Various techniques are provided for training a neural network to classify images. A convolutional neural network (CNN) is trained using training dataset comprising a plurality of synthetic images. The CNN training process tracks image-related metrics and other informative metrics as the training dataset is processed. The trained inference CNN may then be tested using a validation dataset of real images to generate performance results (e.g., whether a training image was properly or improperly labeled by the trained inference CNN). In one or more embodiments, a training dataset and analysis engine extracts and analyzes the informative metrics and performance results, generates parameters for a modified training dataset to improve CNN performance, and generates corresponding instructions to a synthetic image generator to generate a new training dataset. The process repeats in an iterative fashion to build a final training dataset for use in training an inference CNN.

The present disclosure provides a UAV video aesthetic quality evaluation method based on multi-modal deep learning, which establishes a UAV video aesthetic evaluation data set, analyzes the UAV video through a multi-modal neural network, extracts high-dimensional features, and concatenates the extracted features, thereby achieving aesthetic quality evaluation of the UAV video. There are four steps, step one to: establish a UAV video aesthetic evaluation data set, which is divided into positive samples and negative samples according to the video shooting quality; step two to: use SLAM technology to restore the UAV's flight trajectory and to reconstruct a sparse 3D structure of the scene; step three to: through a multi-modal neural network, extract features of the input UAV video on the image branch, motion branch, and structure branch respectively; and step four to: concatenate the features on multiple branches to obtain the final video aesthetic label and video scene type.

Approaches for private and interpretable machine learning systems include a system for processing a query. The system includes one or more teacher modules for receiving a query and generating a respective output, one or more privacy sanitization modules for privacy sanitizing the respective output of each of the one or more teacher modules, and a student module for receiving a query and the privacy sanitized respective output of each of the one or more teacher modules and generating a result. Each of the one or more teacher modules is trained using a respective private data set. The student module is trained using a public data set. In some embodiments, human understandable interpretations of an output from the student module is provided to a model user.

An example system includes a processor to receive an input and a model trained to classify inputs. The processor is to iteratively generate a perturbed input that optimizes a saliency metric including a classification term, a sparsity term, and a smoothness term, while keeping parameters of the model constant. The processor is to also detect that a predefined number of iterations is exceeded or a convergence of values of the perturbed input. The processor is to further generate a saliency mask based on a perturbation of the perturbed input in response to detecting the predefined number of iterations is exceeded or the convergence.

Techniques facilitating causal reasoning for explanation of model predictions are provided. A system can generate one or more explanations of a machine learning model prediction. The one or more explanations can be based on causal relationships determined between feature data of a set of feature data and based on dataset point samples around a trace associated with the causal relationships.

A neural network, trained on a plurality of random size data samples, can receive a plurality of inference data samples including samples of different sizes. The neural network can generate feature maps of the plurality of inference data samples. Pooling can be utilized to generate feature maps having a fixed size. The fixed size feature maps can be utilized to generate an indication of a class for each of the plurality of inference data samples.

In one example, an apparatus comprises: a first sensor layer, including an array of pixel cells configured to generate pixel data; and one or more semiconductor layers located beneath the first sensor layer with the one or more semiconductor layers being electrically connected to the first sensor layer via interconnects. The one or more semiconductor layers comprises on-chip compute circuits configured to receive the pixel data via the interconnects and process the pixel data, the on-chip compute circuits comprising: a machine learning (ML) model accelerator configured to implement a convolutional neural network (CNN) model to process the pixel data; a first memory to store coefficients of the CNN model and instruction codes; a second memory to store the pixel data of a frame; and a controller configured to execute the codes to control operations of the ML model accelerator, the first memory, and the second memory.

The present disclosure relates to systems, non-transitory computer-readable media, and methods for classifying an input image utilizing a classification model conditioned by a generative model and/or self-supervision. For example, the disclosed systems can utilize a generative model to generate a reconstructed image from an input image to be classified. In turn, the disclosed systems can combine the reconstructed image with the input image itself. Using the combination of the input image and the reconstructed image, the disclosed systems utilize a classification model to determine a classification for the input image. Furthermore, the disclosed systems can employ self-supervised learning to cause the classification model to learn discriminative features for better classifying images of both known classes and open-set categories.

The present disclosure relates to systems, non-transitory computer-readable media, and methods for classifying an input image utilizing a classification model conditioned by a generative model and/or self-supervision. For example, the disclosed systems can utilize a generative model to generate a reconstructed image from an input image to be classified. In turn, the disclosed systems can combine the reconstructed image with the input image itself. Using the combination of the input image and the reconstructed image, the disclosed systems utilize a classification model to determine a classification for the input image. Furthermore, the disclosed systems can employ self-supervised learning to cause the classification model to learn discriminative features for better classifying images of both known classes and open-set categories.

The present disclosure provides radio-frequency (RF) systems that can detect the presence of RF signals received by the system, as well as determine characteristics such as the operating frequency of RF signals, the type of RF source that transmitted each RF signal, and/or the location of each RF source with high precision and sensitivity while using low cost, scalable electronics that are versatile enough for deployment in a variety of environments. Such systems can employ a network of RF sensors that can coordinate in response to communication with a computer to perform any such detection and/or determination using trained models executed onboard the RF sensors and/or the computer. RF signals may have unique characteristics when received at one or more RF sensors that may be detected using trained models described herein, even in high noise or non-line of sight (LOS) environments and with low cost, low resolution RF receiver hardware.