic signals of an urban main road. Aiming at characteristics of traffic low of the urban main road of our country, bidirectional vehicle flows of the main road continuously pass through by coordination of the traffic signals on the precondition of ensuring green light utilization time of each intersection and phase on the main road. According to a real-time traffic condition, public signal period time, a phase green signal ratio of each intersection and a phase difference between adjacent intersections can be dynamically calculated; and starting time of a coordination phase is optimized on line, and green conflicts of a bidirectional coordination phase of the main rod are avoided through early stopping and delayed emission of a left-turning signal in the coordination phase. According to the dynamic coordination and control method, average traveling time and an average stopping rate of the traffic flow of the main road can be effectively reduced, and an application effect is better than that of a tradition single-point timed control method and a static coordination control method, so as to provide an effective control method for dynamic coordination and control of the traffic signals of the urban main road.

In order to allow early stopping of codeblock decoding iterations, a cyclic redundancy check (CRC) is attached to each codeblock segment that pertains to the same transport block carrying information bits. The CRC for segment k is calculated for all bits within segments 1 to k. This allows also identifying cases of wrongly assumed CRC check results for segments 1 to k when the CRC attached to segment k+1 is evaluated.

The invention relates to an image single classification method based on a generative confrontation network. The image single classification method based on a generative confrontation network includesthe steps: constructing a generator in the generative confrontation network by means of a dense connection block structure; constructing a discriminator in the generative confrontation network; inputting positive sample training data, and using a gradient punishment algorithm to train the generative confrontation network; according to the classification effect of the model on the verification setduring the training process, adjusting the network parameters, and using the Early Stopping strategy to find the classification optimal iteration number of the model; and after the model training is completed, using the discriminator in the generative confrontation network to test the test set data, and determining the classification effect of the model through a classification recall index CRI determination model. The image single classification method based on a generative confrontation network can automatically generate a negative sample set, and can solve the problem that in a current single classification method, artificial construction of a negative sample data set is likely to cause over-fitting of the classifier.

A method and apparatus for controlling iterative decoding in a turbo decoder are provided, in which a maximum number of iterations is determined for current data to be decoded based on at least one of current HARQ information necessary for a HARQ operation of the current data, previous HARQ information about previous data, and early stop information indicating whether iterative decoding of the previous data was early stopped. A turbo decoder iteratively decodes the current data within the maximum number of iterations.

The invention discloses an HEVC intra-frame prediction coding method and system, and belongs to the technical field of the video coding technique. The method comprises the steps that each image frame is segmented into largest coding units (LCUs) which are not overlapped; the area smoothness of a current LCU is calculated; the similarity between the current LCU and an adjacent LCU is calculated; self-adaption depth prediction is performed according to the similarity between the current LCU and the adjacent LCU; the depth of the current LCU is determined according to the smoothness of the current LCU and the similarity between the current LCU and the adjacent LCU, and coding is performed according to the depth of the current LCU. By the adoption of the HEVC intra-frame prediction coding method and system, CU partition is stopped in advance by early stopping prediction or a certain layer of partition is omitted; meanwhile, as the number of intra-frame prediction modes in an RDO process is reduced in a PU layer, a CU layer and the PU layer are optimized and integrated together, implementation complexity can be reduced, and HEVC intra-frame prediction time is greatly shortened.

Embodiments generate a model of demand of a product that includes an optimized feature set. Embodiments receive sales history for the product and receive a set of relevant features for the product and designate a subset of the relevant features as mandatory features. From the sales history, embodiments form a training dataset and a validation dataset and randomly select from the set of relevant features one or more optional features. Embodiments include the selected optional features with the mandatory features to create a feature test set. Embodiments train an algorithm using the training dataset and the feature test set to generate a trained algorithm and calculate an early stopping metric using the trained algorithm and the validation dataset. When the early stopping metric is below a predefined threshold, the feature test set is the optimized feature set.

The invention provides a cooperative priority control method of double tramcars at a level crossing under a non meeting state. The invention provides a phase position control method of signal lights when two tramcars reach the level crossing in the states except for the meeting state by considering the condition that the tramcars and social vehicles run together at the level crossing. According to the invention, by separating conflicting vehicles based on time, obeying the principle of let the tramcars relatively preferentially go and vehicles free from the conflict go simultaneously, and adopting the control strategies of green-light extension, red-light early stopping, phase position insertion and phase position crossing, mutual conflicts of two types of vehicles are reduced and passing efficiency of the level crossing is improved.

The present invention relates to a general-purpose analogical reasoning system. More specifically, the present invention relates to a high-performance, semantic-based hybrid architecture for analogical reasoning, capable of finding correspondences between a novel situation and a known situation using relational symmetries, object similarities, or a combination of the two. The system is a high-performance symbolic connectionist model which multiplexes activation across a non-temporal dimension and uses controlled activation flow based on an analogical network structure. The system uses incremental inference to stop inference early for object correspondence, uses initial mappings to constrain future mappings, uses inferred mappings to synchronize activation, and independent mapping based on roles, superficial similarity, or composites.

The invention discloses an early stopping method of low-density parity-check (LDPC) code alternating direction multiplier method (ADMM) iterative decoding and aims at mainly solving the problem of slow convergence rate of ADMM iterative decoding in a low signal to noise ratio region in the prior art. A realization method comprises the following steps of initializing a decoding parameter; calculating variable node initial information; updating variable node information; updating an auxiliary variable; updating a lagrange multiplier vector; acquiring hard decision values of codeword bits; calculating a total amount of codeword bits which change between every two continuous iterations in the ADMM iterative decoding; calculating a hard decision change rate of the codeword bits; stopping decision by decoding; and stopping decoding. According to the early stopping method provided by the invention, whether iteration can be stopped as soon as possible or not can be judged according to the sizeof the hard decision change rate (CRHD) of the codeword bits between every two continuous iterations in the ADMM iterative decoding process, so that average iteration times of decoding are reduced, and the decoding speed is increased.

The invention discloses a cross-project software defect prediction method for supervised expression learning. The method comprises the following steps: (1) selecting a defect data set, and preprocessing defect data; (2) training a migration auto-encoder in an unsupervised pre-training mode, wherein the migration auto-encoder comprises a feature coding layer and a label coding layer; (3) with the help of a migration cross validation method, selecting a sample closest to the hidden layer feature distribution of the target project sample from all sample hidden layer feature representations of thesource project as a validation set, and taking the rest as a training set; (4) performing oversampling processing on the training set sample; (5) finely adjusting the migration auto-encoder, and selecting a model hyper-parameter and an early stop strategy; and (6) inputting the preprocessed data of the target project into a migration auto-encoder, and obtaining a final prediction result through the output of a label encoding layer. According to the method, the label information of the source project sample is introduced into the feature representation learning process, so that the predictionperformance of the cross-project software defect prediction model is improved.

The invention provides a power generation device, which comprises an on-off valve; a pressure-sharing flow path, wherein in a circulation flow path, the part between the on-off valve and an evaporator is communicated with the part between a screw expander and a condenser; an expander bypass flow path, wherein in the circulation flow path, the part between the on-off valve and the screw expander is communicated with the part between the screw expander and the condenser; a controller used for turning off the on-off valve upon the shutdown of the device and then opening the bypass valve and the pressure-sharing valve of the expander after the shutdown of a pump action pump. In this way, after the stopping of the circulation of the action medium, the essential early stop of the power generation device is realized. Therefore, no inverse push force is exerted on the screw expander.

The invention provides a cooperative priority control method of double tramcars at a level crossing under a non meeting state. The invention provides a phase position control method of signal lights when two tramcars reach the level crossing in the states except for the meeting state by considering the condition that the tramcars and social vehicles run together at the level crossing. According to the invention, by separating conflicting vehicles based on time, obeying the principle of let the tramcars relatively preferentially go and vehicles free from the conflict go simultaneously, and adopting the control strategies of green-light extension, red-light early stopping, phase position insertion and phase position crossing, mutual conflicts of two types of vehicles are reduced and passing efficiency of the level crossing is improved.

Methods and apparatus are provided for utilizing early stopping rules for non-binary turbo codes. Data can be received from a channel and decoded using a non-binary turbo code to generate extrinsic information and a posteriori information. The generated extrinsic information can be evaluated using an early stopping rule. A hard decision on a value of the received data can be made using the a posteriori information if the early stopping rule is satisfied.

A click stream estimation method of a neural network comprises the following steps: 1) collecting a large amount of historical behavior data of a user, the data including possible features helpful toclick stream estimation, such as advertisement commodity information, user information, context information and store information, and constructing a data set on the basis of the features; 2) constructing a weight matrix decomposition cross neural network model of the associated features, wherein the model comprises a logistic block, a word embedding block, a word embedding vector weight cross block and a hidden layer block; and 3) segmenting the data set into small blocks, sequentially inputting the small blocks into the weight matrix decomposition cross neural network of the associated features, and updating parameters by using a back propagation Adam algorithm until the parameters meet early stop condition convergence; and 4) completing click prediction of the advertisement commodity bythe user in an actual system. Relatively high-dimensional data is mapped to a low-dimensional word vector through a word embedding technology, so that the calculation amount is reduced, and learningof a neural network is facilitated.

An apparatus and a method for raising a data throughput by dynamically changing the maximum number of iterative decoding times of an iteratively decodable code in a mobile communication system are provided. The apparatus includes an early stop controller for determining whether an early stop condition is satisfied during a decoding process according to a maximum number of the iterative decoding times, a Connection IDentifier (CID) controller for detecting a CID of a Media Access Control (MAC) header after the early stop, and an iteration controller for, when determining that a Protocol Data Unit (PDU) is allocated to the receiver apparatus by detecting the CID of the MAC header, resetting the maximum number of iterative decoding times after identifying a decoding block of a next PDU, and for, when the CID is not detected, resetting the maximum number of the iterative decoding times after identifying a next decoding block of the current PDU.

The present invention is aimed to provide oligonucleotide sequence of target activated chronic myeloid leukemia protein kinase PKR and retrovirus dual expression vector possessing the oligonucleotide sequence. The complementary sequence of 20bp SEQ1 and SEQ2 are designed according to BCR-ABL b3a2 type mRNA fusion point upper and down stream, enzyme-cutting point is also designed, a pair of nucleotides T-A is changed to C-G for preventing early stop of transcription.The recombinant retrovirus vector is transformed into K562 cell line of chronic granulocytic leukemia of blast period. The transferred gene is transcribed and hybridizes with BCR-ABL mRNA to form double-chain RNA, activates PKR targetedly, results in K562 cell apoptosis and have no effect on normal cell. The effective ingredient nucleotides sequence is prepared for clinical drug and can treat chronic granulocytic leukemia.

An apparatus and a method for raising a data throughput by dynamically changing the maximum number of iterative decoding times of an iteratively decodable code in a mobile communication system are provided. The apparatus includes an early stop controller for determining whether an early stop condition is satisfied during a decoding process according to a maximum number of the iterative decoding times, a Connection IDentifier (CID) controller for detecting a CID of a Media Access Control (MAC) header after the early stop, and an iteration controller for, when determining that a Protocol Data Unit (PDU) is allocated to the receiver apparatus by detecting the CID of the MAC header, resetting the maximum number of iterative decoding times after identifying a decoding block of a next PDU, and for, when the CID is not detected, resetting the maximum number of the iterative decoding times after identifying a next decoding block of the current PDU.

The invention discloses an etching method and an etching system, wherein the method comprises the following steps of: after main etching is stopped, judging whether etching stopping points are caught when the main etching is stopped, and if not, entering an over-etching step for continuously catching the etching stopping points; when the over-etching step reaches the preset maximum over-etching time, after the etching stopping points are caught and normal over-etching time is passed, automatically stopping; after the over etching is stopped, judging whether etching stopping points are caught or not before the over etching is stopped, and if so, obtaining a correction value of the maximum main etching time according to the etching thickness and the maximum main etching time; and adopting the correction value of the maximum main etching time to carry out next-time etching process. According to the embodiment of the invention, by the introduction of a judgment mechanism and a time feedback mechanism and control for the etching mode of etching equipment, the problems of the etching residual and slightly-large device CDs (Compact Discs) and the like after over-etching due to over-early stopping of the main etching in the prior art are solved.

The invention provides a federal learning model training method and device, electronic equipment and a storage medium, and the training method comprises the steps: carrying out the sample alignment with a data provider server; generating a feature code set, and sending the feature number and the public key of the data provider server to the data provider server; dividing a current sample into a training set and a verification set, and obtaining a parameter set of the federated learning model; performing M times of iterative training on the federal learning model according to the training set, the verification set, the parameter set and the feature coding set; and in the process of each iteration training in the M iteration training, if it is judged that the federated learning model meets the early stop condition, controlling the M iteration training to stop early, and obtaining target parameters of the federated learning model obtained by the last iteration training in the M iteration training. Therefore, training can be more efficient, meanwhile, the modeling effect is improved, an early stop strategy is adopted, overfitting of the model is avoided, and meanwhile complexity is reduced.

The invention discloses an attack stage prediction method based on LSTM and attacker information, and belongs to the field of attack prediction. The method comprises the following steps: collecting warning information of network attacks in a long period of time through an LSTM system; collecting historical information of attackers in a large amount of warning information; preprocessing the historical data to construct a training set, a verification set and a test set required by LSTM model training; then training an LSTM model by using the training set, and determining whether to stop learning of the LSTM on the training set in advance by using the loss of the verification set; and enabling the finally obtained model to be capable of predicting the preprocessed input data, and obtaining the step that the next attack in the multi-stage network attack in the future through prediction.

The invention discloses a slope earthquake slip prediction method and system based on an artificial neural network, and aims to construct an explicit expression for predicting the permanent displacement of a slope through an earthquake oscillation spectrum acceleration and a slope yield acceleration by using a large number of earthquake waves and Newmark slide block analysis. The method comprises the steps of firstly constructing a displacement prediction network model; then selecting seismic waves from a seismic oscillation database NGA-West2 and calculating permanent displacement of the side slope under different conditions; coupling an early stop technology and a 5-fold cross validation technology to select an optimal model hyper-parameter; then, optimal hyper-parameters are configured for the displacement prediction network model, final training is carried out, and the performance of the displacement prediction network model is evaluated; and finally, predicting the permanent displacement under a given earthquake working condition and a slope condition. Based on the method, the invention develops and discloses three slope earthquake slip prediction models with good accuracy, universality and practicability, and has significant guiding significance for slope stability evaluation and aseismic design under the action of an earthquake.

The present invention provides multivariate methods for analyzing microarray gene expression data of high dimensional space and thereby identifying differentially expressed genes. The methods of this invention provide a random search procedure with multiple starts and early stop. Larger sets of differentially expressed genes may be identified using the methods of this invention starting from feature spaces of smaller dimensionality where accurate estimates on covariance matrix can be made.

The present invention provides multivariate methods for analyzing microarray gene expression data of high dimensional space and thereby identifying differentially expressed genes. The methods of this invention provide a random search procedure with multiple starts and early stop. Larger sets of differentially expressed genes may be identified using the methods of this invention starting from feature spaces of smaller dimensionality where accurate estimates on covariance matrix can be made.

The invention relates to a circuit breaker switching-on and switching-off control method. The method includes the following steps that: (1) circuit breaker switching-on and switching-off instructions emitted by a power distribution system are detected, and switching-on pulses are generated based on the switching-on and switching-off instructions; (2) the information of the positions of the switching-on and switching-off of a circuit breaker is obtained, the output duration of switching-on pulses and switching-off pulses is recorded, and the output duration of the switching-on pulses and the switching-off pulses is compared with the current time of the switching-on and switching-off of the circuit breaker; and (3) whether the positions of the switching-on and switching-off of the circuit breaker are qualified is judged, if the positions of the switching-on and switching-off of the circuit breaker are qualified, a driving instruction is emitted so as to drive corresponding switching-on and switching-off coils of the circuit breaker to act, if the positions of the switching-on and switching-off of the circuit breaker are unqualified, the switching-off coil of the circuit breaker acts. With the circuit breaker switching-on and switching-off control method of the invention adopted, it can be ensured that the output duration of the pulses will not be smaller than the required action time of the permanent magnet circuit breaker, and therefore, the failure of switching-on and switching-off caused by the early stopping of the pulses which is further caused by errors of detected switching-on and switching-off position information can be effectively avoided.

The invention discloses a model training method. The method comprises the steps: obtaining multiple groups of hyper-parameters, and constructing models by utilizing each group of hyper-parameters; respectively training the plurality of constructed models by using a training set, and verifying the model being trained by using a verification set; in response to triggering early stop, obtaining evaluation parameters generated when the model being trained is verified, and obtaining a standard parameter from the plurality of evaluation parameters; judging whether the standard parameter is greater than a threshold; in response to the condition that the standard parameter is not greater than the threshold, obtaining a reciprocal of a loss function value corresponding to the currently trained model, and determining a plurality of models which continue to be trained from the models being trained according to the obtained reciprocal of the loss function value and the corresponding evaluation parameters; and in response to the situation that the number of the continuously trained models is greater than 1, returning to the step of training until the number of the continuously trained models isequal to 1. The invention further discloses computer equipment and a readable storage medium.

The invention provides a confidence propagation dynamic flipping decoding method based on a log-likelihood ratio, which can judge each iteration decoding result in advance through bit flipping in combination with a cyclic redundancy check auxiliary early stop standard when confidence propagation algorithm decoding is carried out on a polarization code, and after the preset maximum number of iterations is reached and the CRC verification fails, a bit-flipping decoding program is carried out to perform bit flipping until the CRC verification is passed in advance or the preset maximum number of flipping is reached. According to the method, the number of times of multi-bit flipping attempts can be controlled and reduced under the condition that the flipping bits can be flexibly adjusted, the polar code decoding performance is improved, and the decoding complexity is reduced. A simulation result shows that compared with a BP algorithm of a polarization code based on information post-processing and an original BP algorithm, the BP algorithm has obvious performance gain.