The ability of deep image prior (DIP) to recover high-quality images from incomplete or corrupted measurements has made it popular in inverse problems in image restoration and medical imaging including magnetic resonance imaging (MRI). However, conventional DIP suffers from severe overfitting and spectral bias effects. In this work, we first provide an analysis of how DIP recovers information from undersampled imaging measurements by analyzing the training dynamics of the underlying networks in the kernel regime for different architectures. This study sheds light on important underlying properties for DIP-based recovery. Current research suggests that incorporating a reference image as network input can enhance DIP's performance in image reconstruction compared to using random inputs. However, obtaining suitable reference images requires supervision, and raises practical difficulties. In an attempt to overcome this obstacle, we further introduce a self-driven reconstruction process that concurrently optimizes both the network weights and the input while eliminating the need for training data. Our method incorporates a novel denoiser regularization term which enables robust and stable joint estimation of both the network input and reconstructed image. We demonstrate that our self-guided method surpasses both the original DIP and modern supervised methods in terms of MR image reconstruction performance and outperforms previous DIP-based schemes for image inpainting.
In-Context Learning (ICL) empowers Large Language Models (LLMs) with the capacity to learn in context, achieving downstream generalization without gradient updates but with a few in-context examples. Despite the encouraging empirical success, the underlying mechanism of ICL remains unclear, and existing research offers various viewpoints of understanding. These studies propose intuition-driven and ad-hoc technical solutions for interpreting ICL, illustrating an ambiguous road map. In this paper, we leverage a data generation perspective to reinterpret recent efforts and demonstrate the potential broader usage of popular technical solutions, approaching a systematic angle. For a conceptual definition, we rigorously adopt the terms of skill learning and skill recognition. The difference between them is skill learning can learn new data generation functions from in-context data. We also provide a comprehensive study on the merits and weaknesses of different solutions, and highlight the uniformity among them given the perspective of data generation, establishing a technical foundation for future research to incorporate the strengths of different lines of research.
Fine-tuning pretrained language models (PLMs) for downstream tasks is a large-scale optimization problem, in which the choice of the training algorithm critically determines how well the trained model can generalize to unseen test data, especially in the context of few-shot learning. To achieve good generalization performance and avoid overfitting, techniques such as data augmentation and pruning are often applied. However, adding these regularizations necessitates heavy tuning of the hyperparameters of optimization algorithms, such as the popular Adam optimizer. In this paper, we propose a two-stage fine-tuning method, PAC-tuning, to address this optimization challenge. First, based on PAC-Bayes training, PAC-tuning directly minimizes the PAC-Bayes generalization bound to learn proper parameter distribution. Second, PAC-tuning modifies the gradient by injecting noise with the variance learned in the first stage into the model parameters during training, resulting in a variant of perturbed gradient descent (PGD). In the past, the few-shot scenario posed difficulties for PAC-Bayes training because the PAC-Bayes bound, when applied to large models with limited training data, might not be stringent. Our experimental results across 5 GLUE benchmark tasks demonstrate that PAC-tuning successfully handles the challenges of fine-tuning tasks and outperforms strong baseline methods by a visible margin, further confirming the potential to apply PAC training for any other settings where the Adam optimizer is currently used for training.
Deep learning (DL) methods have been extensively employed in magnetic resonance imaging (MRI) reconstruction, demonstrating remarkable performance improvements compared to traditional non-DL methods. However, recent studies have uncovered the susceptibility of these models to carefully engineered adversarial perturbations. In this paper, we tackle this issue by leveraging diffusion models. Specifically, we introduce a defense strategy that enhances the robustness of DL-based MRI reconstruction methods through the utilization of pre-trained diffusion models as adversarial purifiers. Unlike conventional state-of-the-art adversarial defense methods (e.g., adversarial training), our proposed approach eliminates the need to solve a minimax optimization problem to train the image reconstruction model from scratch, and only requires fine-tuning on purified adversarial examples. Our experimental findings underscore the effectiveness of our proposed technique when benchmarked against leading defense methodologies for MRI reconstruction such as adversarial training and randomized smoothing.
It is widely recognized that the generalization ability of neural networks can be greatly enhanced through carefully designing the training procedure. The current state-of-the-art training approach involves utilizing stochastic gradient descent (SGD) or Adam optimization algorithms along with a combination of additional regularization techniques such as weight decay, dropout, or noise injection. Optimal generalization can only be achieved by tuning a multitude of hyperparameters through grid search, which can be time-consuming and necessitates additional validation datasets. To address this issue, we introduce a practical PAC-Bayes training framework that is nearly tuning-free and requires no additional regularization while achieving comparable testing performance to that of SGD/Adam after a complete grid search and with extra regularizations. Our proposed algorithm demonstrates the remarkable potential of PAC training to achieve state-of-the-art performance on deep neural networks with enhanced robustness and interpretability.
It is well known that the finite step-size ($h$) in Gradient Descent (GD) implicitly regularizes solutions to flatter minima. A natural question to ask is "Does the momentum parameter $\beta$ play a role in implicit regularization in Heavy-ball (H.B) momentum accelerated gradient descent (GD+M)?". To answer this question, first, we show that the discrete H.B momentum update (GD+M) follows a continuous trajectory induced by a modified loss, which consists of an original loss and an implicit regularizer. Then, we show that this implicit regularizer for (GD+M) is stronger than that of (GD) by factor of $(\frac{1+\beta}{1-\beta})$, thus explaining why (GD+M) shows better generalization performance and higher test accuracy than (GD). Furthermore, we extend our analysis to the stochastic version of gradient descent with momentum (SGD+M) and characterize the continuous trajectory of the update of (SGD+M) in a pointwise sense. We explore the implicit regularization in (SGD+M) and (GD+M) through a series of experiments validating our theory.
The remarkable successes of neural networks in a huge variety of inverse problems have fueled their adoption in disciplines ranging from medical imaging to seismic analysis over the past decade. However, the high dimensionality of such inverse problems has simultaneously left current theory, which predicts that networks should scale exponentially in the dimension of the problem, unable to explain why the seemingly small networks used in these settings work as well as they do in practice. To reduce this gap between theory and practice, a general method for bounding the complexity required for a neural network to approximate a Lipschitz function on a high-dimensional set with a low-complexity structure is provided herein. The approach is based on the observation that the existence of a linear Johnson-Lindenstrauss embedding $\mathbf{A} \in \mathbb{R}^{d \times D}$ of a given high-dimensional set $\mathcal{S} \subset \mathbb{R}^D$ into a low dimensional cube $[-M,M]^d$ implies that for any Lipschitz function $f : \mathcal{S}\to \mathbb{R}^p$, there exists a Lipschitz function $g : [-M,M]^d \to \mathbb{R}^p$ such that $g(\mathbf{A}\mathbf{x}) = f(\mathbf{x})$ for all $\mathbf{x} \in \mathcal{S}$. Hence, if one has a neural network which approximates $g : [-M,M]^d \to \mathbb{R}^p$, then a layer can be added which implements the JL embedding $\mathbf{A}$ to obtain a neural network which approximates $f : \mathcal{S} \to \mathbb{R}^p$. By pairing JL embedding results along with results on approximation of Lipschitz functions by neural networks, one then obtains results which bound the complexity required for a neural network to approximate Lipschitz functions on high dimensional sets. The end result is a general theoretical framework which can then be used to better explain the observed empirical successes of smaller networks in a wider variety of inverse problems than current theory allows.
We constructively show, via rigorous mathematical arguments, that GNN architectures outperform those of NN in approximating bandlimited functions on compact $d$-dimensional Euclidean grids. We show that the former only need $\mathcal{M}$ sampled functional values in order to achieve a uniform approximation error of $O_{d}(2^{-\mathcal{M}^{1/d}})$ and that this error rate is optimal, in the sense that, NNs might achieve worse.
Image matting is an important computer vision problem. Many existing matting methods require a hand-made trimap to provide auxiliary information, which is very expensive and limits the real world usage. Recently, some trimap-free methods have been proposed, which completely get rid of any user input. However, their performance lag far behind trimap-based methods due to the lack of guidance information. In this paper, we propose a matting method that use Flexible Guidance Input as user hint, which means our method can use trimap, scribblemap or clickmap as guidance information or even work without any guidance input. To achieve this, we propose Progressive Trimap Deformation(PTD) scheme that gradually shrink the area of the foreground and background of the trimap with the training step increases and finally become a scribblemap. To make our network robust to any user scribble and click, we randomly sample points on foreground and background and perform curve fitting. Moreover, we propose Semantic Fusion Module(SFM) which utilize the Feature Pyramid Enhancement Module(FPEM) and Joint Pyramid Upsampling(JPU) in matting task for the first time. The experiments show that our method can achieve state-of-the-art results comparing with existing trimap-based and trimap-free methods.
Thanks to the combination of state-of-the-art accelerators and highly optimized open software frameworks, there has been tremendous progress in the performance of deep neural networks. While these developments have been responsible for many breakthroughs, progress towards solving large-scale problems, such as video encoding and semantic segmentation in 3D, is hampered because access to on-premise memory is often limited. Instead of relying on (optimal) checkpointing or invertibility of the network layers -- to recover the activations during backpropagation -- we propose to approximate the gradient of convolutional layers in neural networks with a multi-channel randomized trace estimation technique. Compared to other methods, this approach is simple, amenable to analyses, and leads to a greatly reduced memory footprint. Even though the randomized trace estimation introduces stochasticity during training, we argue that this is of little consequence as long as the induced errors are of the same order as errors in the gradient due to the use of stochastic gradient descent. We discuss the performance of networks trained with stochastic backpropagation and how the error can be controlled while maximizing memory usage and minimizing computational overhead.