Exponential Modeling with Deep Learning Features
First Claim
1. A computer system, comprising:
- one or more processors; and
one or more non-transitory computer-readable media that collectively store a machine-learned classification model configured to generate a classification output that comprises a plurality of classification scores respectively for a number of discrete classes based on a set of input data, the classification score for each discrete class indicative of a likelihood that the input data corresponds to the discrete class;
wherein the machine-learned classification model comprises an embedding model and an exponential model;
wherein the embedding model is configured to receive the set of input data and produce an embedding based on the set of input data, wherein the embedding comprises a number of parameter values respectively for a number of parameters included in a final layer of the embedding model, and wherein the number of parameter values is less than the number of discrete classes; and
wherein the exponential model is configured to receive the embedding and apply a mapping to generate the classification output, wherein the mapping describes a plurality of relationships between the number of parameters included in the final layer of the embedding model and the number of discrete classes.
1 Assignment
0 Petitions
Accused Products
Abstract
Aspects of the present disclosure enable humanly-specified relationships to contribute to a mapping that enables compression of the output structure of a machine-learned model. An exponential model such as a maximum entropy model can leverage a machine-learned embedding and the mapping to produce a classification output. In such fashion, the feature discovery capabilities of machine-learned models (e.g., deep networks) can be synergistically combined with relationships developed based on human understanding of the structural nature of the problem to be solved, thereby enabling compression of model output structures without significant loss of accuracy. These compressed models provide improved applicability to “on device” or other resource-constrained scenarios.
21 Citations
20 Claims
-
1. A computer system, comprising:
-
one or more processors; and one or more non-transitory computer-readable media that collectively store a machine-learned classification model configured to generate a classification output that comprises a plurality of classification scores respectively for a number of discrete classes based on a set of input data, the classification score for each discrete class indicative of a likelihood that the input data corresponds to the discrete class; wherein the machine-learned classification model comprises an embedding model and an exponential model; wherein the embedding model is configured to receive the set of input data and produce an embedding based on the set of input data, wherein the embedding comprises a number of parameter values respectively for a number of parameters included in a final layer of the embedding model, and wherein the number of parameter values is less than the number of discrete classes; and wherein the exponential model is configured to receive the embedding and apply a mapping to generate the classification output, wherein the mapping describes a plurality of relationships between the number of parameters included in the final layer of the embedding model and the number of discrete classes. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18)
-
-
19. A computer-implemented method, comprising:
-
obtaining, by one or more computing devices, a set of input data; inputting, by the one or more computing devices, the set of input data into a machine-learned classification model configured to generate a classification output that comprises a plurality of classification scores respectively for a number of discrete classes based on a set of input data, the classification score for each discrete class indicative of a likelihood that the input data corresponds to the discrete class, wherein the machine-learned classification model comprises an embedding model and an exponential model, wherein the embedding model is configured to receive the set of input data and produce an embedding based on the set of input data, wherein the embedding comprises a number of parameter values respectively for a number of parameters included in a final layer of the embedding model, and wherein the number of parameter values is less than the number of discrete classes, and wherein the exponential model is configured to receive the embedding and apply a mapping to generate the classification output, wherein the mapping describes a plurality of relationships between the number of parameters included in the final layer of the embedding model and the number of discrete classes; and receiving, by the one or more computing devices, a classification output of the machine-learned classification model. - View Dependent Claims (20)
-
Specification