Apple Inc.

Cuneyt Oncel Tuzel

Yin Zhou

Ashish Shrivastava

Russell Y Webb

Shreyas Saxena

Jerremy Holland

Tomas Jon Pfister

Shahab Kaynama

Nitish Srivastava

Joshua M Susskind

Erdrin Azemi

Hanlin Goh

Joseph Yitan Cheng

Kaan Emre Dogrusoz

Pavan Kumar Anasosalu Vasu

Ricardo Da Silveira Cabral

Dimitrios Kottas

Seyed Moosavi-Dezfooli

Luca Ballan

Vishwanath A Sindagi

Wenda Wang

Guanhang Wu

Luciano Spinello

The present application relates to apparatus, systems, and methods to perform subject-aware self-supervised learning of a machine-learning model for classification of data, such as classification of biosignals.

Subject-aware contrastive learning for biosignals

A method includes obtaining a time-series of training samples that include one or more states, a ground truth value, an output value produced in the presence of the one or more states, and an actual error value that is defined as a difference between the ground truth value and the output value. The method also includes training a machine learning model using the time-series of training samples such that the machine learning model is configured to determine a condition-dependent error distribution for a current time step based on simulated states for the current time step.

Conditional error models

A generative network may be learned in an adversarial setting with a goal of modifying synthetic data such that a discriminative network may not be able to reliably tell the difference between refined synthetic data and real data. The generative network and discriminative network may work together to learn how to produce more realistic synthetic data with reduced computational cost. The generative network may iteratively learn a function that synthetic data with a goal of generating refined synthetic data that is more difficult for the discriminative network to differentiate from real data, while the discriminative network may be configured to iteratively learn a function that classifies data as either synthetic or real. Over multiple iterations, the generative network may learn to refine the synthetic data to produce refined synthetic data on which other machine learning models may be trained.

Generating more realistic synthetic data with adversarial nets

A method includes determining a detection output that represents an object in a two-dimensional image using a detection model, wherein the detection output includes a shape definition that describes a shape and size of the object; defining a three-dimensional representation based on the shape definition, wherein the three-dimensional representation includes a three-dimensional model that represents the object that is placed in three-dimensional space according to a position and a rotation; determining a three-dimensional detection loss that describes a difference between the three-dimensional representation and three-dimensional sensor information; and updating the detection model based on the three-dimensional detection loss.

Object detection with position, pose, and shape estimation

A method includes obtaining surface samples that represent three-dimensional locations of surfaces of an environment; generating a voxelized representation of the surfaces of the environment in three-dimensional space using the surface samples; obtaining an image that shows the surfaces of the environment; associating each of the surface samples with image information that corresponds to a portion of the image that is spatially correlated with a respective one of the surface samples; determining voxel features for voxels from the voxelized representation based on the surface samples and the image information using a first trained machine learning model, wherein the voxel features each describe three-dimensional shapes present within a respective one of the voxels; and detecting objects based on the voxel features.

Multimodal three-dimensional object detection

A method includes obtaining training samples that include images that depict objects and annotations of annotated key point locations for the objects. The method also includes training a machine learning model to determine estimated key point locations for the objects and key point uncertainty values for the estimated key point locations by minimizing a loss function that is based in part on a key point localization loss value that represents a difference between the annotated key point locations and the estimated key point locations values and is weighted by the key point uncertainty values.

Key point recognition with uncertainty measurement

A neural network is trained to defend against adversarial attacks, such as by preparing an input image for classification by a neural network where the input image includes a noise-based perturbation. The input image is divided into source patches. Replacement patches are selected for the source patches by searching a patch library for candidate patches available for replacing ones of those source patches, such as based on sizes of those source patches. A denoised image reconstructed from a number of replacement patches is then output to the neural network for classification. The denoised image may be produced based on reconstruction errors determined for individual candidate patches identified from the patch library. Alternatively, the denoised image may be selected from amongst a number of candidate denoised images. A set of training images is used to construct the patch library, such as based on salient data within patches of those training images.

Defense against adversarial attacks on neural networks

A voxel feature learning network receives a raw point cloud and converts the point cloud into a sparse 4D tensor comprising three-dimensional coordinates (e.g. X, Y, and Z) for each voxel of a plurality of voxels and a fourth voxel feature dimension for each non-empty voxel. In some embodiments, convolutional mid layers further transform the 4D tensor into a high-dimensional volumetric representation of the point cloud. In some embodiments, a region proposal network identifies 3D bounding boxes of objects in the point cloud based on the high-dimensional volumetric representation. In some embodiments, the feature learning network and the region proposal network are trained end-to-end using training data comprising known ground truth bounding boxes, without requiring human intervention.

Voxel-based feature learning network

A method for detecting objects in an environment includes obtaining, from one or more sensors, distance measurements from the one or more sensors to portions of the environment, generating a representation of a three-dimensional space using the distance measurements, identifying object features in the representation of the three-dimensional space using a neural network, comparing the object features to pre-defined three-dimensional templates to generate scores that represent correspondence of the object features to the pre-defined three-dimensional templates, and determining a location and a rotational orientation for a three-dimensional object based on the scores.

Growing community of inventors

Cupertino, CA, United States of America

Cuneyt Oncel Tuzel