Search

Recent studies have shown that regularization techniques using soft labels, e.g., label smoothing, Mixup, and CutMix, not only enhance image classification accuracy but also improve model calibration and robustness against adversarial attacks. However, the underlying mechanisms of such improvements remain underexplored. In this paper, we offer a novel explanation from the perspective of the representation space (i.e., the space of the features obtained at the penultimate layer). Our investigation first reveals that the decision regions in the representation space form cone-like shapes around the origin after training regardless of the presence of regularization. However, applying regularization causes changes in the distribution of features (or representation vectors). The magnitudes of the representation vectors are reduced and subsequently the cosine similarities between the representation vectors and the class centers (minimal loss points for each class) become higher, which acts as a central mechanism inducing improved calibration and robustness. Our findings provide new insights into the characteristics of the high-dimensional representation space in relation to training and regularization using soft labels.

Logit-based knowledge distillation (KD) for classification is cost-efficient compared to feature-based KD but often subject to inferior performance. Recently, it was shown that the performance of logit-based KD can be improved by effectively delivering the probability distribution for the non-target classes from the teacher model, which is known as `implicit (dark) knowledge', to the student model. Through gradient analysis, we first show that this actually has an effect of adaptively controlling the learning of implicit knowledge. Then, we propose a new loss that enables the student to learn explicit knowledge (i.e., the teacher's confidence about the target class) along with implicit knowledge in an adaptive manner. Furthermore, we propose to separate the classification and distillation tasks for effective distillation and inter-class relationship modeling. Experimental results demonstrate that the proposed method, called adaptive explicit knowledge transfer (AEKT) method, achieves improved performance compared to the state-of-the-art KD methods on the CIFAR-100 and ImageNet datasets., Comment: 19 pages, 5 figures

Recent advancements in low-cost ensemble learning have demonstrated improved efficiency for image classification. However, the existing low-cost ensemble methods show relatively lower accuracy compared to conventional ensemble learning. In this paper, we propose a new low-cost ensemble learning, which can simultaneously achieve high efficiency and classification performance. A CNN is transformed into a multi-branch structure without introduction of additional components, which maintains the computational complexity as that of the original single model and also enhances diversity among the branches' outputs via sufficient separation between different pathways of the branches. In addition, we propose a new strategy that applies grouped convolution in the branches with different numbers of groups in different branches, which boosts the diversity of the branches' outputs. For training, we employ knowledge distillation using the ensemble of the outputs as the teacher signal. The high diversity among the outputs enables to form a powerful teacher, enhancing the individual branch's classification performance and consequently the overall ensemble performance. Experimental results show that our method achieves state-of-the-art classification accuracy and higher uncertainty estimation performance compared to previous low-cost ensemble methods. The code is available at https://github.com/hjdw2/SEMBG.

Recent ensemble learning methods for image classification have been shown to improve classification accuracy with low extra cost. However, they still require multiple trained models for ensemble inference, which eventually becomes a significant burden when the model size increases. In this paper, we propose a low-cost ensemble learning and inference, called Network Fission Ensembles (NFE), by converting a conventional network itself into a multi-exit structure. Starting from a given initial network, we first prune some of the weights to reduce the training burden. We then group the remaining weights into several sets and create multiple auxiliary paths using each set to construct multi-exits. We call this process Network Fission. Through this, multiple outputs can be obtained from a single network, which enables ensemble learning. Since this process simply changes the existing network structure to multi-exits without using additional networks, there is no extra computational burden for ensemble learning and inference. Moreover, by learning from multiple losses of all exits, the multi-exits improve performance via regularization, and high performance can be achieved even with increased network sparsity. With our simple yet effective method, we achieve significant improvement compared to existing ensemble methods. The code is available at https://github.com/hjdw2/NFE.

Magnetic anisotropy in atomically thin correlated heterostructures is essential for exploring quantum magnetic phases for next-generation spintronics. Whereas previous studies have mostly focused on van der Waals systems, here, we investigate the impact of dimensionality of epitaxially-grown correlated oxides down to the monolayer limit on structural, magnetic, and orbital anisotropies. By designing oxide superlattices with a correlated ferromagnetic SrRuO3 and nonmagnetic SrTiO3 layers, we observed modulated ferromagnetic behavior with the change of the SrRuO3 thickness. Especially, for three-unit-cell-thick layers, we observe a significant 1,500% improvement of coercive field in the anomalous Hall effect, which cannot be solely attributed to the dimensional crossover in ferromagnetism. The atomic-scale heterostructures further reveal the systematic modulation of anisotropy for the lattice structure and orbital hybridization, explaining the enhanced magnetic anisotropy. Our findings provide valuable insights into engineering the anisotropic hybridization of synthetic magnetic crystals, offering a tunable spin order for various applications., Comment: 23 pages

Altermagnetism refers to a wide class of compensated magnetic orders featuring magnetic sublattices with opposite spins related by rotational symmetry rather than inversion or translational operations, resulting in non-trivial spin splitting and high-order multipolar orders. Here, by combining theoretical analysis, electrical transport, X-ray and optical spectroscopies, and nonlinear optical measurements, we establish a phase diagram in hybrid molecular beam epitaxy-grown RuO2/TiO2 (110) films, mapping the broken symmetries along the altermagnetic/electronic/structural phase transitions as functions of film thickness and temperature. This phase diagram features a novel altermagnetic metallic polar phase in strained 2 nm samples, extending the concept of multiferroics to altermagnetic systems. These results provide a comprehensive understanding of altermagnetism upon epitaxial heterostructure design for emergent novel phases with multifunctionalities., Comment: 15 pages

Inversion symmetry breaking is critical for many quantum effects and fundamental for spin-orbit torque, which is crucial for next-generation spintronics. Recently, a novel type of gigantic intrinsic spin-orbit torque has been established in the topological van-der-Waals (vdW) magnet iron germanium telluride. However, it remains a puzzle because no clear evidence exists for interlayer inversion symmetry breaking. Here, we report the definitive evidence of broken inversion symmetry in iron germanium telluride directly measured by the second harmonic generation (SHG) technique. Our data show that the crystal symmetry reduces from centrosymmetric P63/mmc to noncentrosymmetric polar P3m1 space group, giving the three-fold SHG pattern with dominant out-of-plane polarization. Additionally, the SHG response evolves from an isotropic pattern to a sharp three-fold symmetry upon increasing Fe deficiency, mainly due to the transition from random defects to ordered Fe vacancies. Such SHG response is robust against temperature, ensuring unaltered crystalline symmetries above and below the ferromagnetic transition temperature. These findings add crucial new information to our understanding of this interesting vdW metal, iron germanium telluride: band topology, intrinsic spin-orbit torque and topological vdW polar metal states., Comment: 32 pages, 9 figures, Accepted by Advanced Materials

With the advancement of generative models, the assessment of generated images becomes more and more important. Previous methods measure distances between features of reference and generated images from trained vision models. In this paper, we conduct an extensive investigation into the relationship between the representation space and input space around generated images. We first propose two measures related to the presence of unnatural elements within images: complexity, which indicates how non-linear the representation space is, and vulnerability, which is related to how easily the extracted feature changes by adversarial input changes. Based on these, we introduce a new metric to evaluating image-generative models called anomaly score (AS). Moreover, we propose AS-i (anomaly score for individual images) that can effectively evaluate generated images individually. Experimental results demonstrate the validity of the proposed approach., Comment: Accepted in CVPR 2024

Human-Object Interaction (HOI) detection, which localizes and infers relationships between human and objects, plays an important role in scene understanding. Although two-stage HOI detectors have advantages of high efficiency in training and inference, they suffer from lower performance than one-stage methods due to the old backbone networks and the lack of considerations for the HOI perception process of humans in the interaction classifiers. In this paper, we propose Vision Transformer based Pose-Conditioned Self-Loop Graph (ViPLO) to resolve these problems. First, we propose a novel feature extraction method suitable for the Vision Transformer backbone, called masking with overlapped area (MOA) module. The MOA module utilizes the overlapped area between each patch and the given region in the attention function, which addresses the quantization problem when using the Vision Transformer backbone. In addition, we design a graph with a pose-conditioned self-loop structure, which updates the human node encoding with local features of human joints. This allows the classifier to focus on specific human joints to effectively identify the type of interaction, which is motivated by the human perception process for HOI. As a result, ViPLO achieves the state-of-the-art results on two public benchmarks, especially obtaining a +2.07 mAP performance gain on the HICO-DET dataset. The source codes are available at https://github.com/Jeeseung-Park/ViPLO., Comment: CVPR 2023

In recent years, many deep neural architectures have been developed for image classification. Whether they are similar or dissimilar and what factors contribute to their (dis)similarities remains curious. To address this question, we aim to design a quantitative and scalable similarity measure between neural architectures. We propose Similarity by Attack Transferability (SAT) from the observation that adversarial attack transferability contains information related to input gradients and decision boundaries widely used to understand model behaviors. We conduct a large-scale analysis on 69 state-of-the-art ImageNet classifiers using our proposed similarity function to answer the question. Moreover, we observe neural architecture-related phenomena using model similarity that model diversity can lead to better performance on model ensembles and knowledge distillation under specific conditions. Our results provide insights into why developing diverse neural architectures with distinct components is necessary., Comment: ECCV 2024; 35pages, 2.56MB

Neural networks with self-attention (a.k.a. Transformers) like ViT and Swin have emerged as a better alternative to traditional convolutional neural networks (CNNs). However, our understanding of how the new architecture works is still limited. In this paper, we focus on the phenomenon that Transformers show higher robustness against corruptions than CNNs, while not being overconfident. This is contrary to the intuition that robustness increases with confidence. We resolve this contradiction by empirically investigating how the output of the penultimate layer moves in the representation space as the input data moves linearly within a small area. In particular, we show the following. (1) While CNNs exhibit fairly linear relationship between the input and output movements, Transformers show nonlinear relationship for some data. For those data, the output of Transformers moves in a curved trajectory as the input moves linearly. (2) When a data is located in a curved region, it is hard to move it out of the decision region since the output moves along a curved trajectory instead of a straight line to the decision boundary, resulting in high robustness of Transformers. (3) If a data is slightly modified to jump out of the curved region, the movements afterwards become linear and the output goes to the decision boundary directly. In other words, there does exist a decision boundary near the data, which is hard to find only because of the curved representation space. This explains the underconfident prediction of Transformers. Also, we examine mathematical properties of the attention operation that induce nonlinear response to linear perturbation. Finally, we share our additional findings, regarding what contributes to the curved representation space of Transformers, and how the curvedness evolves during training., Comment: AAAI 2024

If managers assume a normal or near-normal distribution of Information Technology (IT) project cost overruns, as is common, and cost overruns can be shown to follow a power-law distribution, managers may be unwittingly exposing their organizations to extreme risk by severely underestimating the probability of large cost overruns. In this research, we collect and analyze a large sample comprised of 5,392 IT projects to empirically examine the probability distribution of IT project cost overruns. Further, we propose and examine a mechanism that can explain such a distribution. Our results reveal that IT projects are far riskier in terms of cost than normally assumed by decision makers and scholars. Specifically, we found that IT project cost overruns follow a power-law distribution in which there are a large number of projects with relatively small overruns and a fat tail that includes a smaller number of projects with extreme overruns. A possible generative mechanism for the identified power-law distribution is found in interdependencies among technological components in IT systems. We propose and demonstrate, through computer simulation, that a problem in a single technological component can lead to chain reactions in which other interdependent components are affected, causing substantial overruns. What the power law tells us is that extreme IT project cost overruns will occur and that the prevalence of these will be grossly underestimated if managers assume that overruns follow a normal or near-normal distribution. This underscores the importance of realistically assessing and mitigating the cost risk of new IT projects up front.

Assessing image aesthetics is a challenging computer vision task. One reason is that aesthetic preference is highly subjective and may vary significantly among people for certain images. Thus, it is important to properly model and quantify such \textit{subjectivity}, but there has not been much effort to resolve this issue. In this paper, we propose a novel unified probabilistic framework that can model and quantify subjective aesthetic preference based on the subjective logic. In this framework, the rating distribution is modeled as a beta distribution, from which the probabilities of being definitely pleasing, being definitely unpleasing, and being uncertain can be obtained. We use the probability of being uncertain to define an intuitive metric of subjectivity. Furthermore, we present a method to learn deep neural networks for prediction of image aesthetics, which is shown to be effective in improving the performance of subjectivity prediction via experiments. We also present an application scenario where the framework is beneficial for aesthetics-based image recommendation., Comment: 8 pages, 6 figures

The Vision Transformer has emerged as a powerful tool for image classification tasks, surpassing the performance of convolutional neural networks (CNNs). Recently, many researchers have attempted to understand the robustness of Transformers against adversarial attacks. However, previous researches have focused solely on perturbations in the spatial domain. This paper proposes an additional perspective that explores the adversarial robustness of Transformers against frequency-selective perturbations in the spectral domain. To facilitate comparison between these two domains, an attack framework is formulated as a flexible tool for implementing attacks on images in the spatial and spectral domains. The experiments reveal that Transformers rely more on phase and low frequency information, which can render them more vulnerable to frequency-selective attacks than CNNs. This work offers new insights into the properties and adversarial robustness of Transformers., Comment: Accepted in IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) 2024

Evaluation of generative models is mostly based on the comparison between the estimated distribution and the ground truth distribution in a certain feature space. To embed samples into informative features, previous works often use convolutional neural networks optimized for classification, which is criticized by recent studies. Therefore, various feature spaces have been explored to discover alternatives. Among them, a surprising approach is to use a randomly initialized neural network for feature embedding. However, the fundamental basis to employ the random features has not been sufficiently justified. In this paper, we rigorously investigate the feature space of models with random weights in comparison to that of trained models. Furthermore, we provide an empirical evidence to choose networks for random features to obtain consistent and reliable results. Our results indicate that the features from random networks can evaluate generative models well similarly to those from trained networks, and furthermore, the two types of features can be used together in a complementary way., Comment: Accepted in AAAI 2023

Electric fields have played a key role in discovering and controlling exotic electronic states of condensed matter. However, electric fields usually do not work in metals as free carriers tend to screen electrostatic fields. While a pseudo-electric field generated by inhomogeneous lattice strain, namely a flexoelectric field, can in principle work in all classes of materials, it remains experimentally unexplored in metals. Here, using heteroepitaxy and atomic-scale imaging, we show that flexoelectric fields can polarize a metallic oxide SrRuO3 with unexpectedly large Ru off-center displacements. We also observe that the flexoelectrically induced polar state of SrRuO3 leads to sizable lattice expansion, similar to the electrostrictive expansion caused by ionic displacements in dielectrics under an external electric field. We further suggest that flexoelectrically driven Ru off-centering promotes strong coupling between lattice and electronic degrees of freedom, possibly enhancing the ferromagnetism of SrRuO3. Beyond conventional electric fields, flexoelectric fields may universally engender novel electronic states and their control via pure atomic displacements in a nondestructive and fast manner., Comment: 33 pages, 13 figures

Matter-light interaction is at the center of diverse research fields from quantum optics to condensed matter physics, opening new fields like laser physics. A magnetic exciton is one such rare example found in magnetic insulators. However, it is relatively rare to observe that external variables control matter-light interaction. Here, we report that the broken inversion symmetry of multiferroicity can act as an external knob enabling the magnetic exciton in van der Waals antiferromagnet NiI2. We further discover that this magnetic exciton arises from a transition between Zhang-Rice-triplet and Zhang-Rice-singlet's fundamentally quantum entangled states. This quantum entanglement produces an ultra-sharp optical exciton peak at 1.384 eV with a 5 meV linewidth. Our work demonstrates that NiI2 is two-dimensional magnetically ordered with an intrinsically quantum entangled ground state., Comment: 26 pages, 10 figures. Accepted for publication in Advanced Materials

Recently, video-based action recognition methods using convolutional neural networks (CNNs) achieve remarkable recognition performance. However, there is still lack of understanding about the generalization mechanism of action recognition models. In this paper, we suggest that action recognition models rely on the motion information less than expected, and thus they are robust to randomization of frame orders. Furthermore, we find that motion monotonicity remaining after randomization also contributes to such robustness. Based on this observation, we develop a novel defense method using temporal shuffling of input videos against adversarial attacks for action recognition models. Another observation enabling our defense method is that adversarial perturbations on videos are sensitive to temporal destruction. To the best of our knowledge, this is the first attempt to design a defense method without additional training for 3D CNN-based video action recognition models., Comment: 12 pages, accepted to Neural Networks

While adversarial perturbation of images to attack deep image classification models pose serious security concerns in practice, this paper suggests a novel paradigm where the concept of image perturbation can benefit classification performance, which we call amicable aid. We show that by taking the opposite search direction of perturbation, an image can be modified to yield higher classification confidence and even a misclassified image can be made correctly classified. This can be also achieved with a large amount of perturbation by which the image is made unrecognizable by human eyes. The mechanism of the amicable aid is explained in the viewpoint of the underlying natural image manifold. Furthermore, we investigate the universal amicable aid, i.e., a fixed perturbation can be applied to multiple images to improve their classification results. While it is challenging to find such perturbations, we show that making the decision boundary as perpendicular to the image manifold as possible via training with modified data is effective to obtain a model for which universal amicable perturbations are more easily found., Comment: ICASSP 2023

Recently, learned image compression methods have outperformed traditional hand-crafted ones including BPG. One of the keys to this success is learned entropy models that estimate the probability distribution of the quantized latent representation. Like other vision tasks, most recent learned entropy models are based on convolutional neural networks (CNNs). However, CNNs have a limitation in modeling long-range dependencies due to their nature of local connectivity, which can be a significant bottleneck in image compression where reducing spatial redundancy is a key point. To overcome this issue, we propose a novel entropy model called Information Transformer (Informer) that exploits both global and local information in a content-dependent manner using an attention mechanism. Our experiments show that Informer improves rate--distortion performance over the state-of-the-art methods on the Kodak and Tecnick datasets without the quadratic computational complexity problem. Our source code is available at https://github.com/naver-ai/informer., Comment: CVPR 2022

A significant amount of work has been done on adversarial attacks that inject imperceptible noise to images to deteriorate the image classification performance of deep models. However, most of the existing studies consider attacks in the digital (pixel) domain where an image acquired by an image sensor with sampling and quantization has been recorded. This paper, for the first time, introduces an optical adversarial attack, which physically alters the light field information arriving at the image sensor so that the classification model yields misclassification. More specifically, we modulate the phase of the light in the Fourier domain using a spatial light modulator placed in the photographic system. The operative parameters of the modulator are obtained by gradient-based optimization to maximize cross-entropy and minimize distortions. We present experiments based on both simulation and a real hardware optical system, from which the feasibility of the proposed optical attack is demonstrated. It is also verified that the proposed attack is completely different from common optical-domain distortions such as spherical aberration, defocus, and astigmatism in terms of both perturbation patterns and classification results., Comment: 11 pages, 4 figures, author names corrected

Realizing a state of matter in two dimensions has repeatedly proven a novel route of discovering new physical phenomena. Van der Waals (vdW) materials have been at the center of these now extensive research activities. They offer a natural way of producing a monolayer of matter simply by mechanical exfoliation. This work demonstrates that the possible multiferroic state with coexisting antiferromagnetic and ferroelectric orders possibly persists down to the bilayer flake of NiI$_{2}$. By exploiting the optical second-harmonic generation technique, both magnitude and direction of the ferroelectric order, arising from the cycloidal spin order, are successfully traced. The possible multiferroic state's transition temperature decreases from 58 K for the bulk to about 20 K for the bilayer. Our observation will spur extensive efforts to demonstrate multi-functionality in vdW materials, which have been tried mostly by using heterostructures of singly ferroic ones until now.

Recently, the vulnerability of deep image classification models to adversarial attacks has been investigated. However, such an issue has not been thoroughly studied for image-to-image tasks that take an input image and generate an output image (e.g., colorization, denoising, deblurring, etc.) This paper presents comprehensive investigations into the vulnerability of deep image-to-image models to adversarial attacks. For five popular image-to-image tasks, 16 deep models are analyzed from various standpoints such as output quality degradation due to attacks, transferability of adversarial examples across different tasks, and characteristics of perturbations. We show that unlike image classification tasks, the performance degradation on image-to-image tasks largely differs depending on various factors, e.g., attack methods and task objectives. In addition, we analyze the effectiveness of conventional defense methods used for classification models in improving the robustness of the image-to-image models., Comment: ICPR2022

Evaluating image generation models such as generative adversarial networks (GANs) is a challenging problem. A common approach is to compare the distributions of the set of ground truth images and the set of generated test images. The Frech\'et Inception distance is one of the most widely used metrics for evaluation of GANs, which assumes that the features from a trained Inception model for a set of images follow a normal distribution. In this paper, we argue that this is an over-simplified assumption, which may lead to unreliable evaluation results, and more accurate density estimation can be achieved using a truncated generalized normal distribution. Based on this, we propose a novel metric for accurate evaluation of GANs, named TREND (TRuncated gEneralized Normal Density estimation of inception embeddings). We demonstrate that our approach significantly reduces errors of density estimation, which consequently eliminates the risk of faulty evaluation results. Furthermore, we show that the proposed metric significantly improves robustness of evaluation results against variation of the number of image samples., Comment: Accepted in ECCV 2022

This paper proposes a novel knowledge distillation-based learning method to improve the classification performance of convolutional neural networks (CNNs) without a pre-trained teacher network, called exit-ensemble distillation. Our method exploits the multi-exit architecture that adds auxiliary classifiers (called exits) in the middle of a conventional CNN, through which early inference results can be obtained. The idea of our method is to train the network using the ensemble of the exits as the distillation target, which greatly improves the classification performance of the overall network. Our method suggests a new paradigm of knowledge distillation; unlike the conventional notion of distillation where teachers only teach students, we show that students can also help other students and even the teacher to learn better. Experimental results demonstrate that our method achieves significant improvement of classification performance on various popular CNN architectures (VGG, ResNet, ResNeXt, WideResNet, etc.). Furthermore, the proposed method can expedite the convergence of learning with improved stability. Our code will be available on Github.

In this paper, we propose a novel model-parallel learning method, called local critic training, which trains neural networks using additional modules called local critic networks. The main network is divided into several layer groups and each layer group is updated through error gradients estimated by the corresponding local critic network. We show that the proposed approach successfully decouples the update process of the layer groups for both convolutional neural networks (CNNs) and recurrent neural networks (RNNs). In addition, we demonstrate that the proposed method is guaranteed to converge to a critical point. We also show that trained networks by the proposed method can be used for structural optimization. Experimental results show that our method achieves satisfactory performance, reduces training time greatly, and decreases memory consumption per machine. Code is available at https://github.com/hjdw2/Local-critic-training.