18 results about "Nonlinear neural networks" patented technology

The invention discloses a flotation froth image feature extraction method based on adaptive multi-scale weighted fusion, and belongs to the technical field of intelligent monitoring and image processing in the mineral flotation process, and the method comprises the steps: fusing a color histogram, texture and edge features, and depth features extracted by VGGNet, GoogLeNet and ResNet; and mapping and fusion of multi-modal features are realized by using a self-adaptive multi-scale weighted feature fusion network. And dynamic weighting and fusion are carried out on different characteristic modes through a nonlinear neural network model, and unified characteristic representation is generated to be used for accurate representation of the flotation froth state. The method has the advantages of high recognition precision, high robustness and flexible deployment, online monitoring, optimal control and index prediction of the flotation process can be effectively supported, and reliable technical support is provided for an industrial field.

The invention discloses a secure neural network inference method and system based on linearization during inference, and relates to the technical field of machine learning and information security. A model holder deploys a random feature extraction module of a trained nonlinear neural network at a client, and distributes learnable modules to two non-serial servers; the client inputs the data to a random feature extraction module for feature extraction to obtain random features, the mapped random features are segmented based on arithmetic secret sharing to obtain a first secret share and a second secret share, and the first secret share and the second secret share are respectively distributed to the two servers; the two servers conduct reasoning calculation on the received secret share based on the learnable module, during reasoning, an S-cos activation function is linearized, a first calculation result and a second calculation result are obtained and returned to the client side, and the client side reconstructs a prediction result based on the first calculation result and the second calculation result. And an efficient neural network reasoning service is realized on the premise of protecting data privacy and model privacy.

The invention discloses a double-mechanical-arm model prediction control method and system with disturbance compensation. The method comprises the steps that motion parameters formed between double mechanical arms and an object are collected, and kinematics and dynamics relation solving is conducted through the motion parameters; solving by utilizing a relation between kinematics and dynamics, and designing a double-mechanical-arm hybrid force controller; according to the double-mechanical-arm hybrid force controller, a nonlinear neural network variable structure controller is adopted to conduct Lyapunov model predictive control; through online optimization of model prediction control, system constraints are effectively enhanced, and then the robustness of the system and the performance and stability of a controller are remarkably improved; meanwhile, the overall performance of the system is further improved by combining approximation and compensation of the neural network to white noise.

The invention relates to the technical field of software security and program protection, in particular to a code obfuscation method based on a neural arithmetic unit, which comprises the following steps: step 1, code acquisition and operation verification; step 2, code deep analysis and detailed solution; step 3, model construction and training export; step 4, extracting and confirming confusion parameters; 5, generating and replacing obfuscated codes; step 6, running reasoning and verification; according to the method, an original arithmetic instruction in a program is replaced by a neural arithmetic unit subjected to constraint training, original clear arithmetic semantics are completely embedded into a high-dimensional and nonlinear neural network parameter space, and an attacker faces confused codes, so that the algorithm semantics are completely embedded into the high-dimensional and nonlinear neural network parameter space; the original calculation intention cannot be recovered through the traditional algebraic simplification or logical reasoning means, and the essential improvement from grammar confusion to calculation semantic confusion constructs a deeper defense barrier, and can effectively cope with various advanced reverse analysis technologies including intelligent decompilation.

Various example embodiments relate to mitigation of a non-linearity in a data communication chain. A method may include: obtaining a data communication signal including a plurality of complex-valued input samples; capturing, from the plurality of complex-valued input samples, a current sample and a set of delayed samples; generating a phase-normalized input signal based on normalizing phase of the current sample and the set of delayed samples by a normalization term, wherein the normalization term is common for the current sample and the set of delayed samples; providing the phase-normalized input signal to a neural network configured to mitigate non-linearity of a data communication chain and to output a complex-valued output sample for each of the plurality of complex-valued input samples; and denormalizing phase of the complex-valued output sample by a denormalization term configured to restore phase of the data communication signal.

The application discloses a kind of seabed hydrate induced regional geological disaster chain analysis method, the method comprises: according to hydrate occurrence characteristics, hydrothermal condition and geotechnical exploration data, establish site geological model;Subsequently, through random selection parameter combination through numerical solver calculation hydrate thermal decomposition space-time evolution process;Using nonlinear neural network to construct proxy model, realize the regional rapid prediction of hydrate decomposition space-time evolution in any region.Based on the predicted regional space-time evolution result, the safety factor of hydrate occurrence area is calculated, the maximum sliding surface depth is determined, and further combined with failure dynamics model, the regional disaster chain evolution process after landslide occurs is evaluated.The application realizes the rapid simulation of hydrate decomposition induced geological disaster chain evolution under different occurrence conditions by constructing the integrated model of "occurrence-disturbance-instability", which has important theoretical reference value and practical application significance for deep sea engineering site selection and risk assessment.

Methods and apparatus are provided for a deep learning-based polarization coding scheme. Methods and apparatus are provided in which a processor of an electronic device encodes a segment of a binary message word into a real-valued outer codeword using a non-linear neural network (NN) outer encoding process. The processor combines the real-valued external codewords using a real-field polarization operation to generate codewords of the binary message word.

The application discloses a method for generating nonlinear collaborative financial product alpha factors by using reinforcement learning, relates to the fields of financial quantification and reinforcement learning, and comprises the following steps: step 1, outputting a token of BEG, inputting a current environment state as a feature into a policy gradient network, and outputting a next token; step 2, updating the current environment state, inputting the current environment state into the policy gradient network again, and outputting a next token; step 3, repeating step 2 until a complete alpha factor formula is generated; step 4, placing the complete alpha factor formula into an alpha factor pool, assigning a random weight, inputting all factors in the alpha factor pool into a nonlinear neural network for regression training, and determining the weight of each alpha factor; step 5, composing a total alpha factor with a collaborative effect, and taking a mutual information coefficient as a reward function; step 6, updating the policy gradient network according to the reward function by using a Monte Carlo method; and repeating steps 1 to 5 until an effective prediction result is obtained.

This invention provides a method and system for analyzing the safe operation of grid-connected offshore wind farms. The method includes: acquiring the network topology and operating parameters of the grid-connected offshore wind farm to construct a power flow feasible domain injection space and a node admittance matrix; obtaining the power flow feasible domain based on these two parameters to determine stability criteria; generating an operating sample set by determining the active power injection space of the grid-connected offshore wind system and the number of training samples; and combining the operating sample set and stability criteria to obtain a stability margin set, which is then used to train a margin prediction model under a nonlinear neural network for hyperparameter selection and early shutdown strategy training. Finally, the active power injection space of the grid-connected offshore wind system is used to obtain an operating condition set under the power flow feasible domain, and a stability margin prediction set is obtained under the margin prediction model to construct the low-frequency oscillation safety domain boundary of the grid-connected offshore wind system for safe operation analysis. This invention can improve the accuracy of safe operation analysis of grid-connected offshore wind farms.

The application discloses a large-scale MIMO deep unfolding precoding method and device based on a conjugate gradient method, and belongs to the wireless communication field, wherein the method comprises the following steps: obtaining a downlink air interface channel vector and a transmission vector; inputting the downlink air interface channel vector and the transmission vector into a pre-trained nonlinear neural network to obtain a precoded transmission signal. The nonlinear neural network is a combination of a deep unfolding multi-layer nonlinear subnetwork with a step length in a conjugate gradient algorithm as a training parameter and a single-layer linear subnetwork at a receiver with receiver coefficients as a training parameter, and is obtained by training a downlink air interface channel vector and a transmission vector training set. The internal structure of the conjugate gradient MMSE precoding algorithm is combined with an advanced DNN network, a model-driven supervised neural network is constructed by using deep unfolding, and compared with a data-driven 'black box' DNN network, the network has better generalization ability and interpretability.

To provide a receiver with nonlinear neural network equalizers for a high speed data channel and provide a method of use.SOLUTION: The present invention provides a method performed by a receiver for use in a data channel on an integrated circuit device. The method comprises steps of: performing nonlinear equalization of digitized samples of an input signal on a data channel; determining respective values of each of the output signals from the nonlinear equalized output signals; and adapting parameters of the nonlinear equalization based on the respective values.SELECTED DRAWING: Figure 11

The present disclosure generally relates to handling data of non-linear, multi-variable complex systems. More particularly, the present disclosure relates to methods and systems for training machine learning-based computing devices to ensure adaptive sampling of highly complex data packets. The present invention provides a robust and effective solution to implement a complexity-based sampling methodology that trains the neural network in complex mapping regions, by iteratively sampling the DBMS function and training the neural network in complex regions. The system (110) for training the complex and non-linear neural network may be equipped with a Machine Learning (ML) Engine (214) to solve the problem efficiently.

A receiver for use in a data channel on an integrated circuit device includes a nonlinear equalizer having digitized samples of signals on the data channel as inputs, a decision circuit configured to determine a respective value of each of the signals from an output of the nonlinear equalizer, and an adaptation circuit configured to adapt parameters of the nonlinear equalizer based on each of the values. The nonlinear equalizer may be a neural network equalizer, such as a multi-layer perceptron neural network equalizer or a low-complexity multi-layer perceptron neural network equalizer. A method for detecting data on a data channel on an integrated circuit device includes performing a nonlinear equalization of digitized samples of input signals on the data channel, determining a respective value of each of the output signals from an output signal of the nonlinear equalization, and adapting parameters of the nonlinear equalization based on each of the values.

The present disclosure provides methods and systems for determining information of a sample location. One system includes a computer subsystem configured to input one or more images of a sample location into a zone-based neural network configured to perform digital nonlinear filtering on the one or more images to thereby generate filtered images of the sample location. The computer subsystem is further configured to determine information of the sample location from the filtered images.

The application discloses a reaction condition recommendation method and device for yield prediction ranking, which comprises the following steps: processing an original data set to generate a sample data set; processing samples with several reaction conditions in the sample data set to retain samples with the maximum yield and generate a reaction condition data set; retaining all samples to generate a reaction yield data set; constructing a reaction condition prediction model based on a graph convolution network and a nonlinear neural network and training the model; constructing a reaction yield prediction model based on a multilayer nonlinear neural network and training the model; inputting a molecular SMILES of a target reaction into the reaction condition prediction model to obtain a reaction condition combination; inputting the molecular SMILES and the molecular SMILES in the reaction condition combination into the reaction yield prediction model to obtain a yield corresponding to the reaction condition combination and sort the yield, and taking the reaction condition combination with the maximum yield as a recommendation result. The application provides a reaction condition with high yield for a user, and simultaneously gives a yield value to indicate the yield upper limit of the reaction.

The application provides a learning-based tomographic imaging and reconstruction method to measure the scene density distribution in a light reuse manner. During imaging, the light source emits light according to the intensity obtained through pre-learning, the light from different directions reaches the sensor after being absorbed and attenuated by the scene, and the density information of the scene is obtained by calculating and reconstructing the measurement value. The light emission intensity and the reconstruction algorithm are obtained by learning through a neural network. The method models the CT imaging process as a linear fully connected layer, and the weight corresponds to the light emission intensity of the light source during imaging; the reconstruction algorithm is modeled as a nonlinear neural network, which can be optimized according to the characteristics of the scanning geometry. The method requires a small amount of collected data, and the calculation and reconstruction do not require strong prior assumptions, realizes efficient and high-quality CT acquisition and reconstruction, and can be applied to high-speed dynamic scene three-dimensional reconstruction.

The present application relates to a kind of all-optical nonlinear neural network systems based on linear system, belong to optical neural network technical field.Solve the bottleneck problem that optical neural network relies on nonlinear material or photoelectric conversion to realize nonlinear operation.Techinical scheme includes input module, and input signal is loaded in annular resonant cavity resonance wavelength detuning amount;Coupling module controls interlayer weight;Neuron module passes through multilayer annular resonant cavity linear transmission signal;Output module loads output signal in probe light power.Input and output signal are loaded in different physical quantities, so that linear system realizes nonlinear function.Affordable effect includes breaking through optical nonlinear operation bottleneck, significantly reduce system complexity and energy consumption, enhance compatibility and practicality, improve parallel computing efficiency and system scalability.