Learning graph-based priors for generalized zero-shot learning
An image is received for classification. Thereafter, features are extracted from the image which are used by a machine learning model to classify the image. Thereafter, data is provided that characterizes the classification. The machine learning model can be trained using a training data set labeled, in part, using a generative model conditioned on label attribute information in combination with a directed relation graph having a plurality of nodes in which each node without images at training time are given predefined probability distributions. Related apparatus, systems, techniques and articles are also described.
1. A method for implementation by one or more computing devices comprising:
receiving an image for classification;
extracting features from the image;
classifying, using the extracted features by a machine learning model, the image; and
providing data characterizing the classification;
wherein the machine learning model is trained using a training data set labeled, in part, using a generative model conditioned on label attribute information in combination with a directed relation graph having a plurality of nodes in which each node without images at training time has predefined probability distributions.
2. The method of claim 1 , wherein the machine learning model is a logistic regression model, a support vector machine, a neural network, or a decision tree.
3. The method of claim 1 , wherein the generative model is a variational autoencoder (VAE).
4. The method of claim 3 , wherein the VAE is a cross-alignment and distribution-alignment VAE (CADA-VAE).
5. The method of claim 4 , wherein the predefined probability distributions are Gaussian distributions.
6. The method of claim 5 , wherein an attribute encoder encodes attributes from unseen classes during training.
7. The method of claim 1 , wherein providing the data characterizing the classification comprises: causing the classification to be displayed in a graphical user interface of an electronic visual display, transmitting the classification over a network to a remote computing system, storing the classification in physical persistence, or loading the classification into memory.
8. A system comprising:
at least one data processor; and
memory storing instructions which, when executed by the at least one data processor, result in operations comprising:
receiving an image for classification;
extracting features from the image;
classifying, using the extracted features by a machine learning model, the image; and
providing data characterizing the classification;
wherein the machine learning model is trained using a training data set labeled, in part, using a generative model conditioned on label attribute information in combination with a directed relation graph having a plurality of nodes in which each node without images at training time has predefined probability distributions.
9. The system of claim 8 , wherein the machine learning model is a logistic regression model, a support vector machine, a neural network, or a decision tree.
10. The system of claim 8 , wherein the generative model is a variational autoencoder (VAE).
11. The system of claim 10 , wherein the VAE is a cross-alignment and distribution-alignment VAE (CADA-VAE).
12. The system of claim 11 , wherein the predefined probability distributions are Gaussian distributions.
13. The system of claim 12 , wherein an attribute encoder encodes attributes from unseen classes during training.
14. The system of claim 8 , wherein providing the data characterizing the classification comprises: causing the classification to be displayed in a graphical user interface of an electronic visual display, transmitting the classification over a network to a remote computing system, storing the classification in physical persistence, or loading the classification into memory.
15. A method for implementation by one or more computing devices comprising:
labeling training data using a generative model conditioned on label attribute information in combination with a directed relation graph having a plurality of nodes in which each node without images at training time has predefined probability distributions;
training a machine learning model using the training data;
receiving an image for classification;
extracting features from the image;
classifying, using the extracted features by the machine learning model, the image; and
providing data characterizing the classification.
16. The method of claim 15 , wherein the machine learning model is a logistic regression model, a support vector machine, a neural network, or a decision tree.
17. The method of claim 16 , wherein the generative model is a variational autoencoder (VAE).
18. The method of claim 17 , wherein the VAE is a cross-alignment and distribution-alignment VAE (CADA-VAE).
19. The method of claim 18 , wherein the predefined probability distributions are Gaussian distributions.
20. The method of claim 19 , wherein an attribute encoder encodes attributes from unseen classes during training.