Abstract: This specification describes techniques for determining an inter-cut time range in a media item. In one example method, a set of candidate segmentation points representing time points in a media item is determined, wherein each candidate segmentation point represents at least one of a time point at which feature media data appears in the media item, or a time point at which feature media data disappears from the media item. Based on the candidate segmentation points, an inter-cut time range is identified. The inter-cut time range is associated with a start point from the candidate segmentation points that satisfies a predetermined start point condition, and the inter-cut time range is associated with an end point from the candidate segmentation points that satisfies a predetermined end point condition. The inter-cut time range represents a time period between the start point and the end point in the media item.

Abstract: An image recognition system includes a plurality of sensor devices and a server device. Bach of the plurality of sensor devices includes a first image recognition unit and a transmission data generation unit. The first image recognition unit performs an image recognition process on time-series images captured in chronological order, and estimates whether a predetermined event has occurred in an imaging target captured in the time-series images. On condition that the first image recognition unit has made the estimation that the predetermined event has occurred in the imaging target, the transmission data generation unit generates, as transmission data to be transmitted to the server device, first time-series images obtained by partial images including the image of the imaging target being cut out from respective images included in the time-series images. The server device includes a second image recognition unit.

Abstract: The present invention provides a technique for enhancing the added value of flow line information. The flow line combining device is provided with: an acquisition unit for acquiring first flow line information indicating a trail of positions determined by using a first method and second flow line information indicating a trail of positions determined by using a second method which is different from the first method; a determination unit for assessing overlap in the trails respectively indicated by the acquired first flow line information and the second flow line information; and a combining unit for generating third flow line information which combines the first flow line information and second flow line information if the trail overlap assessed by the determination unit meets a predetermined condition.

Abstract: An information processing apparatus includes a first setting unit configured to set a plurality of observation necessity degrees of positions in a real space and times, a first display unit configured to display the plurality of observation necessity degrees mapped based on a position in the real space and a time, a second display unit configured to display a plurality of targets detected from a captured image, based on observation necessity degrees each corresponding to a different one of the plurality of targets, and a receiving unit configured to receive an input of information corresponding to at least one target of the plurality of targets. The first setting unit resets at least one observation necessity degree of the plurality of observation necessity degrees based on the input information.

Abstract: A video classification platform receives a first message that includes a link to video data including video frames captured by one or more cameras associated with the vehicle, and including information concerning a vehicle and a driving event. The platform obtains the video data using the link, identifies the vehicle and the driving event based on the information, and determines objects in a video frame. The platform constructs a collision cone of the vehicle for the video frame and determines, based on the collision cone, a set of objects that have a potential to collide with the vehicle. The platform assigns a category to the driving event based on the set of objects, the collision cone of the vehicle, and the information, generates a second message that includes the category and the information, and sends the second message to a client device to display the second message.

Abstract: A road surface detecting apparatus may include a camera configured to obtain an image of a front side of a vehicle including a road surface, and a processor configured to classify a road surface event based on a road surface detection model which is generated by performing training based on the obtained image. Accordingly, a stable riding feeling may be provided for the user by quantitatively detecting a road surface event through training and controlling the vehicle based on the detected road surface event.

Abstract: A risk detection and avoidance device comprises one or more image sensors, configured to provide image sensor input data representing a sensor image of a vicinity of the device; and one or more processors, configured to detect from the image sensor data an overpass and one or more persons on the overpass; determine from the image sensor data a falling-object probability associated with the detected one or more persons on the overpass, according to a first logic; determine from the image sensor data one or more falling-object evasion factors associated with a vicinity of the device; and if the falling-object probability exceeds a predetermined threshold, determine a harm avoidance maneuver based at least on the one or more falling-object evasion factors, according to a second logic, and send an instruction comprising the harm avoidance maneuver.

Abstract: A position acquisition unit acquires relative position information between an own vehicle and an obstacle based on a reception wave received by a distance measurement sensor. A shape recognition unit executes shape recognition of the obstacle based on image information acquired by an image capturing unit. A detection processing unit detects the obstacle based on the relative position information acquired by the position acquisition unit and a shape recognition result obtained by the shape recognition unit. The detection processing unit determines whether the height of the obstacle is equal to or greater than a predetermined height or not. The detection processing unit discards the relative position information corresponding to the obstacle in a case where the shape recognition result obtained by the shape recognition unit indicates that the height of the obstacle is less than the predetermined height.

Abstract: Apparatus detecting driving incapability state of a driver includes: a head detection portion that successively detects a head portion, which is higher than a neck of the driver, based on an image of a driver's seat captured by an imaging device mounted in a vehicle; and an out-of-frame state detection portion that detects that the driver is incapable of driving, when the head portion detected by the head detection portion is out of a predetermined scope in the image during travelling of the vehicle.

Abstract: An image capturing device includes a cover plate, a sensing module, a frame body, a first adhesive layer, and a second adhesive layer is provided. The sensing module includes a sensor and a light collimator. The light collimator is disposed between the cover plate and the sensor. The light collimator includes a plurality of light collimating units. Each of the light collimating units includes at least one fiber and a plurality of light absorbing columns. The light absorbing columns are disposed parallel to the at least one fiber and surround the at least one fiber. A thickness of the light collimator is T. A distance between two light absorbing columns farthest from each other among the light absorbing columns in each of the light collimating units is D. A numerical aperture of the at least one fiber is NA, NA?0.7, and D?T×tan[sin?1(NA)].

Abstract: A device may capture a plurality of preview frames of a document, and for each preview frame of the plurality of preview frames, process the preview frame to identify an object in the preview frame. Processing the preview frame may include converting the preview frame into a grayscale image, generating a blurred image based on the grayscale image, detecting a plurality of edges in the blurred image, defining at least one bounding rectangle based on the plurality of edges, and determining an outline of the object based on the at least one bounding rectangle. The device may determine whether a value of an image parameter, associated with the one or more preview frames, satisfies a threshold, and provide feedback to a user of the device, or automatically capture an image of the document, based on determining whether the value of the image parameter satisfies the threshold.

Abstract: A device may capture a plurality of preview frames of a document, and for each preview frame of the plurality of preview frames, process the preview frame to identify an object in the preview frame. Processing the preview frame may include converting the preview frame into a grayscale image, generating a blurred image based on the grayscale image, detecting a plurality of edges in the blurred image, defining at least one bounding rectangle based on the plurality of edges, and determining an outline of the object based on the at least one bounding rectangle. The device may determine whether a value of an image parameter, associated with the one or more preview frames, satisfies a threshold, and provide feedback to a user of the device, or automatically capture an image of the document, based on determining whether the value of the image parameter satisfies the threshold.

Abstract: An electronic device according to various embodiments of the present invention comprises: a display panel; a biometric sensor module disposed on the back surface of the display panel; a processor electrically connected to the display panel and the biometric sensor module, and configured to acquire biometric information by using the biometric sensor module; a first adhesive member filling the gap formed between the back surface of the display panel and the biometric sensor module; and a second adhesive member applied on the first adhesive member, wherein the biometric sensor module can be attached to the back surface of the display panel by using the second adhesive member. In addition, other embodiments are possible.

Abstract: A machine implemented method for extracting information from a receipt, the method constituted of: receiving an image of the receipt; identifying a first text portion within the receipt image which represents a product name abbreviation; abbreviating, in accordance with an abbreviation rule, one of the identified first text portion and a plurality of product names stored on a database, each associated with a product identifier; determining a plurality of candidate product identifiers responsive to the abbreviation; for each of a plurality of predetermined features, assigning a value to each of the identified candidate product identifiers; responsive to the assigned values, assigning a candidate score to each of the identified candidate product identifiers; responsive to the assigned candidate scores and a predetermined scoring rule, selecting one candidate product identifier, which is represented by the identified first text portion; and outputting the selected candidate product identifier.

Abstract: A method and apparatus for detecting objects of interest in images, the method comprising the steps of supplying (S1) at least one input image to a trained deep neural network, DNN, which comprises a stack of layers; and using at least one deconvolved output of at least one learned filter or combining (S2) deconvolved outputs of learned filters of at least one layer of the trained deep neural network, DNN, to detect the objects of interest in the supplied images.

Abstract: A point group data processing device includes: an image data acquisition unit configured to acquire a captured image; a point group data acquisition unit configured to acquire point group data indicating position information of a point group corresponding to a plurality of points included in the image; an area setting unit configured to set a target area which is an area surrounding a subject on the image and an enlargement area which is an area obtained by enlarging the target area; and a target point group specifying unit configured to specify a target point group corresponding to the subject based on depth information of a point group included in the target area and depth information of a point group included in the enlargement area, which are included in the point group data.

Abstract: A classification unit classifies messages included in a text log depending on types, and gives an ID set for each type to each of the classified messages. A creation unit creates, based on dates of occurrence attached to the messages, a matrix indicating an appearance distribution of the messages in the text log for each predetermined duration for each ID. A pattern extraction unit extracts a plurality of patterns, which are combinations of the IDs, from the matrix created by the creation unit. A removal unit removes a part or whole of the patterns from the matrix. A determination unit calculates a degree of importance for each element included in each of the patterns, and determines whether the degree of importance is equal to or higher than a predetermined threshold. A sequence extraction unit extracts a sequence.

Abstract: An image processing apparatus configured to extract an irradiation field from an image obtained through radiation imaging, comprises: an inference unit configured to obtain an irradiation field candidate in the image based on inference processing; a contour extracting unit configured to extract a contour of the irradiation field based on contour extraction processing performed on the irradiation field candidate; and a field extracting unit configured to extract the irradiation field based on the contour.

Abstract: The present invention relates to a method for generating a biometric signature of a subject comprising: obtaining a plurality of sequential video frame images of a moving subject from a video segment; obtaining a portion of each frame comprising a surrounding of the moving subject; carrying out a transformation function to the frequency domain on one or more of said portions of the frames comprising a surrounding of a of the subject; and optionally saving the spectral characteristics of said transformation function in a repository. The present invention also relates to a system for carrying out said method.

Abstract: A method includes receiving a plurality of legacy graphics files associated with a control system for an industrial process, each legacy graphics file comprising a plurality of legacy graphics objects. The method also includes parsing the legacy graphics files to identify primitive graphic objects in each legacy graphics file. The method also includes determining relationships among the identified primitive graphic objects to determine a group of primitive graphic objects that form one component of the industrial process. The method also includes determining one or more patterns among the identified primitive graphic objects. The method also includes saving the one or more patterns in a pattern library.

Abstract: An information processing apparatus includes: a memory; and a processor coupled to the memory and configured to: calculate a positional difference and a degree of similarity in feature value between each feature point in a verification image captured by an imaging device and each of a plurality of feature points in a registered image registered in advance; and match the verification image against the registered image by using the degree of similarity to a feature point having the degree of similarity that is second highest or lower for each feature point in the registered image in a case where a positional difference from a feature point having the degree of similarity that is highest in the verification image is greater than a threshold value.

Abstract: A platform for design of a lighting installation generally includes an automated search engine for retrieving and storing a plurality of lighting objects in a lighting object library and a lighting design environment providing a visual representation of a lighting space containing lighting space objects and lighting objects. The visual representation is based on properties of the lighting space objects and lighting objects obtained from the lighting object library. A plurality of aesthetic filters is configured to permit a designer in a design environment to adjust parameters of the plurality of lighting objects handled in the design environment to provide a desired collective lighting effect using the plurality of lighting objects.

Abstract: Textile matching using color and pattern recognition and methods of use are provided herein. An example method includes analyzing an image (305) of a first article of clothing to obtain color information and pattern information (205-230), comparing the color information and pattern information of the first article of clothing to color information and pattern information for a plurality of other articles of clothing (using Bayesian probability analysis to determine matched pairs, and providing a user with wardrobe suggestions using the matched pairs (705-740).

Abstract: A signal retrieval device includes a modification unit and a signal retrieval unit. The modification unit modifies a value of an attribute of a target represented by an input signal or a stored signals stored in a signal storage unit or a value of an attribute relating to a signal generation source of the input signal to acquire a plurality of modified values of the attribute. The signal retrieval unit retrieves a stored signal of the stored signals similar to the input signal using the input signal or the stored signals in which the attribute is modified according to each of the plurality of modified values of the attribute acquired by the modification unit.

Abstract: Systems and methods are provided for generating sets of candidates comprising images and places within a threshold geographic proximity based on geographic information associated with each of the plurality of images and geographic information associated with each place. For each set of candidates, the systems and methods generate a similarity score based on a similarity between text extracted from each image and a place name, and the geographic information associated with each image and each place. For each place with an associated image as a potential match, the systems and methods generate a name similarity score based on matching the extracted text of the image to the place name, and store an image as place data associated with a place based on determining that the name similarity score for the extracted text associated with the image is higher than a second predetermined threshold.

Abstract: A face clustering system for video face clustering in a video sequence, the system including an inherent supervision summarization device configured to collect group-level supervision and instance level supervision within a same chunklet based on a user input of face images for a person, a discriminative projection learning device configured to embed group constraints of the group-level supervision into a transformed space, and configured to generate an embedding space from the original image feature space, and a clustering device, in the embedding space, configured to execute pair-wise based clustering to cluster the video images into different clusters with the instance level supervision collected by the inherent supervision summarization device.

Abstract: There is provided with an image identification apparatus. An extraction unit extracts a feature value of an image from image data using a Neural Network (NN). A processing unit identifies the image based on the feature value extracted by the extraction unit. The NN comprises a plurality of calculation layers connected hierarchically. The NN includes a plurality of sub-neural networks for performing processing of calculation layers after a specific calculation layer. Mutually different data from an output of the specific calculation layer are respectively inputted to the plurality of sub-neural networks.

Abstract: The invention discloses an object location method, device and storage medium based on image segmentation, the object location method comprises: collecting and labeling training images to obtain a trained database; designing a fully convolutional neural network (FCNN); training the FCNN to obtain a target neural network, by inputting the trained database into the FCNN; labeling and locating object images, based on the target neural network. The method is using the training samples collected in the application scenario to train the FCNN, so it can obtain an optimized FCNN and achieve higher robustness and segmentation accuracy. Particularly, the object segmentation method in the embodiment can perform high-precision segmentation on a plurality of overlapping envelope regions when processing envelopes in a logistics system, and an envelope on the top layer is accurate singulation, allowing the robot to grab only one envelope each time, greatly improving the accuracy and experience of logistics sorting.

Abstract: A computer system trains an artificial intelligence (AI) implemented on the computer system. The computer system includes one or more processors that are configured to receive historical transaction data for a set of resources, where the historical transaction data includes positive examples relating to purchases of one or more resources of the set of resources. The one or more processors are also configured to derive augmented data from the historical transaction data, where the augmented data includes negative examples relating to fictional non-purchase decisions for the set of resources. The one or more processors are further configured to determine a training set for a machine learning algorithm for the AI, where the training set includes the historical transaction data and the augmented data. The one or more processors are configured to train the AI using the machine learning algorithm based on the training set.

Abstract: An apparatus for processing a neural network comprises an image memory into which an input image is written tile-by-tile, each tile overlapping a previous tile to a limited extent; a weights memory for storing weight information for a plurality of convolutional layers of a neural network, including at least two pooling layers; and a layer processing engine configured to combine information from the image and weights memories to generate an output map and to write the output map to image memory. The apparatus is configured to store a limited number of values from adjacent a boundary of an output map for a given layer. The layer processing engine is configured to combine the output map values from a previously processed image tile with the information from the image memory and the weights when generating an output map for a layer of the neural network following the given layer.

Abstract: A computer-implemented system is provided for automatic training of an artificial intelligence (AI) asset. A preprocessing engine is provided for receiving an AI asset by upload and preprocessing the AI asset for training by: associating a set of definition parameters and training criteria with the AI asset; analyzing the training criteria to set a specification for training steps and process; and determining a quotation for the training having regard to known factors associated with the definition parameters and the training specification. A transaction engine is provided for presenting the quotation, receiving an approval of the quotation and a means of payment. A training engine is provided for training the AI asset according to the specification and releasing the AI asset after training.

Abstract: Disclosed are an image classification and conversion method, apparatus, image processor and training method thereof, and medium. The image classification method includes receiving a first input image and a second input image; performing image encoding on the first input image by utilizing n stages of encoding units connected in cascades to produce a first output image, wherein n is an integer greater than 1, and wherein as for 1?i<n, the output of the i-th stage of encoding unit is an input of an (i+1)-th stage of encoding unit, wherein m is an integer greater than 1; outputting a first output image, the first output image comprising mn output sub-images, and each of the mn output sub-images is corresponding to an image category.

Abstract: A system for object and information collection comprises a handheld tool to grab the object. The handheld tool comprises a pole with a handle, a trigger, a grabber mechanism at an opposite end, and a logic system attached to the pole, wherein once a user actuates the trigger to pick up an object with the grabber mechanism, the logic system transmits a capture signal to initiate a classification event. A classification application executing on a processor is responsive to receiving the capture signal from the logic system to capture information associated with the object and assign an object classification to the object based at least in part the captured information. At least one of a portion of the information and the assigned object classification is transmitted over network to an analytics cloud service to perform data aggregation and analysis.

Abstract: An electronic device is provided. The electronic device includes a memory configured to store a computer executable instructions; and a processor configured to execute the executable instructions to: determine a text corresponding to a received command, provide response information on the command based on a first artificial intelligence model classifying the text as a text corresponding to one of a plurality of pre-stored texts, and provide error information on the command based on the first artificial intelligence model classifying the text as an error, wherein the first artificial intelligence model is configured to classify the text as the error based on the text corresponding to the command being a similar text having one of an entity and an intent different from at least one of the plurality of pre-stored texts.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for processing images or features of images using an image classification system that includes a batch normalization layer. One of the systems includes a convolutional neural network configured to receive an input comprising an image or image features of the image and to generate a network output that includes respective scores for each object category in a set of object categories, the score for each object category representing a likelihood that that the image contains an image of an object belonging to the category, and the convolutional neural network comprising: a plurality of neural network layers, the plurality of neural network layers comprising a first convolutional neural network layer and a second neural network layer; and a batch normalization layer between the first convolutional neural network layer and the second neural network layer.

Abstract: An image disambiguation method includes: performing image feature extraction and semantic recognition on at least two images in an image set including similar targets to obtain N K-dimensional semantic feature probability vectors, where the image set includes N images, N and K are both positive integers, and N is greater than or equal to 2; determining a differential feature combination according to the N K-dimensional semantic feature probability vectors, the differential feature combination indicating a difference between the similar targets in the at least two images in the image set; and generating a natural language for representing or prompting the difference between the similar targets in the at least two images in the image set according to the differential feature combination and image features of the at least two images in the image set.

Abstract: A bitmap processing system for creating digital documents on a digital printing press by a fulfiller includes a fulfiller operated raster image processor that receives customer application Page Description Language (PDL) job files containing information for creating the digital documents and generates bitmaps in accordance with the files and a fulfiller operated editor that modifies the bitmaps so that the resulting documents process more efficiently in production steps downstream of a printing process.

Abstract: Systems and methods for utilizing one or more cameras at a mobile device to assist in the implementing anti-counterfeit measures for digital barcodes are disclosed. For example, the systems and methods use a mobile device's front camera to capture photographs (or images) at a certain frequency (e.g., capture a photograph every 0.20 seconds). The systems and methods then scan the captured photograph and retrieve relevant image data (for example, scan image vertically and utilize image data of every fourth pixel). The systems and methods compare certain properties of the retrieved image data with properties of one or more comparison light sources. When the systems and methods detect a match (for example, a match within certain threshold parameters), they identify a barcode among a set of barcodes and display the identified barcode at a display screen of the mobile device.

Abstract: Provided is a code processing method comprising: a code registering step of matching an RGB code and identification information so as to generate a code database; and a recognizing step of extracting at least one RGB code from the color code and recognizing identification information corresponding to the at least one RGB code by using the code database in which the identification information according to the RGB code has been stored in advance, when a color code is received from a recognition request terminal.

Abstract: A transaction card may comprise a first card component having an electrically conductive surface configured to receive an electrically applied coating. An electrically applied coating may be formed on the electrically conductive surface. The transaction card may be manufactured by forming a first card component having an electrically conductive surface configured to receive an electrically applied coating. The method may also include applying an electrically applied coating to the electrically conductive surface.

Abstract: Flexible, stretchable RFID tags are formed by a pocket that is formed from one or more substrates and layers of adhesive, and an electronic circuit that is located within this pocket. The RFID tags can include a stretchable substrate and an electronic circuit attached to the stretchable substrate by one or a finite number of discrete spaced apart attachment locations. When the pocket is formed by relatively thick adhesive layers adhering together one or more flexible substrates to form an internal cavity, the electronic circuit is located within this cavity and either is not adhered to any of the substrates of the cavity, and is free to move about within the cavity, or the electronic circuit can be attached to a substrate by a thin layer of adhesive.

Abstract: A film assembly or a multilayer stack can include an antenna having a spiral form disposed on a stretchable layer. The film assembly includes a stretchable substrate having opposite major top and bottom surfaces, and a first material disposed on the major top surface of the stretchable substrate in the form of a plurality of intersecting rows and columns making the film assembly less stretchable than the stretchable substrate. The rows and columns define a plurality of closed cells. An antenna for use in a radio frequency identification tag is disposed on the stretchable substrate in each closed cell and is isolated from the first material. The antenna includes an elongated electrically conductive electrode having a spiral form.

Abstract: A system for encoding data in an artificial intelligence (AI) integrated circuit solution may include a processor configured to receive voice data comprising at least a segment of an audio waveform, load the voice data into an input array of a cellular neural network (CeNN) in the AI integrated circuit, load one or more wavelet filters into one or more kernels of the CeNN, and perform one or more operations on the voice data to generate a time-spectral diagram. The time-spectral diagram may include a wavelet transformation of the voice data. Each of the one or more filters respectively represents a frequency. The system may also output a voice recognition result based on the time-spectral. Sample training data may be encoded in a similar manner for training the cellular neural network.

Abstract: A network metrics repository stores performance metrics measured during operation of a communication network, and stores fault values indicating types of network operation faults. A neural network circuit has an input layer having input nodes, a sequence of hidden layers each having a plurality of combining nodes, and an output layer having an output node. A processor generates forecasted performance metrics based on extrapolating from measured performance metrics in the network metrics repository, and provides to the input nodes of the neural network circuit the forecasted performance metrics and the measured performance metrics. The processor adapts weights and/or firing thresholds that are used by the input nodes responsive to output of the output node, and controls operation of the communication network based on output of the output node. The output node provides the output responsive to processing through the input nodes a stream of measured performance metrics and forecasted performance metrics.

Abstract: Proposed are a method and apparatus for accelerating data processing in a neural network. The apparatus for accelerating data processing in a neural network may include: a control unit configured to quantize data by at least one method according to a characteristic of data calculated at a node forming at least one layer constituting the neural network, and to separately perform calculation at the node according to the quantized data; and memory configured to store the quantized data.

Abstract: A video action detection method based on a convolutional neural network (CNN) is disclosed in the field of computer vision recognition technologies. A temporal-spatial pyramid pooling layer is added to a network structure, which eliminates limitations on input by a network, speeds up training and detection, and improves performance of video action classification and time location. The disclosed convolutional neural network includes a convolutional layer, a common pooling layer, a temporal-spatial pyramid pooling layer and a full connection layer. The outputs of the convolutional neural network include a category classification output layer and a time localization calculation result output layer. The disclosed method does not require down-sampling to obtain video clips of different durations, but instead utilizes direct input of the whole video at once, improving efficiency.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a neural network. One of the methods includes obtaining a batch of labeled training items and a batch of unlabeled training items; processing the labeled training items and the unlabeled training items using the neural network and in accordance with current values of the network parameters to generate respective embeddings; determining a plurality of similarity values, each similarity value measuring a similarity between the embedding for a respective labeled training item and the embedding for a respective unlabeled training item; determining a respective roundtrip path probability for each of a plurality of roundtrip paths; and performing an iteration of a neural network training procedure to determine a first value update to the current values of the network parameters that decreases roundtrip path probabilities for incorrect roundtrip paths.

Abstract: An electronic apparatus includes a storage configured to store a liquid-state machine (LSM) model and a recurrent neural networks (RNN) model, and a processor configured to input and process a feature data acquired from an input data using the LSM model, to input and process an output value output by the LSM model using the RNN model, and to identify whether a preset object is included in the input data based on an output value output by the RNN model. The RNN model is trained by a sample data related to the preset object. The LSM model includes a plurality of interlinked neurons. A weight applied to a link between the plurality of interlinked neurons is identified based on a spike at which a neuron value is greater than or equal to a preset threshold in a preset unit time.

Abstract: A convolution acceleration and computing processing method includes: reading data to be processed of a preset size in a task to be processed from an off-chip memory of an accelerator sequentially through a FIFO port, and storing the data to be processed into at least one input buffer of a first buffer in an on-chip memory of the accelerator; responsive to data to be processed being stored into an input buffer, reading input window data to be processed from the at least one input buffer sequentially, performing convolution on the input window data to be processed by using the convolution kernel data to obtain output window data and storing the output window data in the third buffer; and storing the output window data in the third buffer into the off-chip memory sequentially through the FIFO port.

Abstract: An abnormality detection device according to an embodiment includes a detector, a remover, and a learner. The detector detects first abnormal data in detection target data which is an abnormality detection target by inputting the detection target data to a first autoencoder which performed learning based on first learning target data which is a learning target. The remover removes data associated with the first abnormal data from the first learning target data to generate second learning target data by inputting the first learning target data to a second autoencoder which performed learning based on the first abnormal data detected by the detector. The learner causes the first autoencoder to perform learning based on the second learning target data generated by the remover.