Abstract:
Machine learning models have excelled in machine translation, face recognition, and recommendation thanks to representation learning. Most representation learning methods require many consistent and noise-free labels. Labels are often limited due to budget constraints and privacy concerns. Standard representation learning methods on small labeled data sets will easily overfit and produce suboptimal solutions. In education, multiple workers with diverse expertise annotate the limited labels, which creates noise and inconsistency in crowdsourcing. This paper proposes a novel framework to learn effective representations from limited data with crowdsourced labels. We use a grouping-based deep neural network to learn embeddings from a few training samples and a Bayesian confidence estimator to capture crowdsourced label inconsistency. To speed up training, we develop a hard example selection procedure to adaptively select model misclassified training examples. Our framework outperforms state-of-the-art baselines in learning representations from limited data with crowdsourced labels in extensive experiments on three real-world data sets. To fully understand our proposed framework, we analyze each of its main components and present its promising results in real production.
Note: Please discuss with our team before submitting this abstract to the college. This Abstract or Synopsis varies based on student project requirements.
Did you like this final year project?
To download this project Code with thesis report and project training... Click Here
