Data Synthesis: A Key to Addressing Privacy Concerns in Medical Research

In this paper, the authors propose a novel approach to combat small datasets in deep learning called "conformal synthesis." The main idea is to generate new data samples that are similar to the original dataset but not necessarily identical, using a generative model. These synthetic samples are then included in the training set to improve the generalization of the model.
The authors identify three challenges associated with small datasets: (1) overfitting, where the model becomes too complex and performs well on the training data but poorly on new data; (2) underfitting, where the model is too simple and fails to capture the underlying patterns in the data; and (3) data privacy concerns, where sensitive information may be exposed if the dataset is not properly anonymized.
To address these challenges, the authors propose a two-stage approach: (1) generating new samples using a generative model, such as Generative Adversarial Networks (GANs), and (2) augmenting the training set with these synthetic samples. They demonstrate the effectiveness of their approach on four real-world datasets and show that it can improve the performance of deep learning models in various applications.
The key insight behind conformal synthesis is that the quality of the generated samples is controlled by a parameter called ε, which measures the distance between the original data and the generated samples. By adjusting this parameter, the authors can control the level of overfitting or underfitting in the model.
To evaluate the performance of their approach, the authors use a feedforward neural network trained on the original data and compare its prediction results with those obtained when the extended training set includes the synthetic samples generated by conformal synthesis. They show that the improved generalization of the model leads to better performance in classification tasks.
In summary, the article proposes a novel approach to combat small datasets in deep learning called conformal synthesis. By generating new data samples similar to the original dataset but not identical, the authors can improve the generalization of deep learning models without compromising data privacy. The proposed approach is demonstrated on four real-world datasets and shows promising results.