Edge-based adaptive machine learning for object recognition
Examples of techniques for interactive generation of labeled data and training instances are provided. According to one or more embodiments of the present invention, a computer-implemented method for interactive generation of labeled data and training instances includes presenting, by the processing device, control labeling options to a user. The method further includes selecting, by a user, one or more of the presented control labeling options. The method further includes selecting, by a processing device, a representative set of unlabeled data samples based at least in part on the control labeling options selected by the user. The method further includes generating, by a processing device, a set of suggested labels for each of the unlabeled data samples.
1. A system for interactive generation of labeled data and training instances, the system comprising:
a memory comprising computer readable instructions; and
a processing device for executing the computer readable instructions for performing a method, the method comprising:
initializing an adaptation task containing a contextual specification of a target visual domain;
receiving, by the processing device, a selection of a set of unlabeled images, wherein the set of unlabeled images are stored on a user's mobile device and describe the target visual domain;
generating, by the processing device, a set of suggested labels for each of the visual objects in the unlabeled images stored on the user's mobile device;
receiving, by the processing device, selected labels for labeling each of the visual objects in the unlabeled images, the selected labels being selected by a user operating the user's mobile device;
labeling, by the processing device, the visual objects in the unlabeled image using the selected labels; and
training, by the processing device, an adaptive model to obtain a recognition result via late fusion techniques using the labeled objects selected by a user as training instances, wherein a late fusion technique parameter is set based on a number of incorrect labels in the set of suggested labels,
wherein the trained adaptive model and training instances are subsequently shared among a plurality users participating in a subsequent adaptation task, and the trained adaptive model is updated in response to performing the training during the subsequent adaptation task, the training during the subsequent adaptation task including using labeled objects selected and received by the plurality of users participating in the subsequent adaptation task.
2. The system of claim 1 , wherein the set of suggested labels for each of the visual objects in the unlabeled images are generated from a generic machine learning model.
3. The system of claim 1 , wherein the method further comprises presenting the generated set of suggested labels to the user.