Modern deep learning models are over-parameterized, where different optima can result in widely varying generalization performance. To account for this, Sharpness-Aware Minimization (SAM) modifies the underlying loss function to guide descent methods towards flatter minima, which arguably have better generalization abilities. In this paper, we focus on a variant of SAM known as micro-batch SAM (mSAM), which, during training, averages the updates generated by adversarial perturbations across several disjoint shards (micro batches) of a mini-batch. We extend a recently developed and well-studied general framework for flatness analysis to show that distributed gradient computation for sharpness-aware minimization theoretically achieves even flatter minima. In order to support this theoretical superiority, we provide a thorough empirical evaluation on a variety of image classification and natural language processing tasks. We also show that contrary to previous work, mSAM can be implemented in a flexible and parallelizable manner without significantly increasing computational costs. Our practical implementation of mSAM yields superior generalization performance across a wide range of tasks compared to SAM, further supporting our theoretical framework.
We propose a simple approach which combines the strengths of probabilistic graphical models and deep learning architectures for solving the multi-label classification task, focusing specifically on image and video data. First, we show that the performance of previous approaches that combine Markov Random Fields with neural networks can be modestly improved by leveraging more powerful methods such as iterative join graph propagation, integer linear programming, and $\ell_1$ regularization-based structure learning. Then we propose a new modeling framework called deep dependency networks, which augments a dependency network, a model that is easy to train and learns more accurate dependencies but is limited to Gibbs sampling for inference, to the output layer of a neural network. We show that despite its simplicity, jointly learning this new architecture yields significant improvements in performance over the baseline neural network. In particular, our experimental evaluation on three video activity classification datasets: Charades, Textually Annotated Cooking Scenes (TACoS), and Wetlab, and three multi-label image classification datasets: MS-COCO, PASCAL VOC, and NUS-WIDE show that deep dependency networks are almost always superior to pure neural architectures that do not use dependency networks.
Image quality assessment (IQA) is a fundamental metric for image processing tasks (e.g., compression). With full-reference IQAs, traditional IQAs, such as PSNR and SSIM, have been used. Recently, IQAs based on deep neural networks (deep IQAs), such as LPIPS and DISTS, have also been used. It is known that image scaling is inconsistent among deep IQAs, as some perform down-scaling as pre-processing, whereas others instead use the original image size. In this paper, we show that the image scale is an influential factor that affects deep IQA performance. We comprehensively evaluate four deep IQAs on the same five datasets, and the experimental results show that image scale significantly influences IQA performance. We found that the most appropriate image scale is often neither the default nor the original size, and the choice differs depending on the methods and datasets used. We visualized the stability and found that PieAPP is the most stable among the four deep IQAs.
This study compares sequential image classification methods based on recurrent neural networks. We describe methods based on recurrent neural networks such as Long-Short-Term memory(LSTM), bidirectional Long-Short-Term memory(BiLSTM) architectures, etc. We also review the state-of-the-art sequential image classification architectures. We mainly focus on LSTM, BiLSTM, temporal convolution network, and independent recurrent neural network architecture in the study. It is known that RNN lacks in learning long-term dependencies in the input sequence. We use a simple feature construction method using orthogonal Ramanujan periodic transform on the input sequence. Experiments demonstrate that if these features are given to LSTM or BiLSTM networks, the performance increases drastically. Our focus in this study is to increase the training accuracy simultaneously reducing the training time for the LSTM and BiLSTM architecture, but not on pushing the state-of-the-art results, so we use simple LSTM/BiLSTM architecture. We compare sequential input with the constructed feature as input to single layer LSTM and BiLSTM network for MNIST and CIFAR datasets. We observe that sequential input to the LSTM network with 128 hidden unit training for five epochs results in training accuracy of 33% whereas constructed features as input to the same LSTM network results in training accuracy of 90% with 1/3 lesser time.
Two key questions in cardiac image analysis are to assess the anatomy and motion of the heart from images; and to understand how they are associated with non-imaging clinical factors such as gender, age and diseases. While the first question can often be addressed by image segmentation and motion tracking algorithms, our capability to model and to answer the second question is still limited. In this work, we propose a novel conditional generative model to describe the 4D spatio-temporal anatomy of the heart and its interaction with non-imaging clinical factors. The clinical factors are integrated as the conditions of the generative modelling, which allows us to investigate how these factors influence the cardiac anatomy. We evaluate the model performance in mainly two tasks, anatomical sequence completion and sequence generation. The model achieves a high performance in anatomical sequence completion, comparable to or outperforming other state-of-the-art generative models. In terms of sequence generation, given clinical conditions, the model can generate realistic synthetic 4D sequential anatomies that share similar distributions with the real data.
Deep neural networks are vulnerable to adversarial attacks. We consider adversarial defense in the case of zero-shot image classification setting, which has rarely been explored because both adversarial defense and zero-shot learning are challenging. We propose LAAT, a novel Language-driven, Anchor-based Adversarial Training strategy, to improve the adversarial robustness in a zero-shot setting. LAAT uses a text encoder to obtain fixed anchors (normalized feature embeddings) of each category, then uses these anchors to perform adversarial training. The text encoder has the property that semantically similar categories can be mapped to neighboring anchors in the feature space. By leveraging this property, LAAT can make the image model adversarially robust on novel categories without any extra examples. Experimental results show that our method achieves impressive zero-shot adversarial performance, even surpassing the previous state-of-the-art adversarially robust one-shot methods in most attacking settings. When models are trained with LAAT on large datasets like ImageNet-1K, they can have substantial zero-shot adversarial robustness across several downstream datasets.
Heteroscedastic classifiers, which learn a multivariate Gaussian distribution over prediction logits, have been shown to perform well on image classification problems with hundreds to thousands of classes. However, compared to standard classifiers, they introduce extra parameters that scale linearly with the number of classes. This makes them infeasible to apply to larger-scale problems. In addition heteroscedastic classifiers introduce a critical temperature hyperparameter which must be tuned. We propose HET-XL, a heteroscedastic classifier whose parameter count when compared to a standard classifier scales independently of the number of classes. In our large-scale settings, we show that we can remove the need to tune the temperature hyperparameter, by directly learning it on the training data. On large image classification datasets with up to 4B images and 30k classes our method requires 14X fewer additional parameters, does not require tuning the temperature on a held-out set and performs consistently better than the baseline heteroscedastic classifier. HET-XL improves ImageNet 0-shot classification in a multimodal contrastive learning setup which can be viewed as a 3.5 billion class classification problem.
In this paper, we propose a method using the fusion of CNN and transformer structure to improve image classification performance. In the case of CNN, information about a local area on an image can be extracted well, but there is a limit to the extraction of global information. On the other hand, the transformer has an advantage in relatively global extraction, but has a disadvantage in that it requires a lot of memory for local feature value extraction. In the case of an image, it is converted into a feature map through CNN, and each feature map's pixel is considered a token. At the same time, the image is divided into patch areas and then fused with the transformer method that views them as tokens. For the fusion of tokens with two different characteristics, we propose three methods: (1) late token fusion with parallel structure, (2) early token fusion, (3) token fusion in a layer by layer. In an experiment using ImageNet 1k, the proposed method shows the best classification performance.
Adversarial attacks aims to perturb images such that a predictor outputs incorrect results. Due to the limited research in structured attacks, imposing consistency checks on natural multi-object scenes is a promising yet practical defense against conventional adversarial attacks. More desired attacks, to this end, should be able to fool defenses with such consistency checks. Therefore, we present the first approach GLOW that copes with various attack requests by generating global layout-aware adversarial attacks where both categorical and geometric layout constraints are explicitly established. Specifically, we focus on object detection task and given a victim image, GLOW first localizes victim objects according to target labels. And then it generates multiple attack plans, together with their context-consistency scores. Our proposed GLOW, on the one hand, is capable of handling various types of requests, including single or multiple victim objects, with or without specified victim objects. On the other hand, it produces a consistency score for each attack plan, reflecting the overall contextual consistency that both semantic category and global scene layout are considered. In experiment, we design multiple types of attack requests and validate our ideas on MS COCO validation set. Extensive experimental results demonstrate that we can achieve about 40$\%$ average relative improvement compared to state-of-the-art methods in conventional single object attack request; Moreover, our method outperforms SOTAs significantly on more generic attack requests by at least 30$\%$; Finally, our method produces superior performance under challenging zero-query black-box setting, or 30$\%$ better than SOTAs. Our code, model and attack requests would be made available.
Prototypical part network (ProtoPNet) methods have been designed to achieve interpretable classification by associating predictions with a set of training prototypes, which we refer to as trivial (i.e., easy-to-learn) prototypes because they are trained to lie far from the classification boundary in the feature space. Note that it is possible to make an analogy between ProtoPNet and support vector machine (SVM) given that the classification from both methods relies on computing similarity with a set of training points (i.e., trivial prototypes in ProtoPNet, and support vectors in SVM). However, while trivial prototypes are located far from the classification boundary, support vectors are located close to this boundary, and we argue that this discrepancy with the well-established SVM theory can result in ProtoPNet models with suboptimal classification accuracy. In this paper, we aim to improve the classification accuracy of ProtoPNet with a new method to learn support prototypes that lie near the classification boundary in the feature space, as suggested by the SVM theory. In addition, we target the improvement of classification interpretability with a new model, named ST-ProtoPNet, which exploits our support prototypes and the trivial prototypes to provide complementary interpretability information. Experimental results on CUB-200-2011, Stanford Cars, and Stanford Dogs datasets demonstrate that the proposed method achieves state-of-the-art classification accuracy and produces more visually meaningful and diverse prototypes.