Abstract
The increasing size of modern datasets combined with the difficulty of obtaining real label information (e.g., class) has made semi-supervised learning a problem of considerable practical importance in modern data analysis. Semi-supervised learning is supervised learning with additional information on the distribution of the examples or, simultaneously, an extension of unsupervised learning guided by some constraints. In this article we present a methodology that bridges between artificial neural network output vectors and logical constraints. In order to do this, we present a semantic loss function and a generalized entropy loss function (Rényi entropy) that capture how close the neural network is to satisfying the constraints on its output. Our methods are intended to be generally applicable and compatible with any feedforward neural network. Therefore, the semantic loss and generalized entropy loss are simply a regularization term that can be directly plugged into an existing loss function. We evaluate our methodology over an artificially simulated dataset and two commonly used benchmark datasets which are MNIST and Fashion-MNIST to assess the relation between the analyzed loss functions and the influence of the various input and tuning parameters on the classification accuracy. The experimental evaluation shows that both losses effectively guide the learner to achieve (near-) state-of-the-art results on semi-supervised multiclass classification.
Highlights
On the one hand, supervised learning uses labeled data to train a model that gives accurate forecasts of data that the model has never seen before, e.g., classification and regression [1,2].On the other hand, unsupervised learning takes unlabeled data as an input and prepares a model based on the patterns or based on the dataset structure, e.g., dimensionality reduction, detecting outliers, and clustering [3,4]
Our motivation in assessing the performance of the generalized entropy and semantic losses is not to achieve the state-of-the-art performance in relation to a specific problem, but rather to highlight their effect
Semi-supervised learning is often considered to be a key challenge for future deep learning tasks
Summary
On the one hand, supervised learning uses labeled (marked) data to train a model that gives accurate forecasts of data that the model has never seen before, e.g., classification and regression [1,2]. Including the parametric weighting of the probabilities that endows data analysis with additional flexibility In this context, the second main goal of this article is to examine, in the same spirit as the first question, whether the addition of the generalized entropy loss function to the loss function provides significant improvements over if this generalized regularization term is not added (i.e., unlabeled data is not utilized). The second main goal of this article is to examine, in the same spirit as the first question, whether the addition of the generalized entropy loss function to the loss function provides significant improvements over if this generalized regularization term is not added (i.e., unlabeled data is not utilized) To these two ends, we evaluate our proposed methods over an artificially created dataset and two commonly used benchmark datasets (i.e., MNIST [19] and Fashion-MNIST [20]) with the expectation that the following research questions can be addressed:.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have