Zero-shot learning using multi-scale manifold alignment
AssigneeHRL Laboratories, LLC
InventorsShay Deutsch, Kyungnam Kim, Yuri Owechko
Life of this patent
Estimating the term…
Abstract
Described is a system for recognition of unseen and untrained patterns. A graph is generated based on visual features from input data, the input data including labeled instances and unseen instances. Semantic representations of the input data are assigned as graph signals based on the visual features. The semantic representations are aligned with visual representations of the input data using a regularization method applied directly in a spectral graph wavelets (SGW) domain. The semantic representations are then used to generate labels for the unseen instances. The unseen instances may represent unknown conditions for an autonomous vehicle.
Claims23
A system for recognition of unseen and untrained patterns, the system comprising:
- one or more processors and a non-transitory computer-readable medium having executable instructions encoded thereon such that when executed, the one or more processors perform operations of:
- generating a graph based on a set of visual features from input data, the input data comprising labeled instances and unseen instances that are unlabeled;
- assigning semantic representations of the input data as graph signals based on the set of visual features, wherein each semantic representation comprises coordinate dimensions;
- aligning the semantic representations with visual representations of the input data using a regularization method applied directly in a spectral graph wavelets (SGW) domain;
- generating labels for the unseen instances using the semantic representations;
- and
- controlling a device based on the labels for the unseen instances.
A computer implemented method for recognition of unseen and untrained patterns, the method comprising an act of:
- causing one or more processors to execute instructions encoded on a non-transitory computer-readable medium, such that upon execution, the one or more processors perform operations of:
- generating a graph based on a set of visual features from input data, the input data comprising labeled instances and unseen instances that are unlabeled;
- assigning semantic representations of the input data as graph signals based on the set of visual features, wherein each semantic representation comprises coordinate dimensions;
- aligning the semantic representations with visual representations of the input data using a regularization method applied directly in a spectral graph wavelets (SGW) domain;
- generating labels for the unseen instances using the semantic representations;
- and
- controlling a device based on the labels for the unseen instances.
A computer program product for recognition of unseen and untrained patterns, the computer program product comprising:
- computer-readable instructions stored on a non-transitory computer-readable medium that are executable by a computer having one or more processors for causing the processor to perform operations of:
- generating a graph based on a set of visual features from input data, the input data comprising labeled instances and unseen instances that are unlabeled;
- assigning semantic representations of the input data as graph signals based on the set of visual features, wherein each semantic representation comprises coordinate dimensions;
- aligning the semantic representations with visual representations of the input data using a regularization method applied directly in a spectral graph wavelets (SGW) domain;
- generating labels for the unseen instances using the semantic representations;
- and
- controlling a device based on the labels for the unseen instances.
Dependencies are inferred from the claim text.