Deep multi-task representation learning
Technologies for analyzing multi-task multimodal data to detect multi-task multimodal events using a deep multi-task representation learning, are disclosed. A combined model with both generative and discriminative aspects is used to share information during both generative and discriminative processes. The technologies can be used to classify data and also to generate data from classification events. The data can then be used to morph data into a desired classification event.
1. A data analyzer comprising instructions embodied in one or more non-transitory machine accessible storage media, the data analyzer configured to cause a computing system comprising one or more computing devices to:
access a set of instances of data having a plurality of modalities, wherein the plurality of modalities are associated with a plurality of tasks;
algorithmically learn a shared representation of the data using a joint optimization of generative and discriminative processes as a single, non-staged framework including parameters learned at the same time and trained in unison; and
learn an inference model using an iterative bottom-up reconstructive and top-down generative approach.
2. The data analyzer of claim 1 , configured to access at least one multi-task multimodal event label and infer data based on the at least one multi-task multimodal event label.
3. The data analyzer of claim 2 , configured to generate or morph at least one event into at least one other event.
4. The data analyzer of claim 1 , configured to classify the data by applying the discriminative process to the shared representation of the multi-task multimodal data.
5. The data analyzer of claim 1 , configured to algorithmically infer missing data both within a modality and across modalities.
6. A method for classifying data, the method comprising, with a computing system comprising one or more computing devices:
accessing a set of instances of data having a plurality of modalities, wherein the plurality of modalities are associated with a plurality of tasks;
algorithmically classifying the data using a dynamic hybrid model that uses joint optimization of generative and discriminative processes as a single, non-staged framework including parameters learned at the same time and trained in unison; and
learning an inference model using an iterative bottom-up reconstructive and top-down generative approach.
7. The method of claim 6 , further comprising inferring an affect-task from a human body pose or a human body activity.
8. The method of claim 6 , further comprising accessing at least one multi-task multimodal event label and inferring data based on the at least one multi-task multimodal event label.
9. A non-transitory computer readable medium having stored thereon at least one program, the at least one program including instructions which, when executed by a processor, cause the processor to perform a method for algorithmically recognizing a multi-task multimodal event in data, comprising:
accessing a set of instances of data, each instance having a plurality of modalities, wherein the plurality of modalities having a plurality of associated tasks;
classifying different instances in the set of data as indicative of different events by applying generative and discriminative processes as a single, non-staged framework including parameters learned at the same time and trained in unison;
generating a semantic label for at least one recognized event; and
learning an inference model using an iterative bottom-up reconstructive and top-down generative approach.
10. The non-transitory computer readable medium of claim 9 , wherein the method further comprises receiving at least one multi-task multimodal event label and determining at least one multi-task multimodal data set.
11. The non-transitory computer readable medium of claim 10 , wherein the at least one multi-task multimodal data set is used to generate a multi-task multimodal event into another multi-task multimodal event.
12. The non-transitory computer readable medium of claim 9 , wherein the generative and discriminative processes determine parameters for the inference model.