Methods and systems for performing diagnostic functions for a deep learning model are provided. One system includes one or more components executed by one or more computer subsystems. The one or more components include a deep learning model configured for determining information from an image generated for a specimen by an imaging tool. The one or more components also include a diagnostic component configured for determining one or more causal portions of the image that resulted in the information being determined and for performing one or more functions based on the determined one or more causal portions of the image.

A visual tracking system for tracking and identifying persons within a monitored location, comprising a plurality of cameras and a visual processing unit, each camera produces a sequence of video frames depicting one or more of the persons, the visual processing unit is adapted to maintain a coherent track identity for each person across the plurality of cameras using a combination of motion data and visual featurization data, and further determine demographic data and sentiment data using the visual featurization data, the visual tracking system further having a recommendation module adapted to identify a customer need for each person using the sentiment data of the person in addition to context data, and generate an action recommendation for addressing the customer need, the visual tracking system is operably connected to a customer-oriented device configured to perform a customer-oriented action in accordance with the action recommendation.

An apparatus and a method use a convolutional neural network (CNN) including a plurality of convolution layers in the field of artificial intelligence (AI) systems and applications thereof. A computing apparatus using a CNN including a plurality of convolution layers includes a memory storing one or more instructions; and one or more processors configured to execute the one or more instructions stored in the memory to obtain input data; identify a filter for performing a convolution operation with respect to the input data, on one of the plurality of convolution layers; identify a plurality of sub-filters corresponding to different filtering regions within the filter; provide a plurality of feature maps based on the plurality of sub-filters; and obtain output data, based on the plurality of feature maps.

Systems and methods for distributed training of deep learning models are disclosed. An example local device to train deep learning models includes a reference generator to label input data received at the local device to generate training data, a trainer to train a local deep learning model and to transmit the local deep learning model to a server that is to receive a plurality of local deep learning models from a plurality of local devices, the server to determine a set of weights for a global deep learning model, and an updater to update the local deep learning model based on the set of weights received from the server.

A method of training an image classification neural network comprises: presenting a first plurality of images to an observer as a visual stimulus, while collecting neurophysiological signals from a brain of the observer; processing the neurophysiological signals to identify a neurophysiological event indicative of a detection of a target by the observer in at least one image of the first plurality of images; training the image classification neural network to identify the target in the image, based on the identification of the neurophysiological event; and storing the trained image classification neural network in a computer-readable storage medium.

A method for dynamically quantizing feature maps of a received image. The method includes convolving an image based on a predicted maximum value, a predicted minimum value, trained kernel weights and the image data. The input data is quantized based on the predicted minimum value and predicted maximum value. The output of the convolution is computed into an accumulator and re-quantized. The re-quantized value is output to an external memory. The predicted min value and the predicted max value are computed based on the previous max values and min values with a weighted average or a pre-determined formula. Initial min value and max value are computed based on known quantization methods and utilized for initializing the predicted min value and predicted max value in the quantization process.

Systems, methods, and computer-readable media for achieving privacy for both data and an algorithm that operates on the data. A system can involve receiving an algorithm from an algorithm provider and receiving data from a data provider, dividing the algorithm into a first algorithm subset and a second algorithm subset and dividing the data into a first data subset and a second data subset, sending the first algorithm subset and the first data subset to the algorithm provider and sending the second algorithm subset and the second data subset to the data provider, receiving a first partial result from the algorithm provider based on the first algorithm subset and first data subset and receiving a second partial result from the data provider based on the second algorithm subset and the second data subset, and determining a combined result based on the first partial result and the second partial result.

A 3-D convolutional autoencoder for low-dose CT via transfer learning from a 2-D trained network is described, A machine learning method for low dose computed tomography (LDCT) image correction is provided. The method includes training, by a training circuitry, a neural network (NN) based, at least in part, on two-dimensional (2-D) training data. The 2-D training data includes a plurality of 2-D training image pairs. Each 2-D image pair includes one training input image and one corresponding target output image. The training includes adjusting at least one of a plurality of 2-D weights based, at least in part, on an objective function. The method further includes refining, by the training circuitry, the NN based, at least in part, on three-dimensional (3-D) training data. The 3-D training data includes a plurality of 3-D training image pairs. Each 3-D training image pair includes a plurality of adjacent 2-D training input images and at least one corresponding target output image. The refining includes adjusting at least one of a plurality of 3-D weights based, at least in part, on the plurality of 2-D weights and based, at least in part, on the objective function. The plurality of 2-D weights includes the at least one adjusted 2-D weight.

A target detection method and apparatus. The method comprises: acquiring an input image, and sending same to a candidate region generation network to generate a plurality of regions of interest; formatting the plurality of regions of interest, and then sending same to a target key point network to generate a thermodynamic diagram; using a global feature map of the input image to perform convolution on the thermodynamic diagram, so as to generate a local depth feature map; and fusing the global feature map and the local depth feature map, and detecting a target therefrom by means of a detector. The present invention can be applied to target detection at different scales, improves the detection accuracy and robustness of a target detection technique for an occluded target in complex scenarios, and achieves, by means of making full use of local key point information of the target, target positioning under occlusion.

The present application provides an action recognition method and apparatus and an electronic device. The method includes: if a target object is detected from a video frame, acquiring a plurality of images containing the target object, and optical-flow images of the plurality of images; extracting an object trajectory feature of the target object from the plurality of images, and extracting an optical-flow trajectory feature of the target object from the optical-flow images of the plurality of images; and according to the object trajectory feature and the optical-flow trajectory feature, recognizing a type of an action of the target object. Because it combines the time-feature information and the spatial-feature information of the target object, effectively increases the accuracy of the detection and recognition on the action type, and may take into consideration the detection efficiency at the same time, thereby improving the overall detection performance.