计算机科学
反问题
算法
人工智能
数学优化
反向
信号处理
应用数学
数学
人工神经网络
估计理论
趋同(经济学)
理论计算机科学
计算复杂性理论
作者
Ismail Alkhouri,Evan A. Bell,Avrajit Ghosh,Shijun Liang,Rongrong Wang,Saiprasad Ravishankar
标识
DOI:10.1109/msp.2025.3632786
摘要
In recent years, deep learning (DL) methods have been extensively developed for inverse imaging problems (IIPs), encompassing supervised, self-supervised, and generative approaches. Most of these methods require large numbers of labeled or unlabeled training data to learn effective models. However, in many practical applications, such as medical image reconstruction, extensive training datasets are often unavailable or limited. A significant milestone in addressing this challenge came in 2018 with the work of Ulyanov et al.[1], which introduced the deep image prior (DIP)—the first training-data-free convolutional neural network (CNN) method for IIPs. Unlike conventional DL approaches, DIP requires only a CNN, the noisy measurements, and a forward operator. By leveraging the implicit regularization of deep networks initialized with random noise, DIP can learn and restore image structures without relying on external datasets. However, a well-known limitation of DIP is its susceptibility to overfitting, primarily due to overparameterization of the network. In this tutorial article, we provide a comprehensive review of DIP, including a theoretical analysis of its training dynamics. We also categorize and discuss recent advancements in DIP-based methods that are aimed at mitigating overfitting, including techniques such as regularization, network reparameterization, and early stopping. Furthermore, we discuss approaches that combine DIP with pretrained neural networks (NNs), present empirical comparison results against data-centric methods, and highlight open research questions and future directions.
科研通智能强力驱动
Strongly Powered by AbleSci AI