55 results about "Neuronal activation" patented technology

The invention discloses a memristor-based neuron circuit, comprising a synaptic circuit, a neuron activation function circuit and a synaptic weight control circuit. In the synaptic circuit, a memristor, under the control of four MOS tubes, changes the memristor value to simulate the change of the synaptic weight in the neuron network. The designed neuron synaptic circuit is capable of being directly connected with a digital logic electrical level so as to achieve convenient and real-time adjustment to the synaptic weight and through the use of the feature that the output voltage of an operational amplifier is restricted by the power supply voltage, the neuron circuit activation function can be realized as a saturated linear function. The neuron synaptic weight change circuit can utilize the existing CMOS micro-controller and at the same time, the micro-controller can be loaded by the neuron network algorithm to change the synaptic weight to realize corresponding functions. According to the invention, a plurality of neuron circuits could be connected into a large-scale neuron network for complicated functions such as mode identification, signal processing, associated memory and non-linear mapping, etc.

The invention provides a BP neural network digital image compression based image watermark embedding and extracting method. The method includes: S1a, performing scrambling treatment on an original watermark image to acquire the scrambled watermark image; S2a, establishing a BP neural network and setting transfer function, training adjustment function, the number of training, neuron activation function thresholds, learning constant and compressibility factors of the BP neural network; S3a, dividing the scrambled watermark image into image embedding blocks, setting input and output expectations as carrier image blocks, and training the carrier image blocks through the BP neural network to acquire output O of a hidden layer; S4a, embedding watermarks in the output O of the hidden layer, and decompressing compressed images embedded with watermark image information to obtain carrier images embedded with the watermarks. The invention further provides a BP neural network digital image compression based image watermark extracting method.

The present invention discloses an IGBT remaining useful life prediction method. By a phase space framing reconstruction technology, on the basis of reconstructing a phase space according to a differential entropy rate, the reconstructed phase space is subjected to Volterra-series input signal vector arrangement; correlation between input data and target output is considered, the optimal selection of each frame of input data is carried out, and the IGBT remaining useful life prediction method adopts a forward-backward algorithm and a least angle regression algorithm which are mature currently to use the optimal input data in input vectors as inputs of a model; a multiresponse sparse regression algorithm and a one-by-one extraction method are added on the basis of an original ELM model to cut off hidden layer nodes which are useless or have little effects, and three mixed neuronal activation functions are used, so that an established network has higher robustness and generalization. According to the present invention, difference of different inputs on a prediction model is sufficiently considered; the prediction model capable of dynamically updating each set of input data by an adaptive algorithm is designed; and prediction accuracy is greatly improved.

The invention discloses a federal learning backdoor attack defense method based on DAGMM, and the method comprises the following steps: (1) a client receives a global model, and trains and uploads a local model and a corresponding neuron activation condition; (2) the server receives the update and calculates the loss of the corresponding client by using the DAGMM; and (3) defense is performed based on multi-round reconstruction errors. According to the invention, the model can be effectively protected from backdoor attack.

The invention relates to the fields of biomedical engineering, computer technology and communication, and specifically discloses an accurate positioning method of a rehabilitation FES signal for stroke patients. The method comprises the steps of wearing a dynamic exoskeleton, defining sensor parameters, performing healthy joint movement, recording exercise data of a healthy joint and the myoelectric signal data of adjacent limbs, and calculating the corresponding motion data of an affected joint and the action time and action point of the FES. The accurate positioning method determines a functional electrical stimulation site and a timing of the affected joint through the healthy joint motion of the patients, instead of determining by the experience of a rehabilitation engineer, by using the principle of time-lapse mirroring, each patient can find electrical stimulation time and action positions suitable for own physical conditions, the FES stimulation on the patients makes the actionof the patient affected joint closer to the action of normal limbs, which can better improve the activation of the brain mirror neurons and promote the rehabilitation of the patients, and the method can help the stroke patients to correct gait during the rehabilitation process to a certain extent.

The present invention is directed to antibodies and antigen binding fragments thereof having binding specificity for PACAP. The antibodies and antigen binding fragments thereof comprise the sequences of the V_H, V_L, and CDR polypeptides described herein, and the polynucleotides encoding them. Antibodies and antigen binding fragments described herein bind to and/or compete for binding to the same linear or conformational epitope(s) on human PACAP as an anti-PACAP antibody. The invention contemplates conjugates of anti-PACAP antibodies and binding fragments thereof conjugated to one or more functional or detectable moieties. Methods of making said anti-PACAP antibodies and antigen binding fragments thereof are also contemplated. Other embodiments of the invention contemplate using anti-PACAP antibodies, and binding fragments thereof, for the diagnosis, assessment, and treatment of diseases and disorders associated with PACAP and conditions where antagonism of PACAP-related activities, such as vasodilation, photophobia, mast cell degranulation, and/or neuronal activation, would be therapeutically beneficial.

The invention discloses a deep neural network test sufficiency method based on a variable intensity combination test. The method includes: utilizing a variable intensity combination test technology toperform relation extraction on neurons in the deep neural network according to the model weight, extracting neuron combinations with different intensities, and assessing the coverage condition of theneuron activation state in the neural network according to the neuron activation state in the neuron combinations; and evaluating the model test sufficiency according to the calculated coverage rate.According to the method, the neuron state space is effectively reduced, corresponding neuron combinations are extracted according to different action relationships, and coverage rate calculation is carried out. If the test case can reach a relatively high coverage rate, the sufficiency of the test case can be better proved, so that the scientificity and credibility of the test criterion can be improved.

The invention provides a fast memory coding method based on a multi-synaptic plasticity pulse neural network. The fast memory coding method comprises the steps of 1, converting external stimulation into an input pulse sequence based on a hierarchical coding strategy; 2, after the pulse neural network receives an input pulse, updating a membrane potential of neurons of an output layer based on an improved SRM model; 3, updating a synaptic weight input to an output layer by using a supervisory group Tempotron, and activating neuron memory input of the output layer; step 4, after the neurons of the output layer are activated, using the unsupervised STDP to update synaptic weights among the activated neurons in the layer, and forming an enhanced cyclic sub-network storage memory; and step 5, while executing the step 4, using unsupervised inhibition synaptic plasticity, updating a synaptic weight between an inhibition layer and an output layer, and inhibiting separation of distribution time of neural populations with different inputs of feedback guarantee memories. The invention further provides a fast memory coding device based on the multi-synaptic plastic spiking neural network. According to the invention, the coding speed and stability of memory are effectively improved.

The invention provides a federated learning security aggregation method and device and electronic equipment. The method comprises the following steps: acquiring model parameter update information of all users participating in federated learning for a current sample category; according to the model parameter updating information, performing model parameter updating on a preset global model to obtain a new global model corresponding to each user; extracting a test sample corresponding to the current sample category from a preset test set, and inputting the test sample into the new global model to obtain a neuron average activation value corresponding to each new global model; determining a user clustering result according to the neuron average activation value corresponding to each new global model; and according to a user clustering result, determining malicious users currently participating in federal learning. The malicious user is determined according to the model neuron activation conditions of different users for a certain sample category, so that the authentication of the federated learning participating user identity is realized, and the reliability of a federated learning aggregation result is improved.

A method, system and apparatus is presented for a wireless neural modulation feedback control system as it relates to an implantable medical device comprised of a radio frequency (RF) receiver circuit, one or more dipole or patch antenna(s), one or more electrode leads connected to at least one dipole or patch antenna(s), and at least one microelectronic neural modulation circuit, and an external or internally implanted RF device to neurally modulate brain tissue in order to treat medical conditions that can be mediated by neuronal activation or inhibition, such as Parkinson's, Alzheimer's, epilepsy, other motor or mood based disorders, and/or pain. The implantable receiver captures energy radiated by the RF transmitter unit and converts this energy to an electrical waveform by the implanted neural modulation circuit to deliver energy that can be utilized by the attached electrode pads in order to activate targeted neurons in the brain.

The invention relates to an artificial neuron construction method based on a quantum circuit, which belongs to the field of quantum machine learning, and comprises the following steps of: firstly, respectively coding the input and the weight of a neuron to a quantum calculation ground state, and then acting a controlled unitary gate containing a neuron weight value on an input quantum state; and finally, neuronal output is obtained through quantum phase estimation. The quantum neuron model for realizing the scheme is mainly composed of three parts of quantum circuits, wherein the first part is an input and weight interaction quantum circuit, and the circuit can well realize the function that neurons receive input values from different connection strengths; the second part is a phase estimation quantum circuit, and the circuit realizes the function of a neuron activation function; and the third part is a weight updating quantum circuit, and the circuit converts updating of the weight value into updating of each quantum bit state representing the weight sub-state. The method has the advantage of quantum information processing.

Neuromorphic circuits are multi-cell networks configured to imitate the behavior of biological neural networks. A neuromorphic circuit is provided which comprises a network of neurons each identified by a neuron address in the network, each neuron being able to receive and process at least one input signal and then later emit on an output of the neuron a signal representing an event which occurs inside the neuron, and a programmable memory composed of elementary memories each associated with a respective neuron. The elementary memory, which is a memory of post-synaptic addresses and weights, comprises an activation input linked by a conductor to the output of the associated neuron to directly receive an event signal emitted by this neuron without passing through an address encoder or decoder. The post-synaptic addresses extracted from an elementary memory activated by a neuron are applied, with associated synaptic weights, as inputs to the neural network.

The deep neural network is improved via prototype factorization. A method may include receiving a set of images, analyzing the images, selecting an internal layer, extracting neuronal activations, factorizing the neuronal activations via a matrix factorization algorithm to select prototypes and generate weights for each of the selected prototypes, replacing the neuronal activations of the internal layer with the selected prototypes and the weights of the selected prototypes, and generating a plurality of neuronal activations of the internal layer. Receiving a second image set, classifying the second image set using the prototype and the weight, displaying the second image set, the selected prototype and the weight, displaying a prediction result and a reference truth value of the second image set, and providing an error image based on the prediction result and the reference truth value; identifying an erroneous prototype of the selected prototype associated with the erroneous image; the error weights of the error prototype are ranked, and a new image class is output based on the error prototype being one of the error weights ranked at the top.

The present disclosure provides a system for improving performance of a neural network model. The system receives the neural network model and a training data associated with the neural network model. In addition, the system examines a first plurality of neuron activations inside the neural network model for the training data. The system examines the first plurality of neurons for creating a statistical profile of the first plurality of neuron activations. Further, the system receives a new set of data samples to improve the neural network model. Furthermore, the system examines a second plurality of neuron activations of each new sample of the new set of data samples. Moreover, the system extracts one or more data samples from the new set of data samples with largest novelty measurements. Also, the system adds the extracted one or more samples to the training data for re-training of the neural network model.

The invention discloses an adversarial defense method based on class activation mapping. The adversarial defense method comprises the following steps: S1, establishing a comparison image set for maximizing a neuron activation value in a prediction model through a gradient rising method; S2, positioning a judgment region based on the class activation mapping graph of the to-be-detected image; S3, calculating the inconsistency between the to-be-detected image judgment area and the same label comparison image judgment area based on a binarization algorithm; s4, judging whether the to-be-detectedimage has disturbance or not, if the inconsistency degree is greater than a threshold value, determining that the to-be-detected image has countermeasure disturbance, and otherwise, determining that the to-be-detected image is a normal image; and S5, removing the countermeasure disturbance in the to-be-detected image. The adversarial defense method is high in universality, can resist different confrontation attacks, is low in data processing cost, and improves the defense efficiency.

Peripherally restricted inhibitors of fatty acid amide hydrolase (FAAH) are provided. The compounds can suppress FAAH activity and increases anandamide levels outside the central nervous system (CNS). Despite their relative inability to access brain and spinal cord, the compounds attenuate behavioral responses indicative of persistent pain in rodent models of inflammation and peripheral nerve injury, and suppresses noxious stimulus-evoked neuronal activation in spinal cord regions implicated in nociceptive processing. CBi receptor blockade prevents these effects. Accordingly, the invention also provides methods, and pharmaceutical compositions for treating conditions in which the inhibition of peripheral FAAH would be of benefit. The compounds of the invention are according to the formula (I): in which R₁ is a polar group. In some embodiments, R₁ is selected from the group consisting of hydroxy and the physiologically hydro lysable esters thereof. R₂ and R₃ are independently selected from the group consisting of hydrogen and substituted or unsubstituted hydrocarbyl; each R₄ is independently selected from the group consisting of halogen and substituted or unsubstituted hydrocarbyl and n is an integer from 0 to 4; each R₅ is independently selected from the group consisting of halo and substituted or unsubstituted hydrocarbyl and m is an integer from 0 to 3; and R₆ is substituted or unsubstituted cyclohexyl; and the pharmaceutically acceptable salts thereof.

Maximum expressivity can be received representing a ratio between maximum and minimum input weights to a neuron of a neural network implementing a weighted real-valued logic gate. Operator arity can be received associated with the neuron. Logical constraints associated with the weighted real-valued logic gate can be determined in terms of weights associated with inputs to the neuron, a threshold-of-truth, and a neuron threshold for activation. The threshold-of-truth can be determined as a parameter used in an activation function of the neuron, based on solving an activation optimization formulated based on the logical constraints, the activation optimization maximizing a product of expressivity representing a distribution width of input weights to the neuron and gradient quality for the neuron given the operator arity and the maximum expressivity. The neural network of logical neurons can be trained using the activation function at the neuron, the activation function using the determined threshold-of-truth.

The invention relates to a method for eliminating a data copy, a neural network processor and an electronic product. When neural network hardware is realized, the condition that weight data are the same but belong to connection between different neurons or neuron clusters often occurs, so that enough storage space needs to be reserved when a chip is designed, and therefore, the technical problem of high resource consumption such as power consumption, storage space and even chip area increase of the brain-like chip can be further solved. In order to solve the technical problem, the invention provides a technical means based on source address translation and an alias mechanism. According to the technical scheme based on the technical means, only the neurons corresponding to the alias need to be triggered to register the corresponding activation output when the neurons are activated and output, and the registered corresponding output is output when the neurons corresponding to the alias are activated and output. According to the method, the technical effects of eliminating redundant weight data copies and reducing resource consumption can be realized, and the method is applicable to both an artificial neural network and a spiking neural network.

The embodiment of the invention provides a method and device for determining target features for business data, and the method comprises the steps: employing a piecewise linear model as an agent modelof a business model for processing the business data, and carrying out training in advance through a historical data processing result of the business model; for business data with target features tobe determined, obtaining a plurality of service features extracted by the business model for current business data; inputting the plurality of service features into a piecewise linear model and acquiring a neuron activation state corresponding to each hidden layer, determining each importance coefficient corresponding to each business feature according to the neuron activation state correspondingto each hidden layer and a weight matrix of the corresponding hidden layer, and determining a plurality of target features from a plurality of business features based on each importance coefficient.According to the embodiment, more accurate and effective explanation can be provided for the business model.

The invention provides a natural occlusion expression recognition method and system with accurate attention, and the method comprises the steps: obtaining a to-be-recognized image, and carrying out the preprocessing of the image; processing the preprocessed to-be-recognized image by using a pre-trained natural occlusion expression recognition network to obtain an expression recognition result; the construction and training process of the natural occlusion expression recognition network comprises the steps of performing face key point detection on an occlusion expression image after preprocessing of a known expression, screening a plurality of interest points in key points, generating a Gaussian graph based on each interest point, and obtaining an occlusion indication graph of the corresponding image; an attention descriptor is calculated according to a neuron activation value of the depth feature of the occlusion expression image, the attention descriptor of the depth feature is forced to approach the occlusion indication graph by constructing attention loss, and attention of the depth feature is forced to adapt to different expressions by constructing expression classification loss. The method is high in recognition accuracy.

The invention discloses a near infrared spectrum imaging system and application thereof. A probe of the near-infrared spectrum imaging system adopts a spring design and can be freely adjusted according to head curvatures of different testees during use, so that comfort and tightness are ensured; a soft rubber sheet is additionally arranged on a mounting seat, so that the hair can be automatically pushed aside and fixed during use, the probe is directly contacted with the skin, and interference of the hair is reduced; meanwhile, a probe cap is made of a composite material, so that comfort, supporting performance and shielding effect on ambient light are ensured; a lossless brain-computer interface technology according to the invention can read neuron activation signals at different positions of the brain of a human body in a non-invasive mode, has advantages of being real-time, low in cost, good in equipment and motion compatibility and the like, and is suitable for almost all tested groups. In practical application, the near infrared spectrum imaging system is very effective for brain signal interpretation and has very high application value.

The invention discloses a longitudinal federated learning backdoor defense method based on neuron activation value clustering. The method comprises the following steps: constructing a longitudinal federated recommendation system comprising a plurality of participants and collaborators; and federated learning: after obtaining a constructed commodity guiding link corresponding to the aggregation embedding representation, the collaboration party classifies the commodity guiding link to effectively screen out a commodity guiding link with potential backdoor attack, and repairs the commodity guiding link with the backdoor attack by using a clustering result. The back door attack commodity sample is guided to learn towards a correct prediction direction, so that the commodity sample of a participant does not need to be obtained, and the repaired commodity recommendation model can defend the back door attack; and the joint embedded representation with the same ID as the aggregation embedded representation of the backdoor attack is filtered or attacked and repaired, so that the parameter optimization of the aggregation embedded representation of the backdoor attack on the commodity recommendation model is prevented or improved, and the defense capability of the commodity recommendation model on the backdoor attack is improved.

The invention discloses a test case priority ranking method based on neuron activation frequency analysis, the input is a neural network to be tested, historical data and a test set, and the output isa test set subjected to test case priority ranking. The main idea of the method is to divide a neuron set in a neural network into a frequently activated neuron set and a non-frequently activated neuron set, and sort test cases by calculating the number of frequently activated neurons activated by new test cases and the number of non-frequently activated neurons activated by the new test cases. The method comprises the following specific steps of (1) determining a neuron frequently-activated neuron subset and a non-frequently-activated neuron subset of each type of data, and (2) performing priority ranking on test cases according to the number of neurons in the frequently-activated subsets and the non-frequently-activated subsets of to-be-ranked data.