Unsupervised, semi-supervised, and supervised learning using deep learning based probabilistic generative models
View Patent ↗Embodiments of the present systems and methods may provide techniques to discover features such as object categories that provide improved accuracy and performance. For example, in an embodiment, a method may comprise extracting, at the computer system, features from a dataset comprising a plurality of data samples using a backbone neural network to form a features vector for each data sample, training, at the computer system, using the features vectors for at least some of the plurality of data samples, an unsupervised generative probabilistic model to perform clustering of extracted features of the at least some of the plurality of data samples by minimizing a negative Log-Likelihood function, wherein clusters of extracted features form categories, and categorizing, at the computer system, at least some different data samples of the plurality of data samples, into the formed categories.
1. A method, implemented in a computer comprising a processor, memory accessible by the processor, and computer program instructions stored in the memory and executable by the processor, the method comprising:
extracting, at a computer system, features from a dataset comprising a plurality of data samples using a backbone neural network to form a features vector for each data sample;
training, at the computer system, using the features vectors for at least some of the plurality of data samples, an unsupervised generative probabilistic model to perform clustering of extracted features of the at least some of the plurality of data samples by minimizing a negative Log-Likelihood function, wherein clusters of extracted features form categories; and
categorizing, at the computer system, at least some different data samples of the plurality of data samples, into the formed categories.
2. The method of claim 1 , wherein the unsupervised generative probabilistic model comprises a mixture component convolutional neural network to accept the features vectors and to output distributions of possible values over unobserved variables of the model, a generative convolutional neural network to accept the features vectors and the distributions over the unobserved variables and to output a negative log likelihood, and a negative log likelihood loss to reduce the negative log likelihood.
3. The method of claim 2 , wherein the unsupervised generative probabilistic model comprises a mixture component convolutional neural network to accept the features vectors and to output a mixture component vector indicating a conditional probability that each feature vector belongs to each mixture component; and
computing, at the computer system, a probability that each topic is associated with each data sample based on the mixture component vector indicating a conditional probability that each feature vector belongs to each mixture component and based on a Gaussian distribution associated with each mixture component.
4. The method of claim 2 , wherein the unsupervised generative probabilistic model comprises a multinomial distribution of topic given document to accept the features vectors and to output a topic vector indicating a conditional probability that each feature vector belongs to each topic.
5. The method of claim 4 , further comprising:
computing, at the computer system, a probability that each topic is associated with each data sample based on the topic vector returned by the mixture component convolutional network and indicating a conditional probability that each feature vector belongs to each topic, multiplied by matrix of model weights, converting topics to words, and defining the log-frequencies of each conversion, and further multiplied by the feature vector.
6. The method of claim 1 , further comprising:
generating, at the computer system, a topic vector indicating a conditional probability that each image represented by a feature vector belongs to each topic, and thus discovering the categories present in the data without any supervision; and
training, at the computer system, a classifier model using a semi-supervised process including at least some labeled data samples, and applying a classification loss function.
7. The method of claim 1 , further comprising using the method of claim 1 between intermediate layers of a neural network to learn a global context to uncontrolled intermediate features.
8. A system comprising a processor, memory accessible by the processor, and computer program instructions stored in the memory and executable by the processor to perform:
extracting, at a computer system, features from a dataset comprising a plurality of data samples using a backbone neural network to form a features vector for each data sample;
training, at the computer system, using the features vectors for at least some of the plurality of data samples, an unsupervised generative probabilistic model to perform clustering of extracted features of the at least some of the plurality of data samples by minimizing a negative Log-Likelihood function, wherein clusters of extracted features form categories; and
categorizing, at the computer system, at least some different data samples of the plurality of data samples, into the formed categories.
9. The system of claim 8 , wherein the unsupervised generative probabilistic model comprises a mixture component convolutional neural network to accept the features vectors and to output distributions of possible values over unobserved variables of the model, a generative convolutional neural network to accept the features vectors and the distributions over the unobserved variables and to output a negative log likelihood, and a negative log likelihood loss to reduce the negative log likelihood.
10. The system of claim 9 , wherein the unsupervised generative probabilistic model comprises a mixture component convolutional neural network to accept the features vectors and to output a mixture component vector indicating a conditional probability that each feature vector belongs to each mixture component; and
computing, at the computer system, a probability that each topic is associated with each data sample based on the mixture component vector indicating a conditional probability that each feature vector belongs to each mixture component and based on a Gaussian distribution associated with each mixture component.
11. The system of claim 9 , wherein the unsupervised generative probabilistic model comprises a multinomial distribution of topic given document to accept the features vectors and to output a topic vector indicating a conditional probability that each feature vector belongs to each topic.
12. The system of claim 11 , further comprising:
computing, at the computer system, a probability that each topic is associated with each data sample based on the topic vector returned by the mixture component convolutional network and indicating a conditional probability that each feature vector belongs to each topic, multiplied by matrix of model weights, converting topics to words, and defining the log-frequencies of each conversion, and further multiplied by the feature vector.
13. The system of claim 8 , further comprising:
generating, at the computer system, a topic vector indicating a conditional probability that each image represented by a feature vector belongs to each topic, and thus discovering the categories present in the data without any supervision; and
training, at the computer system, a classifier model using a semi-supervised process including at least some labeled data samples, and applying a classification loss function.
14. The system of claim 8 , further comprising using the system of claim 8 between intermediate layers of a neural network to learn a global context to uncontrolled intermediate features.
15. A computer program product comprising a non-transitory computer readable storage having program instructions embodied therewith, the program instructions executable by a computer, to cause the computer to perform a method comprising:
extracting, at a computer system, features from a dataset comprising a plurality of data samples using a backbone neural network to form a features vector for each data sample;
training, at the computer system, using the features vectors for at least some of the plurality of data samples, an unsupervised generative probabilistic model to perform clustering of extracted features of the at least some of the plurality of data samples by minimizing a negative Log-Likelihood function, wherein clusters of extracted features form categories; and
categorizing, at the computer system, at least some different data samples of the plurality of data samples, into the formed categories.
16. The computer program product of claim 15 , wherein the unsupervised generative probabilistic model comprises a mixture component convolutional neural network to accept the features vectors and to output distributions of possible values over unobserved variables of the model, a generative convolutional neural network to accept the features vectors and the distributions over the unobserved variables and to output a negative log likelihood, and a negative log likelihood loss to reduce the negative log likelihood.
17. The computer program product of claim 16 , wherein the unsupervised generative probabilistic model comprises a mixture component convolutional neural network to accept the features vectors and to output a mixture component vector indicating a conditional probability that each feature vector belongs to each mixture component; and
computing, at the computer system, a probability that each topic is associated with each data sample based on the mixture component vector indicating a conditional probability that each feature vector belongs to each mixture component and based on a Gaussian distribution associated with each mixture component.
18. The computer program product of claim 16 , wherein the unsupervised generative probabilistic model comprises a multinomial distribution of topic given document to accept the features vectors and to output a topic vector indicating a conditional probability that each feature vector belongs to each topic; and
the method further comprises computing, at the computer system, a probability that each topic is associated with each data sample based on the topic vector returned by the mixture component convolutional network and indicating a conditional probability that each feature vector belongs to each topic, multiplied by matrix of model weights, converting topics to words, and defining the log-frequencies of each conversion, and further multiplied by the feature vector.
19. The computer program product of claim 15 , further comprising:
generating, at the computer system, a topic vector indicating a conditional probability that each image represented by a feature vector belongs to each topic, and thus discovering the categories present in the data without any supervision; and
training, at the computer system, a classifier model using a semi-supervised process including at least some labeled data samples, and applying a classification loss function.
20. The computer program product of claim 15 , further comprising using the computer program product of claim 15 between intermediate layers of a neural network to learn a global context to uncontrolled intermediate features.