Abstract: An apparatus, or corresponding method, for building or updating a map database is described. In one example, the apparatus includes an image correlation module, a training device, and a learned model or neural network. The image correlation module is configured to correlate a first aerial image and terrestrial sensor data collected at a terrestrial vehicle based on at least one control point from the terrestrial data. The learned model training device is configured to define a learned model based using at least one control point from the terrestrial sensor data as ground truth for analysis of the first aerial image. The learned model inference module is configured to receive a second aerial image and apply the learned model on the second aerial image for identification of mapping information for the map data.

Abstract: A method, apparatus, and computer program product are provided for providing improved neural network implementations using a repeated convolution-based attention module. Example embodiments implement a repeated convolution-based attention module that utilizes multiple iterations of a repeated convolutional application layer and subsequent augmentations to generate an attention module output. Example methods may include augmenting an attention input data object based on a previous iteration convolutional output to produce a current iteration input parameter, inputting the input parameter to a repeated convolutional application layer to generate a current iteration input parameter, repeating for multiple iterations, and augmenting the attention input data object based on the final convolutional output to produce an attention module output.

Abstract: Methods and systems for training neural networks using low-bitwidth accelerators are described. The methods described herein use moment-penalization functions. For example, a method comprises producing a modified data set by training a neural network using a moment-penalization function and the data set. The moment-penalization function is configured to penalize a moment associated with the neural network. Training the neural network in turn comprises quantizing the data set to obtain a fixed-point data set so that the fixed-point data set represents the data set in a fixed-point representation, and passing the fixed-point data set through an analog accelerator. The inventors have recognized that training a neural network using a modified objective function augments the accuracy and robustness of the neural network notwithstanding the use of low-bitwidth accelerators.

Abstract: Described herein are techniques of training a machine learning model and performing inference using an analog processor. Some embodiments mitigate the loss in performance of a machine learning model resulting from a lower precision of an analog processor by using an adaptive block floating-point representation of numbers for the analog processor. Some embodiments mitigate the loss in performance of a machine learning model due to noise that is present when using an analog processor. The techniques involve training the machine learning model such that it is robust to noise.

Abstract: An approach is provided for selecting training observations for machine learning models. The approach involves determining a first distribution of a plurality of features observed in the training data set, and a second distribution of the plurality of features observed in the candidate pool of observations. The approach further involves selecting one or more observations in the candidate pool of observations for annotation based on the first distribution and the second distribution. The approach further involves adding the one or more observations to the training data set after annotation. The training data set is used for training the machine learning model.

Abstract: A training system for training a machine learning model such as a neural network may have a different configuration and/or hardware components than a target device that employs the trained neural network. For example, the training system may use a higher precision format to represent neural network parameters than the target device. In another example, the target device may use analog and digital processing hardware to compute an output of the neural network whereas the training system may have used only digital processing hardware to train the neural network. The difference in configuration and/or hardware components of the target device may introduce quantization error into parameters of the neural network, and thus affect performance of the neural network on the target device. Described herein is a training system that trains a neural network for use on a target device that reduces loss in performance resulting from quantization error.

Abstract: Changes in a road network may be detected and information/data regarding the change may be locally propagated in at least near real time with respect to the detection of the change. A vehicle apparatus onboard a vehicle analyzes sensor data collected as the vehicle traverses at least a portion of a road network. The sensor data is captured by sensors onboard the vehicle. The sensor data is analyzed in at least near real time with respect to the capturing of the sensor data. The vehicle apparatus detects an inconsistency between a result of the analysis of the sensor data and map data stored in the memory. Responsive to the detected inconsistency satisfying a vehicle-to-vehicle criterion, the vehicle apparatus generates a notification message comprising an indication of the detected inconsistency. The vehicles apparatus transmits the notification message via a communication interface using a short or medium range communication protocol.

Abstract: A method, apparatus, and computer program product are provided for providing improved neural network implementations using a repeated convolution-based attention module. Example embodiments implement a repeated convolution-based attention module that utilizes multiple iterations of a repeated convolutional application layer and subsequent augmentations to generate an attention module output. Example methods may include augmenting an attention input data object based on a previous iteration convolutional output to produce a current iteration input parameter, inputting the input parameter to a repeated convolutional application layer to generate a current iteration input parameter, repeating for multiple iterations, and augmenting the attention input data object based on the final convolutional output to produce an attention module output.

Abstract: A distributed processing system is disclosed herein. The distributed processing system includes a server, a database server, and an application server that are interconnected via a network, and connected via the network to a plurality of independent processing units. The independent processing units can include an analysis engine that is machine-learning-capable, and thus uniquely completes its processing tasks. The server can provide one or several pieces of data to one or several of the independent processing units, can receive analysis results from these one or several independent processing units, and can update the result based on a value characterizing the machine learning of the independent processing unit.

Abstract: Synthetic training information/data of a second probe style is generated based on first probe information/data of a first probe style using a style transfer model. First probe information/data is defined. An instance of first probe information/data comprises labels and first probe style sensor information/data. A style transfer model generates training information/data based on at least a portion of the first probe information/data. An instance of training information/data corresponds to an instance of first probe information/data and comprises second probe style sensor information/data. The first and second probe styles are different. A second probe style model is trained using machine learning and the training information/data. The second probe style model is used to analyze second probe style second probe information/data to extract map information/data from the second probe information/data. Each instance of second probe data is captured by one or more second probe sensors of a second probe apparatus.

Abstract: An apparatus, method and computer program product are provided to train a feature detector to identify a respective feature from images captured by a camera. With respect to an apparatus, the apparatus causes at least one feature from one or more images that have been labelled to be projected onto a map. The apparatus is also caused to refine a representation of a path of a vehicle that carries a camera that captured the one or more images based upon registration of the at least one feature that has been projected with the map. Based upon the path of the vehicle following refinement, the apparatus projects one or more other features that have not been labelled from the map into the one or more images and then utilizes the images to train a feature detector.

Abstract: An apparatus, or corresponding method, for building or updating a map database is described. In one example, the apparatus includes an image correlation module, a training device, and a learned model or neural network. The image correlation module is configured to correlate a first aerial image and terrestrial sensor data collected at a terrestrial vehicle based on at least one control point from the terrestrial data. The learned model training device is configured to define a learned model based using at least one control point from the terrestrial sensor data as ground truth for analysis of the first aerial image. The learned model inference module is configured to receive a second aerial image and apply the learned model on the second aerial image for identification of mapping information for the map data.

Abstract: Synthetic training information/data of a second probe style is generated based on first probe information/data of a first probe style using a style transfer model. First probe information/data is defined. An instance of first probe information/data comprises labels and first probe style sensor information/data. A style transfer model generates training information/data based on at least a portion of the first probe information/data. An instance of training information/data corresponds to an instance of first probe information/data and comprises second probe style sensor information/data. The first and second probe styles are different. A second probe style model is trained using machine learning and the training information/data. The second probe style model is used to analyze second probe style second probe information/data to extract map information/data from the second probe information/data. Each instance of second probe data is captured by one or more second probe sensors of a second probe apparatus.

Abstract: Changes in a road network may be detected and information/data regarding the change may be locally propagated in at least near real time with respect to the detection of the change. A vehicle apparatus onboard a vehicle analyzes sensor data collected as the vehicle traverses at least a portion of a road network. The sensor data is captured by sensors onboard the vehicle. The sensor data is analyzed in at least near real time with respect to the capturing of the sensor data. The vehicle apparatus detects an inconsistency between a result of the analysis of the sensor data and map data stored in the memory. Responsive to the detected inconsistency satisfying a vehicle-to-vehicle criterion, the vehicle apparatus generates a notification message comprising an indication of the detected inconsistency. The vehicles apparatus transmits the notification message via a communication interface using a short or medium range communication protocol.

Abstract: An approach is provided for dynamic adaptation of an in-vehicle feature detector. The approach involves embedding a feature detection model, precomputed weights for the feature detection model, or a combination thereof in a data layer of map data representing a geographic area from which a training data set was collected to generate the feature detection model, the precomputed weights, or a combination thereof. The approach also involves deploying the feature detection model, the precomputed weights, or a combination thereof to adapt an in-vehicle feature detector based on determining that the in-vehicle feature detector is in the geographic area, plans to travel in the geographic area, or a combination thereof. The in-vehicle feature detector can then use the feature detection model, the precomputed weights, or a combination thereof to process sensor data collected while in the geographic area to detect one or more features.

Abstract: An apparatus, method and computer program product are provided to train a feature detector to identify a respective feature from images captured by a camera. With respect to an apparatus, the apparatus causes at least one feature from one or more images that have been labelled to be projected onto a map. The apparatus is also caused to refine a representation of a path of a vehicle that carries a camera that captured the one or more images based upon registration of the at least one feature that has been projected with the map. Based upon the path of the vehicle following refinement, the apparatus projects one or more other features that have not been labelled from the map into the one or more images and then utilizes the images to train a feature detector.

Abstract: A distributed processing system is disclosed herein. The distributed processing system includes a server, a database server, and an application server that are interconnected via a network, and connected via the network to a plurality of independent processing units. The independent processing units can include an analysis engine that is machine-learning-capable, and thus uniquely completes its processing tasks. The server can provide one or several pieces of data to one or several of the independent processing units, can receive analysis results from these one or several independent processing units, and can update the result based on a value characterizing the machine learning of the independent processing unit.

Abstract: A distributed processing system is disclosed herein. The distributed processing system includes a server, a database server, and an application server that are interconnected via a network, and connected via the network to a plurality of independent processing units. The independent processing units can include an analysis engine that is machine-learning-capable, and thus uniquely completes its processing tasks. The server can provide one or several pieces of data to one or several of the independent processing units, can receive analysis results from these one or several independent processing units, and can update the result based on a value characterizing the machine learning of the independent processing unit.

Abstract: A distributed processing system is disclosed herein. The distributed processing system includes a server, a database server, and an application server that are interconnected via a network, and connected via the network to a plurality of independent processing units. The independent processing units can include an analysis engine that is machine-learning-capable, and thus uniquely completes its processing tasks. The server can provide one or several pieces of data to one or several of the independent processing units, can receive analysis results from these one or several independent processing units, and can update the result based on a value characterizing the machine learning of the independent processing unit.

Abstract: Synthetic training information/data of a second probe style is generated based on first probe information/data of a first probe style using a style transfer model. First probe information/data is defined. An instance of first probe information/data comprises labels and first probe style sensor information/data. A style transfer model generates training information/data based on at least a portion of the first probe information/data. An instance of training information/data corresponds to an instance of first probe information/data and comprises second probe style sensor information/data. The first and second probe styles are different. A second probe style model is trained using machine learning and the training information/data. The second probe style model is used to analyze second probe style second probe information/data to extract map information/data from the second probe information/data. Each instance of second probe data is captured by one or more second probe sensors of a second probe apparatus.