Shin'ya Yamaguchi

Daiki Chijiwa, Shin'ya Yamaguchi, Atsutoshi Kumagai et al.

Few-shot learning for neural networks (NNs) is an important problem that aims to train NNs with a few data. The main challenge is how to avoid overfitting since over-parameterized NNs can easily overfit to such small dataset. Previous work (e.g. MAML by Finn et al. 2017) tackles this challenge by meta-learning, which learns how to learn from a few data by using various tasks. On the other hand, one conventional approach to avoid overfitting is restricting hypothesis spaces by endowing sparse NN structures like convolution layers in computer vision. However, although such manually-designed sparse structures are sample-efficient for sufficiently large datasets, they are still insufficient for few-shot learning. Then the following questions naturally arise: (1) Can we find sparse structures effective for few-shot learning by meta-learning? (2) What benefits will it bring in terms of meta-generalization? In this work, we propose a novel meta-learning approach, called Meta-ticket, to find optimal sparse subnetworks for few-shot learning within randomly initialized NNs. We empirically validated that Meta-ticket successfully discover sparse subnetworks that can learn specialized features for each given task. Due to this task-wise adaptation ability, Meta-ticket achieves superior meta-generalization compared to MAML-based methods especially with large NNs. The code is available at: https://github.com/dchiji-ntt/meta-ticket

Shin'ya Yamaguchi, Daiki Chijiwa, Sekitoshi Kanai et al.

This paper investigates methods for improving generative data augmentation for deep learning. Generative data augmentation leverages the synthetic samples produced by generative models as an additional dataset for classification with small dataset settings. A key challenge of generative data augmentation is that the synthetic data contain uninformative samples that degrade accuracy. This is because the synthetic samples do not perfectly represent class categories in real data and uniform sampling does not necessarily provide useful samples for tasks. In this paper, we present a novel strategy for generative data augmentation called meta generative regularization (MGR). To avoid the degradation of generative data augmentation, MGR utilizes synthetic samples in the regularization term for feature extractors instead of in the loss function, e.g., cross-entropy. These synthetic samples are dynamically determined to minimize the validation losses through meta-learning. We observed that MGR can avoid the performance degradation of naïve generative data augmentation and boost the baselines. Experiments on six datasets showed that MGR is effective particularly when datasets are smaller and stably outperforms baselines.

Shin'ya Yamaguchi, Sekitoshi Kanai, Atsutoshi Kumagai et al.

Transfer learning is crucial in training deep neural networks on new target tasks. Current transfer learning methods always assume at least one of (i) source and target task label spaces overlap, (ii) source datasets are available, and (iii) target network architectures are consistent with source ones. However, holding these assumptions is difficult in practical settings because the target task rarely has the same labels as the source task, the source dataset access is restricted due to storage costs and privacy, and the target architecture is often specialized to each task. To transfer source knowledge without these assumptions, we propose a transfer learning method that uses deep generative models and is composed of the following two stages: pseudo pre-training (PP) and pseudo semi-supervised learning (P-SSL). PP trains a target architecture with an artificial dataset synthesized by using conditional source generative models. P-SSL applies SSL algorithms to labeled target data and unlabeled pseudo samples, which are generated by cascading the source classifier and generative models to condition them with target samples. Our experimental results indicate that our method can outperform the baselines of scratch training and knowledge distillation.

Shin'ya Yamaguchi, Kosuke Nishida

Recent concept-based interpretable models have succeeded in providing meaningful explanations by pre-defined concept sets. However, the dependency on the pre-defined concepts restricts the application because of the limited number of concepts for explanations. This paper proposes a novel interpretable deep neural network called explanation bottleneck models (XBMs). XBMs generate a text explanation from the input without pre-defined concepts and then predict a final task prediction based on the generated explanation by leveraging pre-trained vision-language encoder-decoder models. To achieve both the target task performance and the explanation quality, we train XBMs through the target task loss with the regularization penalizing the explanation decoder via the distillation from the frozen pre-trained decoder. Our experiments, including a comparison to state-of-the-art concept bottleneck models, confirm that XBMs provide accurate and fluent natural language explanations without pre-defined concept sets. Code is available at https://github.com/yshinya6/xbm/.

Satoshi Suzuki, Shin'ya Yamaguchi, Shoichiro Takeda et al.

This paper addresses the tradeoff between standard accuracy on clean examples and robustness against adversarial examples in deep neural networks (DNNs). Although adversarial training (AT) improves robustness, it degrades the standard accuracy, thus yielding the tradeoff. To mitigate this tradeoff, we propose a novel AT method called ARREST, which comprises three components: (i) adversarial finetuning (AFT), (ii) representation-guided knowledge distillation (RGKD), and (iii) noisy replay (NR). AFT trains a DNN on adversarial examples by initializing its parameters with a DNN that is standardly pretrained on clean examples. RGKD and NR respectively entail a regularization term and an algorithm to preserve latent representations of clean examples during AFT. RGKD penalizes the distance between the representations of the standardly pretrained and AFT DNNs. NR switches input adversarial examples to nonadversarial ones when the representation changes significantly during AFT. By combining these components, ARREST achieves both high standard accuracy and robustness. Experimental results demonstrate that ARREST mitigates the tradeoff more effectively than previous AT-based methods do.

Sekitoshi Kanai, Shin'ya Yamaguchi, Masanori Yamada et al.

This paper proposes a new loss function for adversarial training. Since adversarial training has difficulties, e.g., necessity of high model capacity, focusing on important data points by weighting cross-entropy loss has attracted much attention. However, they are vulnerable to sophisticated attacks, e.g., Auto-Attack. This paper experimentally reveals that the cause of their vulnerability is their small margins between logits for the true label and the other labels. Since neural networks classify the data points based on the logits, logit margins should be large enough to avoid flipping the largest logit by the attacks. Importance-aware methods do not increase logit margins of important samples but decrease those of less-important samples compared with cross-entropy loss. To increase logit margins of important samples, we propose switching one-vs-the-rest loss (SOVR), which switches from cross-entropy to one-vs-the-rest loss for important samples that have small logit margins. We prove that one-vs-the-rest loss increases logit margins two times larger than the weighted cross-entropy loss for a simple problem. We experimentally confirm that SOVR increases logit margins of important samples unlike existing methods and achieves better robustness against Auto-Attack than importance-aware methods.

Masanori Yamada, Tomoya Yamashita, Shin'ya Yamaguchi et al.

Model merging is attracting attention as a novel method for creating a new model by combining the weights of different trained models. While previous studies reported that model merging works well for models trained on a single dataset with different random seeds, model merging between different datasets remains unsolved. In this paper, we attempt to reveal the difficulty in merging such models trained on different datasets and alleviate it. Our empirical analyses show that, in contrast to the single-dataset scenarios, dataset information needs to be accessed to achieve high accuracy when merging models trained on different datasets. However, the requirement to use full datasets not only incurs significant computational costs but also becomes a major limitation when integrating models developed and shared by others. To address this, we demonstrate that dataset reduction techniques, such as coreset selection and dataset condensation, effectively reduce the data requirement for model merging. In our experiments with SPLIT-CIFAR10 model merging, the accuracy is significantly improved by $31%$ when using the full dataset and $24%$ when using the sampled subset compared with not using the dataset.

Satoshi Suzuki, Shin'ya Yamaguchi, Shoichiro Takeda et al.

Contrastive pre-trained vision-language models, such as CLIP, demonstrate strong generalization abilities in zero-shot classification by leveraging embeddings extracted from image and text encoders. This paper aims to robustly fine-tune these vision-language models on in-distribution (ID) data without compromising their generalization abilities in out-of-distribution (OOD) and zero-shot settings. Current robust fine-tuning methods tackle this challenge by reusing contrastive learning, which was used in pre-training, for fine-tuning. However, we found that these methods distort the geometric structure of the embeddings, which plays a crucial role in the generalization of vision-language models, resulting in limited OOD and zero-shot performance. To address this, we propose Difference Vector Equalization (DiVE), which preserves the geometric structure during fine-tuning. The idea behind DiVE is to constrain difference vectors, each of which is obtained by subtracting the embeddings extracted from the pre-trained and fine-tuning models for the same data sample. By constraining the difference vectors to be equal across various data samples, we effectively preserve the geometric structure. Therefore, we introduce two losses: average vector loss (AVL) and pairwise vector loss (PVL). AVL preserves the geometric structure globally by constraining difference vectors to be equal to their weighted average. PVL preserves the geometric structure locally by ensuring a consistent multimodal alignment. Our experiments demonstrate that DiVE effectively preserves the geometric structure, achieving strong results across ID, OOD, and zero-shot metrics.

Sekitoshi Kanai, Yasutoshi Ida, Kazuki Adachi et al.

This study investigates a method to evaluate time-series datasets in terms of the performance of deep neural networks (DNNs) with state space models (deep SSMs) trained on the dataset. SSMs have attracted attention as components inside DNNs to address time-series data. Since deep SSMs have powerful representation capacities, training datasets play a crucial role in solving a new task. However, the effectiveness of training datasets cannot be known until deep SSMs are actually trained on them. This can increase the cost of data collection for new tasks, as a trial-and-error process of data collection and time-consuming training are needed to achieve the necessary performance. To advance the practical use of deep SSMs, the metric of datasets to estimate the performance early in the training can be one key element. To this end, we introduce the concept of data evaluation methods used in system identification. In system identification of linear dynamical systems, the effectiveness of datasets is evaluated by using the spectrum of input signals. We introduce this concept to deep SSMs, which are nonlinear dynamical systems. We propose the K-spectral metric, which is the sum of the top-K spectra of signals inside deep SSMs, by focusing on the fact that each layer of a deep SSM can be regarded as a linear dynamical system. Our experiments show that the K-spectral metric has a large absolute value of the correlation coefficient with the performance and can be used to evaluate the quality of training datasets.

Shohei Enomoto, Shin'ya Yamaguchi

In this paper, we address a fundamental gap between pre-training and fine-tuning of deep neural networks: while pre-training has shifted from unimodal to multimodal learning with enhanced visual understanding, fine-tuning predominantly remains unimodal, limiting the benefits of rich pre-trained representations. To bridge this gap, we propose a novel approach that transforms unimodal datasets into multimodal ones using Multimodal Large Language Models (MLLMs) to generate synthetic image captions for fine-tuning models with a multimodal objective. Our method employs carefully designed prompts incorporating class labels and domain context to produce high-quality captions tailored for classification tasks. Furthermore, we introduce a supervised contrastive loss function that explicitly encourages clustering of same-class representations during fine-tuning, along with a new inference technique that leverages class-averaged text embeddings from multiple synthetic captions per image. Extensive experiments across 13 image classification benchmarks demonstrate that our approach outperforms baseline methods, with particularly significant improvements in few-shot learning scenarios. Our work establishes a new paradigm for dataset enhancement that effectively bridges the gap between multimodal pre-training and fine-tuning. Our code is available at https://github.com/s-enmt/MMFT.

Shin'ya Yamaguchi, Takuma Fukuda

Synthetic samples from diffusion models are promising for leveraging in training discriminative models as replications of real training datasets. However, we found that the synthetic datasets degrade classification performance over real datasets even when using state-of-the-art diffusion models. This means that modern diffusion models do not perfectly represent the data distribution for the purpose of replicating datasets for training discriminative tasks. This paper investigates the gap between synthetic and real samples by analyzing the synthetic samples reconstructed from real samples through the diffusion and reverse process. By varying the time steps starting the reverse process in the reconstruction, we can control the trade-off between the information in the original real data and the information added by diffusion models. Through assessing the reconstructed samples and trained models, we found that the synthetic data are concentrated in modes of the training data distribution as the reverse step increases, and thus, they are difficult to cover the outer edges of the distribution. Our findings imply that modern diffusion models are insufficient to replicate training data distribution perfectly, and there is room for the improvement of generative modeling in the replication of training datasets.

Shin'ya Yamaguchi, Kosuke Nishida, Daiki Chijiwa et al.

Concept bottleneck models (CBMs) are inherently interpretable and intervenable neural network models, which explain their final label prediction by the intermediate prediction of high-level semantic concepts. However, they require target task training to learn input-to-concept and concept-to-label mappings, incurring target dataset collections and training resources. In this paper, we present \textit{zero-shot concept bottleneck models} (Z-CBMs), which predict concepts and labels in a fully zero-shot manner without training neural networks. Z-CBMs utilize a large-scale concept bank, which is composed of millions of vocabulary extracted from the web, to describe arbitrary input in various domains. For the input-to-concept mapping, we introduce concept retrieval, which dynamically finds input-related concepts by the cross-modal search on the concept bank. In the concept-to-label inference, we apply concept regression to select essential concepts from the retrieved concepts by sparse linear regression. Through extensive experiments, we confirm that our Z-CBMs provide interpretable and intervenable concepts without any additional training. Code will be available at https://github.com/yshinya6/zcbm.

Shin'ya Yamaguchi, Daiki Chijiwa, Tamao Sakao et al.

Large vision-language models (LVLMs) employ multi-modal in-context learning (MM-ICL) to adapt to new tasks by leveraging demonstration examples. While increasing the number of demonstrations boosts performance, they incur significant inference latency due to the quadratic computational cost of Transformer attention with respect to the context length. To address this trade-off, we propose Parallel In-Context Learning (Parallel-ICL), a plug-and-play inference algorithm. Parallel-ICL partitions the long demonstration context into multiple shorter, manageable chunks. It processes these chunks in parallel and integrates their predictions at the logit level, using a weighted Product-of-Experts (PoE) ensemble to approximate the full-context output. Guided by ensemble learning theory, we introduce principled strategies for Parallel-ICL: (i) clustering-based context chunking to maximize inter-chunk diversity and (ii) similarity-based context compilation to weight predictions by query relevance. Extensive experiments on VQA, image captioning, and classification benchmarks demonstrate that Parallel-ICL achieves performance comparable to full-context MM-ICL, while significantly improving inference speed. Our work offers an effective solution to the accuracy-efficiency trade-off in MM-ICL, enabling dynamic task adaptation with substantially reduced inference overhead.

Daiki Chijiwa, Shin'ya Yamaguchi, Yasutoshi Ida et al.

Pruning the weights of randomly initialized neural networks plays an important role in the context of lottery ticket hypothesis. Ramanujan et al. (2020) empirically showed that only pruning the weights can achieve remarkable performance instead of optimizing the weight values. However, to achieve the same level of performance as the weight optimization, the pruning approach requires more parameters in the networks before pruning and thus more memory space. To overcome this parameter inefficiency, we introduce a novel framework to prune randomly initialized neural networks with iteratively randomizing weight values (IteRand). Theoretically, we prove an approximation theorem in our framework, which indicates that the randomizing operations are provably effective to reduce the required number of the parameters. We also empirically demonstrate the parameter efficiency in multiple experiments on CIFAR-10 and ImageNet. The code is available at: https://github.com/dchiji-ntt/iterand

Shin'ya Yamaguchi

Semi-supervised learning (SSL) is a promising approach for training deep classification models using labeled and unlabeled datasets. However, existing SSL methods rely on a large unlabeled dataset, which may not always be available in many real-world applications due to legal constraints (e.g., GDPR). In this paper, we investigate the research question: Can we train SSL models without real unlabeled datasets? Instead of using real unlabeled datasets, we propose an SSL method using synthetic datasets generated from generative foundation models trained on datasets containing millions of samples in diverse domains (e.g., ImageNet). Our main concepts are identifying synthetic samples that emulate unlabeled samples from generative foundation models and training classifiers using these synthetic samples. To achieve this, our method is formulated as an alternating optimization problem: (i) meta-learning of generative foundation models and (ii) SSL of classifiers using real labeled and synthetic unlabeled samples. For (i), we propose a meta-learning objective that optimizes latent variables to generate samples that resemble real labeled samples and minimize the validation loss. For (ii), we propose a simple unsupervised loss function that regularizes the feature extractors of classifiers to maximize the performance improvement obtained from synthetic samples. We confirm that our method outperforms baselines using generative foundation models on SSL. We also demonstrate that our methods outperform SSL using real unlabeled datasets in scenarios with extremely small amounts of labeled datasets. This suggests that synthetic samples have the potential to provide improvement gains more efficiently than real unlabeled data.

Shin'ya Yamaguchi, Dewei Feng, Sekitoshi Kanai et al.

Contrastive language image pre-training (CLIP) is an essential component of building modern vision-language foundation models. While CLIP demonstrates remarkable zero-shot performance on downstream tasks, the multi-modal feature spaces still suffer from a modality gap, which is a gap between image and text feature clusters and limits downstream task performance. Although existing works attempt to address the modality gap by modifying pre-training or fine-tuning, they struggle with heavy training costs with large datasets or degradations of zero-shot performance. This paper presents CLIP-Refine, a post-pre-training method for CLIP models at a phase between pre-training and fine-tuning. CLIP-Refine aims to align the feature space with 1 epoch training on small image-text datasets without zero-shot performance degradations. To this end, we introduce two techniques: random feature alignment (RaFA) and hybrid contrastive-distillation (HyCD). RaFA aligns the image and text features to follow a shared prior distribution by minimizing the distance to random reference vectors sampled from the prior. HyCD updates the model with hybrid soft labels generated by combining ground-truth image-text pair labels and outputs from the pre-trained CLIP model. This contributes to achieving both maintaining the past knowledge and learning new knowledge to align features. Our extensive experiments with multiple classification and retrieval tasks show that CLIP-Refine succeeds in mitigating the modality gap and improving the zero-shot performance.

Shin'ya Yamaguchi, Sekitoshi Kanai, Kazuki Adachi et al.

While fine-tuning is a de facto standard method for training deep neural networks, it still suffers from overfitting when using small target datasets. Previous methods improve fine-tuning performance by maintaining knowledge of the source datasets or introducing regularization terms such as contrastive loss. However, these methods require auxiliary source information (e.g., source labels or datasets) or heavy additional computations. In this paper, we propose a simple method called adaptive random feature regularization (AdaRand). AdaRand helps the feature extractors of training models to adaptively change the distribution of feature vectors for downstream classification tasks without auxiliary source information and with reasonable computation costs. To this end, AdaRand minimizes the gap between feature vectors and random reference vectors that are sampled from class conditional Gaussian distributions. Furthermore, AdaRand dynamically updates the conditional distribution to follow the currently updated feature extractors and balance the distance between classes in feature spaces. Our experiments show that AdaRand outperforms the other fine-tuning regularization, which requires auxiliary source information and heavy computation costs.

Shohei Enomoto, Naoya Hasegawa, Kazuki Adachi et al.

Deep neural networks have achieved remarkable success in a variety of computer vision applications. However, there is a problem of degrading accuracy when the data distribution shifts between training and testing. As a solution of this problem, Test-time Adaptation~(TTA) has been well studied because of its practicality. Although TTA methods increase accuracy under distribution shift by updating the model at test time, using high-uncertainty predictions is known to degrade accuracy. Since the input image is the root of the distribution shift, we incorporate a new perspective on enhancing the input image into TTA methods to reduce the prediction's uncertainty. We hypothesize that enhancing the input image reduces prediction's uncertainty and increase the accuracy of TTA methods. On the basis of our hypothesis, we propose a novel method: Test-time Enhancer and Classifier Adaptation~(TECA). In TECA, the classification model is combined with the image enhancement model that transforms input images into recognition-friendly ones, and these models are updated by existing TTA methods. Furthermore, we found that the prediction from the enhanced image does not always have lower uncertainty than the prediction from the original image. Thus, we propose logit switching, which compares the uncertainty measure of these predictions and outputs the lower one. In our experiments, we evaluate TECA with various TTA methods and show that TECA reduces prediction's uncertainty and increases accuracy of TTA methods despite having no hyperparameters and little parameter overhead.

Kazuki Adachi, Shohei Enomoto, Taku Sasaki et al.

Person re-identification (re-id), which aims to retrieve images of the same person in a given image from a database, is one of the most practical image recognition applications. In the real world, however, the environments that the images are taken from change over time. This causes a distribution shift between training and testing and degrades the performance of re-id. To maintain re-id performance, models should continue adapting to the test environment's temporal changes. Test-time adaptation (TTA), which aims to adapt models to the test environment with only unlabeled test data, is a promising way to handle this problem because TTA can adapt models instantly in the test environment. However, the previous TTA methods are designed for classification and cannot be directly applied to re-id. This is because the set of people's identities in the dataset differs between training and testing in re-id, whereas the set of classes is fixed in the current TTA methods designed for classification. To improve re-id performance in changing test environments, we propose TEst-time similarity Modification for Person re-identification (TEMP), a novel TTA method for re-id. TEMP is the first fully TTA method for re-id, which does not require any modification to pre-training. Inspired by TTA methods that refine the prediction uncertainty in classification, we aim to refine the uncertainty in re-id. However, the uncertainty cannot be computed in the same way as classification in re-id since it is an open-set task, which does not share person labels between training and testing. Hence, we propose re-id entropy, an alternative uncertainty measure for re-id computed based on the similarity between the feature vectors. Experiments show that the re-id entropy can measure the uncertainty on re-id and TEMP improves the performance of re-id in online settings where the distribution changes over time.

Kazuki Adachi, Shin'ya Yamaguchi, Tomoki Hamagami

Pre-trained vision-language models such as contrastive language-image pre-training (CLIP) have demonstrated a remarkable generalizability, which has enabled a wide range of applications represented by zero-shot classification. However, vision-language models still suffer when they face datasets with large gaps from training ones, i.e., distribution shifts. We found that CLIP is especially vulnerable to sensor degradation, a type of realistic distribution shift caused by sensor conditions such as weather, light, or noise. Collecting a new dataset from a test distribution for fine-tuning highly costs since sensor degradation occurs unexpectedly and has a range of variety. Thus, we investigate test-time adaptation (TTA) of zero-shot classification, which enables on-the-fly adaptation to the test distribution with unlabeled test data. Existing TTA methods for CLIP mainly focus on modifying image and text embeddings or predictions to address distribution shifts. Although these methods can adapt to domain shifts, such as fine-grained labels spaces or different renditions in input images, they fail to adapt to distribution shifts caused by sensor degradation. We found that this is because image embeddings are "corrupted" in terms of uniformity, a measure related to the amount of information. To make models robust to sensor degradation, we propose a novel method called uniformity-aware information-balanced TTA (UnInfo). To address the corruption of image embeddings, we introduce uniformity-aware confidence maximization, information-aware loss balancing, and knowledge distillation from the exponential moving average (EMA) teacher. Through experiments, we demonstrate that our UnInfo improves accuracy under sensor degradation by retaining information in terms of uniformity.

Daiki Chijiwa, Taku Hasegawa, Kyosuke Nishida et al.

Tokenization -- the process of decomposing a given text into a sequence of subwords called tokens -- is one of the key components in the development of language models. Particularly, auto-regressive language models generate texts token by token, i.e., by predicting the next-token distribution given the previous ones, and thus tokenization directly affects their efficiency in text generation. Since each language model has their own vocabulary as a set of possible tokens, they struggle to cooperate with each other at the level of next-token distributions such as model ensemble. In this paper, we establish a theoretical framework of lossless vocabulary reduction, which efficiently converts a given auto-regressive language model into the one with an arbitrarily small vocabulary without any loss in accuracy. As an application, we demonstrate that language models with different tokenization can cooperate with each other efficiently through their maximal common vocabulary.

Shin'ya Yamaguchi, Kosuke Nishida, Daiki Chijiwa

Large vision-language models (LVLMs) have demonstrated remarkable capabilities by integrating pre-trained vision encoders with large language models (LLMs). Similar to single-modal LLMs, chain-of-thought (CoT) prompting has been adapted for LVLMs to enhance multi-modal reasoning by generating intermediate rationales based on visual and textual inputs. While CoT is assumed to improve grounding and accuracy in LVLMs, our experiments reveal a key challenge: existing LVLMs often ignore the contents of generated rationales in CoT reasoning. To address this, we re-formulate multi-modal CoT reasoning as a KL-constrained reward maximization focused on rationale-conditional log-likelihood. As the optimal solution, we propose rationale-enhanced decoding (RED), a novel plug-and-play inference-time decoding strategy. RED harmonizes visual and rationale information by multiplying distinct image-conditional and rationale-conditional next token distributions. Extensive experiments show that RED consistently and significantly improves reasoning over standard CoT and other decoding methods across multiple benchmarks and LVLMs. Our work offers a practical and effective approach to improve both the faithfulness and accuracy of CoT reasoning in LVLMs, paving the way for more reliable rationale-grounded multi-modal systems.

Kazuki Adachi, Shin'ya Yamaguchi, Atsutoshi Kumagai

Real-world image recognition systems often face corrupted input images, which cause distribution shifts and degrade the performance of models. These systems often use a single prediction model in a central server and process images sent from various environments, such as cameras distributed in cities or cars. Such single models face images corrupted in heterogeneous ways in test time. Thus, they require to instantly adapt to the multiple corruptions during testing rather than being re-trained at a high cost. Test-time adaptation (TTA), which aims to adapt models without accessing the training dataset, is one of the settings that can address this problem. Existing TTA methods indeed work well on a single corruption. However, the adaptation ability is limited when multiple types of corruption occur, which is more realistic. We hypothesize this is because the distribution shift is more complicated, and the adaptation becomes more difficult in case of multiple corruptions. In fact, we experimentally found that a larger distribution gap remains after TTA. To address the distribution gap during testing, we propose a novel TTA method named Covariance-Aware Feature alignment (CAFe). We empirically show that CAFe outperforms prior TTA methods on image corruptions, including multiple types of corruptions.

Kazuki Adachi, Shin'ya Yamaguchi

Existing image recognition techniques based on convolutional neural networks (CNNs) basically assume that the training and test datasets are sampled from i.i.d distributions. However, this assumption is easily broken in the real world because of the distribution shift that occurs when the co-occurrence relations between objects and backgrounds in input images change. Under this type of distribution shift, CNNs learn to focus on features that are not task-relevant, such as backgrounds from the training data, and degrade their accuracy on the test data. To tackle this problem, we propose relevant feature focusing (ReFF). ReFF detects task-relevant features and regularizes CNNs via explanation outputs (e.g., Grad-CAM). Since ReFF is composed of post-hoc explanation modules, it can be easily applied to off-the-shelf CNNs. Furthermore, ReFF requires no additional inference cost at test time because it is only used for regularization while training. We demonstrate that CNNs trained with ReFF focus on features relevant to the target task and that ReFF improves the test-time accuracy.

Shin'ya Yamaguchi, Sekitoshi Kanai

Generative adversarial networks built from deep convolutional neural networks (GANs) lack the ability to exactly replicate the high-frequency components of natural images. To alleviate this issue, we introduce two novel training techniques called frequency dropping (F-Drop) and frequency matching (F-Match). The key idea of F-Drop is to filter out unnecessary high-frequency components from the input images of the discriminators. This simple modification prevents the discriminators from being confused by perturbations of the high-frequency components. In addition, F-Drop makes the GANs focus on fitting in the low-frequency domain, in which there are the dominant components of natural images. F-Match minimizes the difference between real and fake images in the frequency domain for generating more realistic images. F-Match is implemented as a regularization term in the objective functions of the generators; it penalizes the batch mean error in the frequency domain. F-Match helps the generators to fit in the high-frequency domain filtered out by F-Drop to the real image. We experimentally demonstrate that the combination of F-Drop and F-Match improves the generative performance of GANs in both the frequency and spatial domain on multiple image benchmarks.

Sekitoshi Kanai, Masanori Yamada, Shin'ya Yamaguchi et al.

We propose a method for improving adversarial robustness by addition of a new bounded function just before softmax. Recent studies hypothesize that small logits (inputs of softmax) by logit regularization can improve adversarial robustness of deep learning. Following this hypothesis, we analyze norms of logit vectors at the optimal point under the assumption of universal approximation and explore new methods for constraining logits by addition of a bounded function before softmax. We theoretically and empirically reveal that small logits by addition of a common activation function, e.g., hyperbolic tangent, do not improve adversarial robustness since input vectors of the function (pre-logit vectors) can have large norms. From the theoretical findings, we develop the new bounded function. The addition of our function improves adversarial robustness because it makes logit and pre-logit vectors have small norms. Since our method only adds one activation function before softmax, it is easy to combine our method with adversarial training. Our experiments demonstrate that our method is comparable to logit regularization methods in terms of accuracies on adversarially perturbed datasets without adversarial training. Furthermore, it is superior or comparable to logit regularization methods and a recent defense method (TRADES) when using adversarial training.

Shin'ya Yamaguchi, Sekitoshi Kanai, Tetsuya Shioda et al.

The rotation prediction (Rotation) is a simple pretext-task for self-supervised learning (SSL), where models learn useful representations for target vision tasks by solving pretext-tasks. Although Rotation captures information of object shapes, it hardly captures information of textures. To tackle this problem, we introduce a novel pretext-task called image enhanced rotation prediction (IE-Rot) for SSL. IE-Rot simultaneously solves Rotation and another pretext-task based on image enhancement (e.g., sharpening and solarizing) while maintaining simplicity. Through the simultaneous prediction of rotation and image enhancement, models learn representations to capture the information of not only object shapes but also textures. Our experimental results show that IE-Rot models outperform Rotation on various standard benchmarks including ImageNet classification, PASCAL-VOC detection, and COCO detection/segmentation.

Shin'ya Yamaguchi, Sekitoshi Kanai, Takeharu Eda

For deep learning applications, the massive data development (e.g., collecting, labeling), which is an essential process in building practical applications, still incurs seriously high costs. In this work, we propose an effective data augmentation method based on generative adversarial networks (GANs), called Domain Fusion. Our key idea is to import the knowledge contained in an outer dataset to a target model by using a multi-domain learning GAN. The multi-domain learning GAN simultaneously learns the outer and target dataset and generates new samples for the target tasks. The simultaneous learning process makes GANs generate the target samples with high fidelity and variety. As a result, we can obtain accurate models for the target tasks by using these generated samples even if we only have an extremely low volume target dataset. We experimentally evaluate the advantages of Domain Fusion in image classification tasks on 3 target datasets: CIFAR-100, FGVC-Aircraft, and Indoor Scene Recognition. When trained on each target dataset reduced the samples to 5,000 images, Domain Fusion achieves better classification accuracy than the data augmentation using fine-tuned GANs. Furthermore, we show that Domain Fusion improves the quality of generated samples, and the improvements can contribute to higher accuracy.