Deep Generative Modelling: A Comparative Review of VAEs, GANs, Normalizing Flows, Energy-Based and Autoregressive Models

Article Computer Science, Artificial Intelligence

Potential Flow Generator With L2 Optimal Transport Regularity for Generative Models

Liu Yang et al.

Summary: We propose a potential flow generator with L-2 optimal transport regularity and demonstrate its effectiveness in various generative models and image translation tasks.

IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS (2022)

Add to Collection

Article Computer Science, Artificial Intelligence

Regularisation of neural networks by enforcing Lipschitz continuity

Henry Gouk et al.

Summary: The study investigates the effect of enforcing Lipschitz continuity of neural networks with respect to inputs, providing a technique for computing upper bound of Lipschitz constant for multiple p-norms. It formulates training with bounded Lipschitz constant as a constrained optimization problem and shows that resulting models outperform those trained with common regularizers. The study also demonstrates intuitive tuning of hyperparameters, impact of norm choice on model, and significant performance gains with limited training data.

MACHINE LEARNING (2021)

Add to Collection

Review Computer Science, Artificial Intelligence

Normalizing Flows: An Introduction and Review of Current Methods

Ivan Kobyzev et al.

Summary: Normalizing Flows are generative models that can produce tractable distributions, enabling efficient and accurate sampling and density evaluation. This survey article aims to provide a coherent and comprehensive review of the literature on the construction and use of Normalizing Flows for distribution learning, offering context and explanation of the models, reviewing current state-of-the-art literature, and identifying open questions and promising future directions.

IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE (2021)

Add to Collection

Article Computer Science, Artificial Intelligence

Self-Supervised Visual Feature Learning With Deep Neural Networks: A Survey

Longlong Jing et al.

Summary: This paper reviews deep learning-based self-supervised general visual feature learning methods, covering motivation, pipeline, architectures, schema, evaluation metrics, datasets, performance comparisons, and future directions.

IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE (2021)

Add to Collection

Article Physics, Multidisciplinary

A Neural Network MCMC Sampler That Maximizes Proposal Entropy

Zengyi Li et al.

Summary: The study proposes a method of maximizing proposal entropy through neural networks to adapt to distributions of any shape. By using gradients to generate proposals, the model achieved significantly higher efficiency in multiple sampling tasks, and demonstrated the reliability of the sampler through training a convergent energy-based model of natural images.

ENTROPY (2021)

Add to Collection

Article Computer Science, Interdisciplinary Applications

Learning to Segment From Scribbles Using Multi-Scale Adversarial Attention Gates

Gabriele Valvano et al.

Summary: This study explores image segmentation in medical imaging using scribble annotations, with the use of multi-scale GAN for generating segmentation masks and a novel attention gating mechanism for improved object localization. Experimental results demonstrate that the model performs comparably to fully annotated models on medical and non-medical datasets, with potential extensions to various settings.

IEEE TRANSACTIONS ON MEDICAL IMAGING (2021)

Add to Collection

Proceedings Paper Computer Science, Artificial Intelligence

Anycost GANs for Interactive Image Synthesis and Editing

Ji Lin et al.

Summary: This paper proposes Anycost GAN, which supports elastic resolutions and channels for faster image generation while providing outputs perceptually similar to the full generator, enabling interactive natural image editing. By utilizing techniques such as sampling-based multi-resolution training, adaptive-channel training, and a generator-conditioned discriminator, the anycost generator can achieve better image quality at various configurations.

2021 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION, CVPR 2021 (2021)

Add to Collection

Article Computer Science, Artificial Intelligence

On Data Augmentation for GAN Training

Ngoc-Trung Tran et al.

Summary: The study demonstrates that optimizing data augmentation in Generative Adversarial Networks can help the generator better learn the distribution of the original data, which is crucial for various fields such as medical applications.

IEEE TRANSACTIONS ON IMAGE PROCESSING (2021)

Add to Collection

Article Computer Science, Artificial Intelligence

Cooperative Training of Descriptor and Generator Networks

Jianwen Xie et al.

IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE (2020)

Add to Collection

Article Computer Science, Artificial Intelligence

Generative Adversarial Networks are special cases of Artificial Curiosity (1990) and also closely related to Predictability Minimization (1991)

Jurgen Schmidhuber

NEURAL NETWORKS (2020)

Add to Collection

Article Computer Science, Software Engineering

Neural Importance Sampling

Thomas Mueller et al.

ACM TRANSACTIONS ON GRAPHICS (2019)

Add to Collection

Review Computer Science, Artificial Intelligence

Generative adversarial network in medical imaging: A review

Xin Yi et al.

MEDICAL IMAGE ANALYSIS (2019)

Add to Collection

Proceedings Paper Engineering, Electrical & Electronic

DepthwiseGANs: Fast Training Generative Adversarial Networks for Realistic Image Synthesis

Mkhuseli Ngxande et al.

2019 SOUTHERN AFRICAN UNIVERSITIES POWER ENGINEERING CONFERENCE/ROBOTICS AND MECHATRONICS/PATTERN RECOGNITION ASSOCIATION OF SOUTH AFRICA (SAUPEC/ROBMECH/PRASA) (2019)

Add to Collection

Proceedings Paper Computer Science, Artificial Intelligence