Abstract: An estimator generation apparatus may include a first estimator and a second estimator sharing a common encoder. The first estimator may be trained to determine a target person's state from face image data. The second estimator may be trained to reconstruct physiological data from face image data. The machine learning may allow the common encoder to have its parameters converging toward higher-accuracy local solutions for estimating the target person's state, thus generating the estimator that may estimate the target person's state more accurately.

Abstract: Technology is disclosed herein for learning motion in video. In an implementation, an artificial neural network extracts features from a video. A correspondence proposal (CP) module performs, for at least some of the features, a search for corresponding features in the video based on a semantic similarity of a given feature to others of the features. The CP module then generates a joint semantic vector for each of the features based at least on the semantic similarity of the given feature to one or more of the corresponding features and a spatiotemporal distance of the given feature to the one or more of the corresponding features. The artificial neural network is able to identify motion in the video using the joint semantic vectors generated for the features extracted from the video.

Abstract: One example method for identifying useful segments in surgical videos includes accessing a video of a surgical procedure and user activities of a plurality of users who have watched the video of the surgical procedure. The user activities include operations performed during playback of the video. The method further includes dividing the video into multiple segments and determining a popularity score for each of the multiple segments based on the operations. Useful segments are identified from the segments based on the popularity scores. The method further includes generating metadata for the video of the surgical procedure to include an indication of the identified useful segments and associating the metadata with the video of the surgical procedure.

Abstract: In a variety of embodiments, machine classifiers may model multi-turn dialogue as a one-to-many prediction task. The machine classifier may be trained using adversarial bootstrapping between a generator and a discriminator with multi-turn capabilities. The machine classifiers may be trained in both auto-regressive and traditional teacher-forcing modes, with the generator including a hierarchical recurrent encoder-decoder network and the discriminator including a bi-directional recurrent neural network. The discriminator input may include a mixture of ground truth labels, the teacher-forcing outputs of the generator, and/or noise data. This mixture of input data may allow for richer feedback on the autoregressive outputs of the generator. The outputs can be ranked based on the discriminator feedback and a response selected from the ranked outputs.

Abstract: Methods and systems are provided for processing different-modality digital images of tissue. The method includes, for each image, detecting biological entities in the image and generating an entity graph comprising entity nodes, representing respective biological entities, interconnected by edges representing interactions between entities represented by the entity nodes. The method also includes selecting, from each image, anchor elements comprising elements corresponding to anchor elements of at least one other image, and generating an anchor graph in which anchor nodes, representing respective anchor elements, are interconnected with entity nodes of the entity graph for the image by edges indicating relations between entity nodes and anchor nodes.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for associating a new measurement of an object surrounding a vehicle with a maintained track. One of the methods includes receiving an object track for a particular object, receiving a new measurement characterizing a new object at a new time step, and determining whether the new object is the same as the particular object, comprising: generating a representation of the new object at the new and preceding time steps; generating a representation of the particular object at the new and preceding time steps; processing a first network input comprising the representations using a first neural network to generate an embedding of the first network input; and processing the embedding of the first network input using a second neural network to generate a predicted likelihood that the new object and the particular object are the same.

Abstract: A computer implemented method of executing a plurality of discrete software modules each including a machine learning algorithm as an executable software component configurable to approximate a function relating a domain data set to a range data set; a data store; and a message handler as an executable software component arranged to receive input data and communicate output data for the module, wherein the message handler is adapted to determine domain parameters for the algorithm based on the input data and to generate the output data based on a result generated by the algorithm, the method including providing a communication channel between modules in order that at least part of output data for a first module constitutes at least part of input data for a second module so as to create a network of modules for combining machine learning algorithms to refine the approximation of the function.

Abstract: An autoencoder may be trained to predict 3D pose labels using simulation data extracted from a simulated environment, which may be configured to represent an environment in which the 3D pose estimator is to be deployed. Assets may be used to mimic the deployment environment such as 3D models or textures and parameters used to define deployment scenarios and/or conditions that the 3D pose estimator will operate under in the environment. The autoencoder may be trained to predict a segmentation image from an input image that is invariant to occlusions. Further, the autoencoder may be trained to exclude areas of the input image from the object that correspond to one or more appendages of the object. The 3D pose may be adapted to unlabeled real-world data using a GAN, which predicts whether output of the 3D pose estimator was generated from real-world data or simulated data.

Abstract: The present disclosure provides for two-dimensional mode matching by receiving an optical signal traveling in a first direction; and scattering the optical signal according lto a scattering strength that progressively changes in the first direction. In various embodiments, the scattering strength progressively changes by increasing or decreasing in the first direction. A plurality of scatterers disposed in a path of the optical signal change in widths that progressively increase or decrease along the first direction. In various embodiments, a second optical signal is received in the grating coupler from a second direction; and is scattered into a surface of a photonic chip via a grating coupler. In some embodiments, the second direction is perpendicular to the first direction.

Abstract: In one example, a non-transitory computer readable medium stores instructions that, when executed by one or more hardware processors, cause the one or more hardware processors to: load a first weight data element of an array of weight data elements from a memory into a systolic array; select a subset of input data elements from the memory into the systolic array to perform first computations of a dilated convolution operation, the subset being selected based on a rate of the dilated convolution operation and coordinates of the weight data element within the array of weight data elements; and control the systolic array to perform the first computations based on the first weight data element and the subset to generate first output data elements of an output data array. An example of a compiler that generates the instructions is also provided.

Abstract: At least one processor may capture a plurality of image snapshots containing information about a monitored system at a plurality of sequential times, each snapshot having the same vertical and horizontal dimensions. The processor may label the plurality of image snapshots as indicative of an event that took place in the monitored system, may receive additional data describing the event, may cluster the labeled plurality of image snapshots and the additional data using at least one machine learning clustering algorithm, and may merge the clustered plurality of image snapshots and the clustered additional data into merged data. The processors may create a model by processing the merged data using at least one neural network, the model being configured to detect future events of a same type as the event in the monitored system. The processor may store the model in a memory in communication with the processor.

Abstract: A method for image segmentation includes receiving an input image. The method further includes obtaining a deep learning model having a triad of predictors. Furthermore, the method includes processing the input image by a shape model in the triad of predictors to generate a segmented shape image. Moreover, the method includes presenting the segmented shape image via a display unit.

Abstract: Apparatuses, systems, and techniques to collaboratively train one or more machine learning models. Parameter reviewers may be configured to compare sets of machine learning model parameter information in order to generate one or more machine learning models, such as neural networks.

Abstract: A learning-based model is trained using a plurality of attributes of media. Depth estimation is performed using the learning-based model. The depth estimation supports performing a computer vision task on the media. Attributes used in the depth estimation include scene understanding, depth correctness, and processing of sharp edges and gaps. The media may be processed to perform media restoration or the media quality enhancement. A computer vision task may include semantic segmentation.

Abstract: In certain embodiments, cable patching may be facilitated. For example, identifiers corresponding to available ports for devices or objects within racks and required ports for patching devices into the racks may be determined. Cable patching schemes for the racks may be retrieved from sensors. The available ports, the required ports, and the cable patching schemes may be input into a prediction model, the prediction model being configured based on the inputs. New identifiers corresponding to new available ports for a new rack and new required ports for patching new devices into the new rack may be determined. A predicted cable patching scheme for patching the new rack may be obtained from the prediction model. In some embodiments, the predicted cable patching scheme may be transmitted to a user device.

Abstract: Systems and methods include, responsive to a request for capacity change of X channels, X is an integer >1, on an optical section (14) and at an Optical Add/Drop Multiplexer (OADM) node (12) in an optical network (10), dividing optical spectrum on the optical section into M slots, M is an integer >1, such that the capacity change of X channels takes a maximum of N steps, N is an integer >1; and performing the capacity change of X channels in up to the N steps in an interleaved manner that changes a subset of the X channels in each of the N steps. For each step, the performing can include a maximum of M/N slots of the M slots with spacing between each of the M/N slots not used for the capacity change in a corresponding step. The spacing can be f, (N+f), (2N+f), . . . , M over the optical spectrum, where f is each step, f=1, 2, . . . , N.

Abstract: A computing system for decimating video data includes: a processor; a persistent storage system coupled to the processor; and memory storing instructions that, when executed by the processor, cause the processor to decimate a batch of frames of video data by: receiving the batch of frames of video data; mapping, by a feature extractor, the frames of the batch to corresponding feature vectors in a feature space, each of the feature vectors having a lower dimension than a corresponding one of the frames of the batch; selecting a set of dissimilar frames from the plurality of frames of video data based on dissimilarities between corresponding ones of the feature vectors; and storing the selected set of dissimilar frames in the persistent storage system, the size of the selected set of dissimilar frames being smaller than the number of frames in the batch of frames of video data.

Abstract: The techniques disclosed herein improve the efficiency of a system by providing intelligent agents for managing data associated with objects that are displayed within mixed-reality and virtual-reality collaboration environments and facilitating communication between the agents and computing devices associated with the collaboration environments. Individual agents are configured to collect, analyze, and store data associated with individual objects in a database associated with the agent. The agents are further configured to receive queries from the computing devices regarding an individual object. Agents can respond to queries by presenting relevant information collected by the agent with a view of the queried object of interest In addition, the data stored by the agents can be shared between different collaboration environments without requiring users to manually store and present a collection of content for each object.

Abstract: Disclosed herein is a swappable modular-based radiofrequency (RF) frontend that is reconfigurable to form transmitting (TX) and receiving (RX) phased array systems for diverse applications. Such swappable RF frontend may be used with unique spatial and spectral optical processing of complex RF signals over an ultra-wide frequency band. The swappable RF front end may be used in conjunction with an optically upconverted imaging receiver and/or in conjunction with optically addressed phased array technologies transmitters.

Abstract: The present disclosure provide for a radio frequency over glass (RFoG) system having an optical node and an RFoG extender residing in a first service area coupled to the optical node. The RFoG functions to transmit an upstream (US) radio frequency (RF) signal to a head end, receive a downstream (DS) RF signal from the head end and extend the DS RF signal to the second service area. The second service area is different from the first service area and the second service area is remote from the first service area.