Controllable Data Generation by Deep Learning: A Review

Abstract

Designing and generating new data under targeted properties has been attracting various critical applications such as molecule design, image editing and speech synthesis. Traditional hand-crafted approaches heavily rely on expertise experience and intensive human efforts, yet still suffer from the insufficiency of scientific knowledge and low throughput to support effective and efficient data generation. Recently, the advancement of deep learning induces expressive methods that can learn the underlying representation and properties of data. Such capability provides new opportunities in figuring out the mutual relationship between the structural patterns and functional properties of the data and leveraging such relationship to generate structural data given the desired properties. This article provides a systematic review of this promising research area, commonly known as controllable deep data generation. Firstly, the potential challenges are raised and preliminaries are provided. Then the controllable deep data generation is formally defined, a taxonomy on various techniques is proposed and the evaluation metrics in this specific domain are summarized. After that, exciting applications of controllable deep data generation are introduced and existing works are experimentally analyzed and compared. Finally, the promising future directions of controllable deep data generation are highlighted and five potential challenges are identified.

Fast and Scalable Unsupervised Deep Subgraph Anomaly Detection

“…By reconstructing data from low-dimensional embeddings, the nature of the majority pattern in the data is captured while outliers and noise are filtered out. Reconstruction error, which is defined as the degree of mismatch between the original data and the reconstructed data from lowdimensional embeddings, can be considered a strong indicator for describing anomalies in the dataset [23,[34][35][36][37]. Specifically, larger reconstruction errors indicate a higher probability of anomalies, as they deviate significantly from the majority patterns.…”

Multi-objective Deep Data Generation with Correlated Property Control

“…Extensive efforts have been spent on learning underlying low-dimensional representation and the generation process of high-dimensional data through deep generative models such as variational autoencoders (VAE) [27,35,9], generative adversarial networks (GANs) [11,12], normalizing flows [40,5], etc [48,17,8]. Particularly, enhancing the disentanglement and independence of latent dimensions has been attracting the attention of the community [4,43,3,34,45,23], enabling controllable generation that generates data with desired properties by interpolating latent variables [44,13,29,25,38,20,7,49]. For instance, CSVAE transfers image attributes by correlating latent variables with desired properties [28].…”