The inherent slow imaging speed of Magnetic Resonance Image (MRI) has spurred the development of various acceleration methods, typically through heuristically undersampling the MRI measurement domain known as k-space. Recently, deep neural networks have been applied to reconstruct undersampled k-space data and have shown improved reconstruction performance. While most of these methods focus on designing novel reconstruction networks or new training strategies for a given undersampling pattern, e.g., Cartesian undersampling or Non-Cartesian sampling, to date, there is limited research aiming to learn and optimize k-space sampling strategies using deep neural networks. This work proposes a novel optimization framework to learn k-space sampling trajectories by considering it as an Ordinary Differential Equation (ODE) problem that can be solved using neural ODE. In particular, the sampling of k-space data is framed as a dynamic system, in which neural ODE is formulated to approximate the system with additional constraints on MRI physics. In addition, we have also demonstrated that trajectory optimization and image reconstruction can be learned collaboratively for improved imaging efficiency and reconstruction performance. Experiments were conducted on different in-vivo datasets (e.g., brain and knee images) acquired with different sequences. Initial results have shown that our proposed method can generate better image quality in accelerated MRI than conventional undersampling schemes in Cartesian and Non-Cartesian acquisitions.
Low-rankness plays an important role in traditional machine learning, but is not so popular in deep learning. Most previous low-rank network compression methods compress the networks by approximating pre-trained models and re-training. However, optimal solution in the Euclidean space may be quite different from the one in the low-rank manifold. A well pre-trained model is not a good initialization for the model with low-rank constraint. Thus, the performance of low-rank compressed network degrades significantly. Compared to other network compression methods such as pruning, low-rank methods attracts less attention in recent years. In this paper, we devise a new training method, low-rank projection with energy transfer (LRPET), that trains low-rank compressed networks from scratch and achieves competitive performance. First, we propose to alternately perform stochastic gradient descent training and projection onto the low-rank manifold. This asymptotically approaches the optimal solution in the low-rank manifold. Compared to re-training on compact model, this enables fully utilization of model capacity since solution space is relaxed back to Euclidean space after projection. Second, the matrix energy (the sum of squares of singular values) reduction caused by projection is compensated by energy transfer. We uniformly transfer the energy of the pruned singular values to the remaining ones. We theoretically show that energy transfer eases the trend of gradient vanishing caused by projection. Comprehensive experiment on CIFAR-10 and ImageNet have justified that our method is superior to other low-rank compression methods and also outperforms recent state-of-the-art pruning methods.
Dynamic attention mechanism and global modeling ability make Transformer show strong feature learning ability. In recent years, Transformer has become comparable to CNNs methods in computer vision. This review mainly investigates the current research progress of Transformer in image and video applications, which makes a comprehensive overview of Transformer in visual learning understanding. First, the attention mechanism is reviewed, which plays an essential part in Transformer. And then, the visual Transformer model and the principle of each module are introduced. Thirdly, the existing Transformer-based models are investigated, and their performance is compared in visual learning understanding applications. Three image tasks and two video tasks of computer vision are investigated. The former mainly includes image classification, object detection, and image segmentation. The latter contains object tracking and video classification. It is significant for comparing different models' performance in various tasks on several public benchmark data sets. Finally, ten general problems are summarized, and the developing prospects of the visual Transformer are given in this review.
Approximating probability distributions can be a challenging task, particularly when they are supported over regions of high geometrical complexity or exhibit multiple modes. Annealing can be used to facilitate this task which is often combined with constant a priori selected increments in inverse temperature. However, using constant increments limit the computational efficiency due to the inability to adapt to situations where smooth changes in the annealed density could be handled equally well with larger increments. We introduce AdaAnn, an adaptive annealing scheduler that automatically adjusts the temperature increments based on the expected change in the Kullback-Leibler divergence between two distributions with a sufficiently close annealing temperature. AdaAnn is easy to implement and can be integrated into existing sampling approaches such as normalizing flows for variational inference and Markov chain Monte Carlo. We demonstrate the computational efficiency of the AdaAnn scheduler for variational inference with normalizing flows on a number of examples, including density approximation and parameter estimation for dynamical systems.
Training speaker-discriminative and robust speaker verification systems without speaker labels is still challenging and worthwhile to explore. In this study, we propose an effective self-supervised learning framework and a novel regularization strategy to facilitate self-supervised speaker representation learning. Different from contrastive learning-based self-supervised learning methods, the proposed self-supervised regularization (SSReg) focuses exclusively on the similarity between the latent representations of positive data pairs. We also explore the effectiveness of alternative online data augmentation strategies on both the time domain and frequency domain. With our strong online data augmentation strategy, the proposed SSReg shows the potential of self-supervised learning without using negative pairs and it can significantly improve the performance of self-supervised speaker representation learning with a simple Siamese network architecture. Comprehensive experiments on the VoxCeleb datasets demonstrate that our proposed self-supervised approach obtains a 23.4% relative improvement by adding the effective self-supervised regularization and outperforms other previous works.
The COVID-19 pandemic has affected societies and human health and well-being in various ways. In this study, we collected Reddit data from 2019 (pre-pandemic) and 2020 (pandemic) from the subreddits communities associated with 8 universities, applied natural language processing (NLP) techniques, and trained graphical neural networks with social media data, to study how the pandemic has affected people's emotions and psychological states compared to the pre-pandemic era. Specifically, we first applied a pre-trained Robustly Optimized BERT pre-training approach (RoBERTa) to learn embedding from the semantic information of Reddit messages and trained a graph attention network (GAT) for sentiment classification. The usage of GAT allows us to leverage the relational information among the messages during training. We then applied subgroup-adaptive model stacking to combine the prediction probabilities from RoBERTa and GAT to yield the final classification on sentiment. With the manually labeled and model-predicted sentiment labels on the collected data, we applied a generalized linear mixed-effects model to estimate the effects of pandemic and online teaching on people's sentiment in a statistically significant manner. The results suggest the odds of negative sentiments in 2020 is $14.6\%$ higher than the odds in 2019 ($p$-value $<0.001$), and the odds of negative sentiments are $41.6\%$ higher with in-person teaching than with online teaching in 2020 ($p$-value $=0.037$) in the studied population.
In recent years, Transformer has achieved good results in Natural Language Processing (NLP) and has also started to expand into Computer Vision (CV). Excellent models such as the Vision Transformer and Swin Transformer have emerged. At the same time, the platform for Transformer models was extended to embedded devices to meet some resource-sensitive application scenarios. However, due to the large number of parameters, the complex computational flow and the many different structural variants of Transformer models, there are a number of issues that need to be addressed in its hardware design. This is both an opportunity and a challenge. We propose Vis-TOP (Visual Transformer Overlay Processor), an overlay processor for various visual Transformer models. It differs from coarse-grained overlay processors such as CPU, GPU, NPE, and from fine-grained customized designs for a specific model. Vis-TOP summarizes the characteristics of all visual Transformer models and implements a three-layer and two-level transformation structure that allows the model to be switched or changed freely without changing the hardware architecture. At the same time, the corresponding instruction bundle and hardware architecture are designed in three-layer and two-level transformation structure. After quantization of Swin Transformer tiny model using 8-bit fixed points (fix_8), we implemented an overlay processor on the ZCU102. Compared to GPU, the TOP throughput is 1.5x higher. Compared to the existing Transformer accelerators, our throughput per DSP is between 2.2x and 11.7x higher than others. In a word, the approach in this paper meets the requirements of real-time AI in terms of both resource consumption and inference speed. Vis-TOP provides a cost-effective and power-effective solution based on reconfigurable devices for computer vision at the edge.
We propose Noise-Augmented Privacy-Preserving Empirical Risk Minimization (NAPP-ERM) that solves ERM with differential privacy guarantees. Existing privacy-preserving ERM approaches may be subject to over-regularization with the employment of an l2 term to achieve strong convexity on top of the target regularization. NAPP-ERM improves over the current approaches and mitigates over-regularization by iteratively realizing target regularization through appropriately designed augmented data and delivering strong convexity via a single adaptively weighted dual-purpose l2 regularizer. When the target regularization is for variable selection, we propose a new regularizer that achieves both privacy and sparsity guarantees simultaneously. Finally, we propose a strategy to retrieve privacy budget when the strong convexity requirement is met, which can be returned to users such that the DP of ERM is guaranteed at a lower privacy cost than originally planned, or be recycled to the ERM optimization procedure to reduce the injected DP noise and improve the utility of DP-ERM. From an implementation perspective, NAPP-ERM can be achieved by optimizing a non-perturbed object function given noise-augmented data and can thus leverage existing tools for non-private ERM optimization. We illustrate through extensive experiments the mitigation effect of the over-regularization and private budget retrieval by NAPP-ERM on variable selection and prediction.
This report summarizes the results of Learning to Understand Aerial Images (LUAI) 2021 challenge held on ICCV 2021, which focuses on object detection and semantic segmentation in aerial images. Using DOTA-v2.0 and GID-15 datasets, this challenge proposes three tasks for oriented object detection, horizontal object detection, and semantic segmentation of common categories in aerial images. This challenge received a total of 146 registrations on the three tasks. Through the challenge, we hope to draw attention from a wide range of communities and call for more efforts on the problems of learning to understand aerial images.
Fast inference of numerical model parameters from data is an important prerequisite to generate predictive models for a wide range of applications. Use of sampling-based approaches such as Markov chain Monte Carlo may become intractable when each likelihood evaluation is computationally expensive. New approaches combining variational inference with normalizing flow are characterized by a computational cost that grows only linearly with the dimensionality of the latent variable space, and rely on gradient-based optimization instead of sampling, providing a more efficient approach for Bayesian inference about the model parameters. Moreover, the cost of frequently evaluating an expensive likelihood can be mitigated by replacing the true model with an offline trained surrogate model, such as neural networks. However, this approach might generate significant bias when the surrogate is insufficiently accurate around the posterior modes. To reduce the computational cost without sacrificing inferential accuracy, we propose Normalizing Flow with Adaptive Surrogate (NoFAS), an optimization strategy that alternatively updates the normalizing flow parameters and the weights of a neural network surrogate model. We also propose an efficient sample weighting scheme for surrogate model training that ensures some global accuracy of the surrogate while capturing the likely regions of the parameters that yield the observed data. We demonstrate the inferential and computational superiority of NoFAS against various benchmarks, including cases where the underlying model lacks identifiability. The source code and numerical experiments used for this study are available at https://github.com/cedricwangyu/NoFAS.