Abstract: A method for guiding a robot equipped with a camera to facilitate three-dimensional (3D) reconstruction through sampling based planning includes recognizing and localizing an object in a two-dimensional (2D) image. The method also includes computing 3D depth maps for the localized object. A 3D object map is constructed from the depth maps. A sampling based structure is grown around the 3D object map and a cost is assigned to each edge of the sampling based structure. The sampling based structure may be searched to determine a lowest cost sequence of edges that may, in turn be used to guide the robot.

Abstract: A method of motion planning includes observing an object from a first pose of an agent having a controllable camera. The method also includes determining one or more subsequent control inputs to move the agent and the camera to observe the object from at least one subsequent pose. The subsequent control input(s) are determined so as to minimize an expected enclosing measure of the object based on visual data collected from the camera. The method further includes controlling the agent and the camera based on the subsequent control input(s).

Abstract: The balance of training data between classes for a machine learning model is modified. Adjustments are made at the gradient stage where selective backpropagation is utilized to modify a cost function to adjust or selectively apply the gradient based on the class example frequency in the data sets. The factor for modifying the gradient may be determined based on a ratio of the number of examples of the class with a fewest members to the number of examples of a present class. The gradient associated with the present class is modified based on the above determined factor.

Abstract: An intelligent camera network cooperatively acquires images over a wireless network. The network automatically captures images based on a trigger. The trigger may include messages from other image capturing devices. A first image capture devices is triggered to acquire an image based on a message from at least one other image capture device.

Abstract: An apparatus for classifying an input includes a classifier and a feature extractor. The feature extractor is configured to generate a feature vector based on the input. The feature vector is also configured to set a number of elements of the feature vector to zero to produce a sparse feature vector. The sparse feature vector has the same dimensions as the feature vector generated by the feature extractor. However, the sparse feature vector includes fewer non-zero elements than the feature vector generated by the feature extractor. The feature vector is further configured to forward the sparse feature vector to the classifier to classify the input.

Abstract: A method of detecting unknown classes is presented and includes generating a first classifier for multiple first classes. In one configuration, an output of the first classifier has a dimension of at least two. The method also includes designing a second classifier to receive the output of the first classifier to decide whether input data belongs to the multiple first classes or at least one second class.

Abstract: A method of blink and averted gaze avoidance with a camera includes detecting an averted gaze of a subject and/or one or more closed eyes of the subject in response to receiving an input to actuate a camera shutter. The method also includes scheduling actuation of the camera shutter to a future estimated time period to capture an image of the subject when a gaze direction of the subject is centered on the camera and/or both eyes of the subject are open.

Abstract: Context-based priors are utilized in machine learning networks (e.g., neural networks) for detecting objects in images. The likely locations of objects are estimated based on context labels. A machine learning network identifies a context label of an entire image. Based on the, the network selects a set of likely regions for detecting objects of interest in the image.

Abstract: A method of adaptively selecting a configuration for a machine learning process includes determining current system resources and performance specifications of a current system. A new configuration for the machine learning process is determined based at least in part on the current system resources and the performance specifications. The method also includes dynamically selecting between a current configuration and the new configuration based at least in part on the current system resources and the performance specifications.

Abstract: A method of biasing a deep neural network includes determining whether an element has an increased probability of being present in an input to the network. The method also includes adjusting a bias of activation functions of neurons in the network to increase sensitivity to the element. In one configuration, the bias is adjusted without adjusting weights of the network. The method further includes adjusting an output of the network based on the biasing.

Abstract: A method of training a neural network model includes determining a specificity of multiple filters after a predetermined number of training iterations. The method also includes training each of the filters based on the specificity.

Abstract: Feature extraction includes determining a reference model for feature extraction and fine-tuning the reference model for different tasks. The method also includes storing a set of weight differences calculated during the fine-tuning. Each set may correspond to a different task.

Abstract: A method of managing memory usage of a stored training set for classification includes calculating one or both of a first similarity metric and a second similarity metric. The first similarity metric is associated with a new training sample and existing training samples of a same class as the new training sample. The second similarity metric is associated with the new training sample and existing training samples of a different class than the new training sample. The method also includes selectively storing the new training sample in memory based on the first similarity metric, and/or the second similarity metric.

Abstract: Certain aspects of the present disclosure provide methods and apparatus for creating tags (static or dynamic) for input/output classes of a neural network model using supervised learning. The method includes augmenting a neural network model with a plurality of neurons and training the augmented network using spike timing dependent plasticity (STDP) to determine one or more tags.

Abstract: A method of operating a spiking neural network having neurons coupled together with a synapse includes monitoring a timing of a presynaptic spike and monitoring a timing of a postsynaptic spike. The method also includes determining a time difference between the postsynaptic spike and the presynaptic spike. The method further includes calculating a stochastic update of a delay for the synapse based on the time difference between the postsynaptic spike and the presynaptic spike.

Abstract: A method for classifying an object includes applying multiple confidence values to multiple objects. The method also includes determining a metric based on the multiple confidence values. The method further includes determining a classification of a first object from the multiple objects based on a knowledge-graph when the metric is above a threshold.

Abstract: Differential encoding in a neural network includes predicting an activation value for a neuron in the neural network based on at least one previous activation value for the neuron. The encoding further includes encoding a value based on a difference between the predicted activation value and an actual activation value for the neuron in the neural network.

Abstract: A method of blink and averted gaze avoidance with a camera includes detecting an averted gaze of a subject and/or one or more closed eyes of the subject in response to receiving an input to actuate a camera shutter. The method also includes scheduling actuation of the camera shutter to a future estimated time period to capture an image of the subject when a gaze direction of the subject is centered on the camera and/or both eyes of the subject are open.

Abstract: A method of operating a spiking neural network having neurons coupled together with a synapse includes monitoring a timing of a presynaptic spike and monitoring a timing of a postsynaptic spike. The method also includes determining a time difference between the postsynaptic spike and the presynaptic spike. The method further includes calculating a stochastic update of a delay for the synapse based on the time difference between the postsynaptic spike and the presynaptic spike.

Abstract: Certain aspects of the present disclosure provide methods and apparatus for creating tags (static or dynamic) for input/output classes of a neural network model using supervised learning. The method includes augmenting a neural network model with a plurality of neurons and training the augmented network using spike timing dependent plasticity (STDP) to determine one or more tags.