Resumen de: GB2597129A

A machine learning system 600 for processing transaction data. The machine learning system 600 has a first processing stage 603 with: an interface 622 to receive a vector representation of a previous state for the first processing stage; a time difference encoding 628 to generate a vector representation of a time difference between the current and previous iterations/transactions. Combinatory logic (632, fig 6b) modifies the vector representation of the previous state based on the time difference encoding. Logic (634, fig 6b) combines the modified vector representation and a representation of the current transaction data to generate a vector representation of the current state. The machine learning system 600 also has a second processing stage 604 with a neural network architecture 660, 690 to receive data from the first processing stage and to map said data to a scalar value 602 representative of a likelihood that the proposed transaction presents an anomaly within a sequence of actions. The scalar value is used to determine whether to approve or decline the proposed transaction. The first stage may comprise a recurrent neural network and the second stage may comprise multiple attention heads.

Resumen de: CN110084281A

The invention discloses an image generation method, a neural network compression method, a related device and equipment in the field of artificial intelligence, and the method comprises the steps: inputting a first matrix into an initial image generator, and obtaining a generated image; inputting the generated image into a preset discriminator to obtain a discrimination result, the preset discriminator being obtained through a real image and classification training corresponding to the real image; updating the initial image generator according to the judgment result to obtain a target image generator; and inputting the second matrix into a target image generator to obtain a sample image. Furthermore, the invention also discloses a neural network compression method, which compresses the preset discriminator based on the sample image obtained by the image generation method.

Resumen de: WO2017124641A1

A device for executing reversal training of an artificial neural network comprises an instruction caching unit (1), a controller unit (2), a direct memory access unit (3), an H tree module (4), a primary calculation module (5), and multiple secondary calculation modules (6). By using the device, reversal training of a multiple-layer artificial neural network can be implemented. For each layer, firstly, weighted summation is performed on input gradient vectors to calculate an output gradient vector of this layer. The output gradient vector is multiplied by a derivative value of a next-layer activation function on which forward calculation is performed, so that a next-layer input gradient vector can be obtained. The input gradient vector is multiplied by an input neuron counterpoint in forward calculation to obtain the gradient of a weight value of this layer, and the weight value of this layer can be updated according to the gradient of the obtained weight value of this layer.

Resumen de: WO2020182989A1

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating a schedule for a computation graph. One of the methods includes obtaining data representing an input computation graph; processing the data representing the input computation graph using a graph neural network to generate one or more instance-specific proposal distributions; and generating a schedule for the input computation graph by performing an optimization algorithm in accordance with the one or more instance-specific proposal distributions.

Resumen de: WO2020190466A1

A spatially sparse convolutional neural network (CNN) framework is introduced to that leverages high sparsity of input data to significantly reduce the computational cost of applications that employ CNNs (e.g., inking applications and others) by avoiding unnecessary floating point mathematical operations. The framework, which is compatible with parallelized operations, includes (1) a data structure for sparse tensors that both (a) reduces storage burden and (b) speeds computations; (2) a set of sparse tensor operations that accelerate convolution computations; and (3) the merging of pooling and convolutional layers. Practical applications involving handwriting recognition and/or stroke analysis demonstrate a notable reduction in storage and computational burdens.

Resumen de: US2022012850A1

A trained deep neural network transforms an image of a sample obtained with a holographic microscope to an image that substantially resembles a microscopy image obtained with a microscope having a different microscopy image modality. Examples of different imaging modalities include bright-field, fluorescence, and dark-field. For bright-field applications, deep learning brings bright-field microscopy contrast to holographic images of a sample, bridging the volumetric imaging capability of holography with the speckle-free and artifact-free image contrast of bright-field microscopy. Holographic microscopy images obtained with a holographic microscope are input into a trained deep neural network to perform cross-modality image transformation from a digitally back-propagated hologram corresponding to a particular depth within a sample volume into an image that substantially resembles a microscopy image of the sample obtained at the same particular depth with a microscope having the different microscopy image modality.

Resumen de: US2022012594A1

A computer-implemented method for training a neural network, which, in particular, is configured to classify physical measuring variables, a fitting of parameters of the neural network occurring as a function of an output signal of the neural network, when the input signal is supplied, and as a function of an associated desired output signal, the fitting of the parameters occurs as a function of an ascertained gradient. The components of the ascertained gradient are scaled as a function of to which layer of the neural network the parameters corresponding to these components belong.

Resumen de: US2022012586A1

An inference engine for a neural network uses a compute-in-memory array storing a kernel coefficients. A clamped input matrix is provided to the compute-in-memory array to produce an output vector representing a function of the clamped input vector and the kernel. A circuit is included receiving an input vector, where elements of the input vector have values in a first range of values. The circuit clamps the values of the elements of the input vector a limit of a second range of values to provide the clamped input vector. The second range of values is more narrow than the first range of values, and set according to the characteristics of the compute-in-memory array. The first range of values can be used in training using digital computation resources, and the second range of values can be used in inference using the compute-in-memory array.

Resumen de: US2022012876A1

The invention provides for a method of training a neural network (322) configured for providing a further processing location (326). The method comprises providing (200) a labeled medical image (100), wherein the labeled medical image comprises multiple labels each indicating a truth processing location (102, 104, 106). The method further comprises inputting (202) the labeled medical image into the neural network to obtain one trial processing location. The one trial processing location comprises a most likely trial processing location (108). The method further comprises determine (204) the closest truth processing location (106) for the most likely trial processing location. The method further comprises calculating (206) an error vector (110) using the closest truth processing location and the most likely trial processing location. The method further comprises training (208) the neural network using the error vector.

Resumen de: US2022012898A1

A computer-implemented neural network system for decomposing input video data. A video data input receives a sequence of video image frames. The sequence is encoded, using a 3D spatio-temporal encoder neural network, into a set of latent variables representing a compressed version of the sequence. A 3D spatio-temporal decoder neural network processes the set of latent variables to generate two or more sets of decomposed video data; these may be stored, communicated, and/or made available to a user interface. Input video including undesired features such as reflections, shadows, and occlusions may thus be decomposed into two or more video sequences, one in which the undesired features are suppressed, and another containing the undesired features.

Resumen de: US2022012817A1

Aspects of the disclosure provide a computerized method and system that utilizes reference expense reports to build and train one or more neural network learning models that intelligently determine the riskiness of to-be-determined expense reports submitted for reimbursement. In examples, a determined riskiness may inform a reimbursement entity manager when determining whether to approve, reject, and/or flag for further review a to-be-determined expense report. In instances, computerized expense report resolution systems and methods may be further automated in order to omit user interactions with to-be-determined expense reports, such that an intelligent computer determines whether to approve, reject, and/or flag a to-be-determined expense report based on the intelligently determined riskiness of the to-be-determined expense report.

Resumen de: US2022012479A1

Methods, systems, and apparatuses, including computer programs encoded on a computer-readable storage medium for estimating the shape, size, and mass of fish are described. A pair of stereo cameras may be utilized to obtain right and left images of fish in a defined area. The right and left images may be processed, enhanced, and combined. Object detection may be used to detect and track a fish in images. A pose estimator may be used to determine key points and features of the detected fish. Based on the key points, a three-dimensional (3-D) model of the fish is generated that provides an estimate of the size and shape of the fish. A regression model or neural network model can be applied to the 3-D model to determine a likely weight of the fish.

Resumen de: US2022012507A1

A system and method for autonomous navigation. The system includes a computing device having a processor and a storage device storing computer executable code. The computer executable code, when executed at the processor, is configured to: provide a planned path having intersections in an environment, where the intersections and roads therebetween are represented by sequential place identifications (IDs); receive images of the environment; perform convolutional neural network on the images to obtain predicted place IDs; when a predicted place ID of a current image is next to a place ID of a previous image, and is the same as predicted place IDs of a predetermined number of following images, define the predicted place ID as place IDs of the current and the following images; and perform autonomous navigation based on the planned path and the image place IDs.

Resumen de: US2022012274A1

Methods and systems of training and using a neural network model include training a time series embedding model and a text embedding model with unsupervised clustering to translate time series and text, respectively, to a shared latent space. The time series embedding model and the text embedding model are further trained using semi-supervised clustering that samples training data pairs of time series information and associated text for annotation.

Resumen de: US2022011392A1

The invention provides for a medical imaging system (100, 300). The medical imaging system comprises a memory (110) for storing machine executable instructions (120). The memory further contains an implementation of a trained convolutional neural network (122, 122′, 122″, 122″′, 122″″). The trained convolutional neural network comprises more than one spherical convolutional neural network portions (502, 502′). The trained convolutional neural network is configured for receiving diffusion magnetic resonance imaging data (124). The diffusion magnetic resonance imaging data comprises a spherical diffusion portion (500, 500′). The more than one spherical convolutional neural network portions are configured for receiving the spherical diffusion portion. The trained convolutional neural network comprises an output layer (508) configured for generating a neural network output (126) in response to inputting the diffusion magnetic resonance imaging data into the trained convolutional neural network. The medical imaging system further comprises a processor (104) for controlling the machine executable instructions. Execution of the machine executable instructions causes the processor to: receive (200) the diffusion magnetic resonance imaging data; and generate (202) the neural network output by inputting the diffusion magnetic resonance imaging data into the trained convolutional neural network.

Resumen de: US2022011122A1

Provided are a trajectory prediction method and device, a storage medium, and a computer program to avoid low accuracy and low reliability of a prediction result in conventional trajectory prediction methods. A trajectory prediction neural network acquires input current trajectory data and current map data of current environment when a moving subject moves in the current environment. The current trajectory data and the current map data are expressed as a current trajectory point set and a current map point set in a high-dimensional space. A global scene feature is extracted according to the current trajectory point set and the current map point set. The global scene feature has a trajectory feature and a map feature of the current environment. Multiple prediction trajectory point sets of the moving subject and a probability corresponding to each prediction trajectory point set are predicted and output according to the global scene feature.

Resumen de: US2022012597A1

A generator for converting an input vector from a latent space to one or more records x of measurement data that is realistic with respect to a given application domain. The generator includes: a trained neural network that is configured to map the input vector to a set of distribution parameters that characterize a random distribution of realistic measurement data, where this random distribution is configured such that given said set of distribution parameters and at least one source of randomness, samples of realistic measurement data may be obtained; and a sampling module including a random or pseudo-random number generator as a source of randomness and configured to sample the realistic measurement data from the random distribution.

Resumen de: US2022012567A1

Techniques for training a neural network classifier using classification metadata from another, non-neural network (non-NN) classifier are provided. In one set of embodiments, a computer system can train the non-NN classifier using a training data set, where the training results in a trained version of the non-NN network classifier. The computer system can further classify a data instance in the plurality of data instances using the trained non-NN classifier, the classifying generating a first class distribution for the data instance, and provide the data instance's feature set as input to a neural network classifier, the providing causing the neural network classifier to generate a second class distribution for the data instance. The computer system can then compute a loss value indicating a degree of divergence between the first and second class distributions and provide the loss value as feedback to the neural network classifier, which can cause the neural network classifier to adjust one or more internal edge weights in an manner that reduces the degree of divergence.

Resumen de: US2022012568A1

Apparatuses, systems, and techniques are presented to generate image or video content. In at least one embodiment, one or more neural networks are used to add one or more first objects to an image including one or more second objects, wherein one or more poses of the one or more first objects in the image is determined with respect to the one or more second objects.

Resumen de: US2022010744A1

Using machine learning for cylinder misfire detection in a dynamic firing level modulation controlled internal combustion engine is described. In a classification embodiment, cylinder misfires are differentiated from intentional skips based on a measured exhaust manifold pressure. In a regressive model embodiment, the measured exhaust manifold pressure is compared to a predicted exhaust manifold pressure generated by neural network in response to one or more inputs indicative of the operation of the vehicle. Based on the comparison, a prediction is made if a misfire has occurred or not. In yet other alternative embodiment, angular crank acceleration is used as well for misfire detection.

Resumen de: US2022012511A1

A computing machine accesses conventional image data comprising a photograph of a first person. The computing machine converts, using a conventional image neural network adapter engine, the conventional image data into a model data format representing the first person, wherein the model data format is a format that is standardized for both conventional image data and stereo pair image data. The computing machine generates, using a biometric task neural network engine and based on the model data format data representing the first person, a task output representing the first person. The computing machine transmits a representation of the task output representing the first person.

Resumen de: US2022012745A1

Systems and methods are provided for authorizing an electronic transaction. In one implementation at least one processor is programmed to receive electronic transaction data and historical transaction data, the electronic transaction data including an entity identifier component and an amount component of an electronic transaction; determine, based on the entity identifier component and the amount component, a location of the electronic transaction in a space of a distributed representation space, the distributed representation space comprising a mapping of electronic transaction components in a high-order space; determine locations of the historical transaction data in the distributed representation space; determine a decision boundary in the distributed representation space based on the locations of the historical transaction data; and authorize the electronic transaction based on the location of the electronic transaction being within the decision boundary.

Resumen de: US2022012532A1

A myocardium image analysis method comprising acquiring a target image including precontrast-enhanced myocardium, based on a type of coronary artery related to the myocardium, distinguishing the myocardium included in the target image, using an artificial neural network, providing information on the distinguished myocardium, wherein the artificial neural network is trained based on a training database generated by matching images for training of precontrast-enhanced coronary artery and myocardium and images for training of postcontrast-enhanced coronary artery and myocardium.

Resumen de: US2022012427A1

Systems and methods that access an online networked resource using a locator are disclosed. A first item of content on the networked resource is identified. A trigger rule comprising keywords and a sentiment classifier is accessed. A neural network, including input, hidden, and output layers, is used to assign a sentiment classification to the first item of content. The trigger rule, the sentiment classification, and identified keywords, are used to determine whether response content is to be posted to the online networked resource. In response to determining, using the trigger rule, the assigned sentiment classification, and keywords identified in the first item of content, that response content is to be posted to the online networked resource, the sentiment classification and identified keywords are used to select and/or generate a second item of content, and the second item of content is enabled to be posted to the online networked resource.

Resumen de: US2022012531A1

The aim of the invention is to configure an image analysis device (BA). This is achieved in that a plurality of training images (TPIC) assigned to an object type (OT) and an object sub-type (OST) are fed into a first neural network module (CNN) in Order to detect image features. Furthermore, training output data sets (FEA) of the first neural network module (CNN) are fed into a second neural network module (MLP) in Order to detect object types using image features. According to the invention, the first and second neural network module (CNN, MLP) are trained together such that training output data sets (OOT) of the second neural network module (MLP) at least approximately reproduce the object types (OT) assigned to the training images (TPIC). Furthermore, for each object type (OT1, OT2):—training images (TPIC) assigned to the object type (OT1, OT2) are fed into the trained first neural network module (CNN),—the first neural network module training output data set (FEA1, FEA2) generated for the respective training image (TPIC) is assigned to the object sub-type (OST) of the respective training image (TPIC), and—by means of the aforementioned sub-type assignments, a sub-type detection module (BMLP1, BMLP2) is configured to detect object sub-types (OST) using image features for the image analysis device (BA).