Atsutoshi Kumagai

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.

Tomoharu Iwata, Atsutoshi Kumagai

We propose a meta-learning method for semi-supervised learning that learns from multiple tasks with heterogeneous attribute spaces. The existing semi-supervised meta-learning methods assume that all tasks share the same attribute space, which prevents us from learning with a wide variety of tasks. With the proposed method, the expected test performance on tasks with a small amount of labeled data is improved with unlabeled data as well as data in various tasks, where the attribute spaces are different among tasks. The proposed method embeds labeled and unlabeled data simultaneously in a task-specific space using a neural network, and the unlabeled data's labels are estimated by adapting classification or regression models in the embedding space. For the neural network, we develop variable-feature self-attention layers, which enable us to find embeddings of data with different attribute spaces with a single neural network by considering interactions among examples, attributes, and labels. Our experiments on classification and regression datasets with heterogeneous attribute spaces demonstrate that our proposed method outperforms the existing meta-learning and semi-supervised learning methods.

Tomoharu Iwata, Atsutoshi Kumagai

Many neural network-based out-of-distribution (OoD) detection methods have been proposed. However, they require many training data for each target task. We propose a simple yet effective meta-learning method to detect OoD with small in-distribution data in a target task. With the proposed method, the OoD detection is performed by density estimation in a latent space. A neural network shared among all tasks is used to flexibly map instances in the original space to the latent space. The neural network is meta-learned such that the expected OoD detection performance is improved by using various tasks that are different from the target tasks. This meta-learning procedure enables us to obtain appropriate representations in the latent space for OoD detection. For density estimation, we use a Gaussian mixture model (GMM) with full covariance for each class. We can adapt the GMM parameters to in-distribution data in each task in a closed form by maximizing the likelihood. Since the closed form solution is differentiable, we can meta-learn the neural network efficiently with a stochastic gradient descent method by incorporating the solution into the meta-learning objective function. In experiments using six datasets, we demonstrate that the proposed method achieves better performance than existing meta-learning and OoD detection methods.

Yasutoshi Ida, Sekitoshi Kanai, Kazuki Adachi et al.

Regularized discrete optimal transport (OT) is a powerful tool to measure the distance between two discrete distributions that have been constructed from data samples on two different domains. While it has a wide range of applications in machine learning, in some cases the sampled data from only one of the domains will have class labels such as unsupervised domain adaptation. In this kind of problem setting, a group-sparse regularizer is frequently leveraged as a regularization term to handle class labels. In particular, it can preserve the label structure on the data samples by corresponding the data samples with the same class label to one group-sparse regularization term. As a result, we can measure the distance while utilizing label information by solving the regularized optimization problem with gradient-based algorithms. However, the gradient computation is expensive when the number of classes or data samples is large because the number of regularization terms and their respective sizes also turn out to be large. This paper proposes fast discrete OT with group-sparse regularizers. Our method is based on two ideas. The first is to safely skip the computations of the gradients that must be zero. The second is to efficiently extract the gradients that are expected to be nonzero. Our method is guaranteed to return the same value of the objective function as that of the original method. Experiments show that our method is up to 8.6 times faster than the original method without degrading accuracy.

Akira Ito, Masanori Yamada, Atsutoshi Kumagai

Recently, Ainsworth et al. showed that using weight matching (WM) to minimize the $L^2$ distance in a permutation search of model parameters effectively identifies permutations that satisfy linear mode connectivity (LMC), where the loss along a linear path between two independently trained models with different seeds remains nearly constant. This paper analyzes LMC using WM, which is useful for understanding stochastic gradient descent's effectiveness and its application in areas like model merging. We first empirically show that permutations found by WM do not significantly reduce the $L^2$ distance between two models, and the occurrence of LMC is not merely due to distance reduction by WM itself. We then demonstrate that permutations can change the directions of the singular vectors, but not the singular values, of the weight matrices in each layer. This finding shows that permutations found by WM primarily align the directions of singular vectors associated with large singular values across models. This alignment brings the singular vectors with large singular values, which determine the model's functionality, closer between the original and merged models, allowing the merged model to retain functionality similar to the original models, thereby satisfying LMC. This paper also analyzes activation matching (AM) in terms of singular vectors and finds that the principle of AM is likely the same as that of WM. Finally, we analyze the difference between WM and the straight-through estimator (STE), a dataset-dependent permutation search method, and show that WM can be more advantageous than STE in achieving LMC among three or more models.

Tomoharu Iwata, Atsutoshi Kumagai

Although Gaussian processes (GPs) with deep kernels have been successfully used for meta-learning in regression tasks, its uncertainty estimation performance can be poor. We propose a meta-learning method for calibrating deep kernel GPs for improving regression uncertainty estimation performance with a limited number of training data. The proposed method meta-learns how to calibrate uncertainty using data from various tasks by minimizing the test expected calibration error, and uses the knowledge for unseen tasks. We design our model such that the adaptation and calibration for each task can be performed without iterative procedures, which enables effective meta-learning. In particular, a task-specific uncalibrated output distribution is modeled by a GP with a task-shared encoder network, and it is transformed to a calibrated one using a cumulative density function of a task-specific Gaussian mixture model (GMM). By integrating the GP and GMM into our neural network-based model, we can meta-learn model parameters in an end-to-end fashion. Our experiments demonstrate that the proposed method improves uncertainty estimation performance while keeping high regression performance compared with the existing methods using real-world datasets in few-shot settings.

Hiroshi Takahashi, Tomoharu Iwata, Atsutoshi Kumagai et al.

Diffusion models are powerful generative models but often generate sensitive data that are unwanted by users, mainly because the unlabeled training data frequently contain such sensitive data. Since labeling all sensitive data in the large-scale unlabeled training data is impractical, we address this problem by using a small amount of labeled sensitive data. In this paper, we propose positive-unlabeled diffusion models, which prevent the generation of sensitive data using unlabeled and sensitive data. Our approach can approximate the evidence lower bound (ELBO) for normal (negative) data using only unlabeled and sensitive (positive) data. Therefore, even without labeled normal data, we can maximize the ELBO for normal data and minimize it for labeled sensitive data, ensuring the generation of only normal data. Through experiments across various datasets and settings, we demonstrated that our approach can prevent the generation of sensitive images without compromising image quality.

Hiroshi Takahashi, Tomoharu Iwata, Atsutoshi Kumagai et al.

Aligning language models with human preferences is essential for ensuring their safety and reliability. Although most existing approaches assume specific human preference models such as the Bradley-Terry model, this assumption may fail to accurately capture true human preferences, and consequently, these methods lack statistical consistency, i.e., the guarantee that language models converge to the true human preference as the number of samples increases. In contrast, direct density ratio optimization (DDRO) achieves statistical consistency without assuming any human preference models. DDRO models the density ratio between preferred and non-preferred data distributions using the language model, and then optimizes it via density ratio estimation. However, this density ratio is unstable and often diverges, leading to training instability of DDRO. In this paper, we propose a novel alignment method that is both stable and statistically consistent. Our approach is based on the relative density ratio between the preferred data distribution and a mixture of the preferred and non-preferred data distributions. Our approach is stable since this relative density ratio is bounded above and does not diverge. Moreover, it is statistically consistent and yields significantly tighter convergence guarantees than DDRO. We experimentally show its effectiveness with Qwen 2.5 and Llama 3.

Akira Ito, Masanori Yamada, Daiki Chijiwa et al.

Recently, Ainsworth et al. empirically demonstrated that, given two independently trained models, applying a parameter permutation that preserves the input-output behavior allows the two models to be connected by a low-loss linear path. When such a path exists, the models are said to achieve linear mode connectivity (LMC). Prior studies, including Ainsworth et al., have reported that achieving LMC requires not only an appropriate permutation search but also sufficiently wide models (e.g., a 32 $\times$ width multiplier for ResNet-20). This is broadly believed to be because increasing the model width ensures a large enough space of candidate permutations, increasing the chance of finding one that yields LMC. In this work, we empirically demonstrate that, even without any permutations, simply widening the models is sufficient for achieving LMC when using a suitable softmax temperature calibration. We further explain why this phenomenon arises by analyzing intermediate layer outputs. Specifically, we introduce layerwise exponentially weighted connectivity (LEWC), which states that the output of each layer of the merged model can be represented as an exponentially weighted sum of the outputs of the corresponding layers of the original models. Consequently the merged model's output matches that of an ensemble of the original models, which facilitates LMC. To the best of our knowledge, this work is the first to show that widening the model not only facilitates nonlinear mode connectivity, as suggested in prior research, but also significantly increases the possibility of achieving linear mode connectivity.

Atsutoshi Kumagai, Tomoharu Iwata, Taishi Nishiyama et al.

We propose a meta-learning method for learning from multiple noisy annotators. In many applications such as crowdsourcing services, labels for supervised learning are given by multiple annotators. Since the annotators have different skills or biases, given labels can be noisy. To learn accurate classifiers, existing methods require many noisy annotated data. However, sufficient data might be unavailable in practice. To overcome the lack of data, the proposed method uses labeled data obtained in different but related tasks. The proposed method embeds each example in tasks to a latent space by using a neural network and constructs a probabilistic model for learning a task-specific classifier while estimating annotators' abilities on the latent space. This neural network is meta-learned to improve the expected test classification performance when the classifier is adapted to a given small amount of annotated data. This classifier adaptation is performed by maximizing the posterior probability via the expectation-maximization (EM) algorithm. Since each step in the EM algorithm is easily computed as a closed-form and is differentiable, the proposed method can efficiently backpropagate the loss through the EM algorithm to meta-learn the neural network. We show the effectiveness of our method with real-world datasets with synthetic noise and real-world crowdsourcing datasets.

Atsutoshi Kumagai, Tomoharu Iwata, Yasuhiro Fujiwara

We propose a meta-learning method for positive and unlabeled (PU) classification, which improves the performance of binary classifiers obtained from only PU data in unseen target tasks. PU learning is an important problem since PU data naturally arise in real-world applications such as outlier detection and information retrieval. Existing PU learning methods require many PU data, but sufficient data are often unavailable in practice. The proposed method minimizes the test classification risk after the model is adapted to PU data by using related tasks that consist of positive, negative, and unlabeled data. We formulate the adaptation as an estimation problem of the Bayes optimal classifier, which is an optimal classifier to minimize the classification risk. The proposed method embeds each instance into a task-specific space using neural networks. With the embedded PU data, the Bayes optimal classifier is estimated through density-ratio estimation of PU densities, whose solution is obtained as a closed-form solution. The closed-form solution enables us to efficiently and effectively minimize the test classification risk. We empirically show that the proposed method outperforms existing methods with one synthetic and three real-world datasets.

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.

Kentaro Ohno, Atsutoshi Kumagai

Recurrent neural networks with a gating mechanism such as an LSTM or GRU are powerful tools to model sequential data. In the mechanism, a forget gate, which was introduced to control information flow in a hidden state in the RNN, has recently been re-interpreted as a representative of the time scale of the state, i.e., a measure how long the RNN retains information on inputs. On the basis of this interpretation, several parameter initialization methods to exploit prior knowledge on temporal dependencies in data have been proposed to improve learnability. However, the interpretation relies on various unrealistic assumptions, such as that there are no inputs after a certain time point. In this work, we reconsider this interpretation of the forget gate in a more realistic setting. We first generalize the existing theory on gated RNNs so that we can consider the case where inputs are successively given. We then argue that the interpretation of a forget gate as a temporal representation is valid when the gradient of loss with respect to the state decreases exponentially as time goes back. We empirically demonstrate that existing RNNs satisfy this gradient condition at the initial training phase on several tasks, which is in good agreement with previous initialization methods. On the basis of this finding, we propose an approach to construct new RNNs that can represent a longer time scale than conventional models, which will improve the learnability for long-term sequential data. We verify the effectiveness of our method by experiments with real-world datasets.

Atsutoshi Kumagai, Tomoharu Iwata, Yasuhiro Fujiwara

We propose a few-shot learning method for unsupervised feature selection, which is a task to select a subset of relevant features in unlabeled data. Existing methods usually require many instances for feature selection. However, sufficient instances are often unavailable in practice. The proposed method can select a subset of relevant features in a target task given a few unlabeled target instances by training with unlabeled instances in multiple source tasks. Our model consists of a feature selector and decoder. The feature selector outputs a subset of relevant features taking a few unlabeled instances as input such that the decoder can reconstruct the original features of unseen instances from the selected ones. The feature selector uses the Concrete random variables to select features via gradient descent. To encode task-specific properties from a few unlabeled instances to the model, the Concrete random variables and decoder are modeled using permutation-invariant neural networks that take a few unlabeled instances as input. Our model is trained by minimizing the expected test reconstruction error given a few unlabeled instances that is calculated with datasets in source tasks. We experimentally demonstrate that the proposed method outperforms existing feature selection methods.

Atsutoshi Kumagai, Tomoharu Iwata, Yasuhiro Fujiwara

The ratio of two probability densities, called a density-ratio, is a vital quantity in machine learning. In particular, a relative density-ratio, which is a bounded extension of the density-ratio, has received much attention due to its stability and has been used in various applications such as outlier detection and dataset comparison. Existing methods for (relative) density-ratio estimation (DRE) require many instances from both densities. However, sufficient instances are often unavailable in practice. In this paper, we propose a meta-learning method for relative DRE, which estimates the relative density-ratio from a few instances by using knowledge in related datasets. Specifically, given two datasets that consist of a few instances, our model extracts the datasets' information by using neural networks and uses it to obtain instance embeddings appropriate for the relative DRE. We model the relative density-ratio by a linear model on the embedded space, whose global optimum solution can be obtained as a closed-form solution. The closed-form solution enables fast and effective adaptation to a few instances, and its differentiability enables us to train our model such that the expected test error for relative DRE can be explicitly minimized after adapting to a few instances. We empirically demonstrate the effectiveness of the proposed method by using three problems: relative DRE, dataset comparison, and outlier detection.

Tomoharu Iwata, Atsutoshi Kumagai

Neural network-based anomaly detection methods have shown to achieve high performance. However, they require a large amount of training data for each task. We propose a neural network-based meta-learning method for supervised anomaly detection. The proposed method improves the anomaly detection performance on unseen tasks, which contains a few labeled normal and anomalous instances, by meta-training with various datasets. With a meta-learning framework, quick adaptation to each task and its effective backpropagation are important since the model is trained by the adaptation for each epoch. Our model enables them by formulating adaptation as a generalized eigenvalue problem with one-class classification; its global optimum solution is obtained, and the solver is differentiable. We experimentally demonstrate that the proposed method achieves better performance than existing anomaly detection and few-shot learning methods on various datasets.

Masanori Yamada, Sekitoshi Kanai, Tomoharu Iwata et al.

Adversarial training is actively studied for learning robust models against adversarial examples. A recent study finds that adversarially trained models degenerate generalization performance on adversarial examples when their weight loss landscape, which is loss changes with respect to weights, is sharp. Unfortunately, it has been experimentally shown that adversarial training sharpens the weight loss landscape, but this phenomenon has not been theoretically clarified. Therefore, we theoretically analyze this phenomenon in this paper. As a first step, this paper proves that adversarial training with the L2 norm constraints sharpens the weight loss landscape in the linear logistic regression model. Our analysis reveals that the sharpness of the weight loss landscape is caused by the noise aligned in the direction of increasing the loss, which is used in adversarial training. We theoretically and experimentally confirm that the weight loss landscape becomes sharper as the magnitude of the noise of adversarial training increases in the linear logistic regression model. Moreover, we experimentally confirm the same phenomena in ResNet18 with softmax as a more general case.

Tomoharu Iwata, Atsutoshi Kumagai

Time-series forecasting is important for many applications. Forecasting models are usually trained using time-series data in a specific target task. However, sufficient data in the target task might be unavailable, which leads to performance degradation. In this paper, we propose a few-shot learning method that forecasts a future value of a time-series in a target task given a few time-series in the target task. Our model is trained using time-series data in multiple training tasks that are different from target tasks. Our model uses a few time-series to build a forecasting function based on a recurrent neural network with an attention mechanism. With the attention mechanism, we can retrieve useful patterns in a small number of time-series for the current situation. Our model is trained by minimizing an expected test error of forecasting next timestep values. We demonstrate the effectiveness of the proposed method using 90 time-series datasets.

Atsutoshi Kumagai, Tomoharu Iwata, Yasuhiro Fujiwara

We propose a simple yet effective method for detecting anomalous instances on an attribute graph with label information of a small number of instances. Although with standard anomaly detection methods it is usually assumed that instances are independent and identically distributed, in many real-world applications, instances are often explicitly connected with each other, resulting in so-called attributed graphs. The proposed method embeds nodes (instances) on the attributed graph in the latent space by taking into account their attributes as well as the graph structure based on graph convolutional networks (GCNs). To learn node embeddings specialized for anomaly detection, in which there is a class imbalance due to the rarity of anomalies, the parameters of a GCN are trained to minimize the volume of a hypersphere that encloses the node embeddings of normal instances while embedding anomalous ones outside the hypersphere. This enables us to detect anomalies by simply calculating the distances between the node embeddings and hypersphere center. The proposed method can effectively propagate label information on a small amount of nodes to unlabeled ones by taking into account the node's attributes, graph structure, and class imbalance. In experiments with five real-world attributed graph datasets, we demonstrate that the proposed method achieves better performance than various existing anomaly detection methods.

Atsutoshi Kumagai, Tomoharu Iwata

We propose a method to infer domain-specific models such as classifiers for unseen domains, from which no data are given in the training phase, without domain semantic descriptors. When training and test distributions are different, standard supervised learning methods perform poorly. Zero-shot domain adaptation attempts to alleviate this problem by inferring models that generalize well to unseen domains by using training data in multiple source domains. Existing methods use observed semantic descriptors characterizing domains such as time information to infer the domain-specific models for the unseen domains. However, it cannot always be assumed that such metadata can be used in real-world applications. The proposed method can infer appropriate domain-specific models without any semantic descriptors by introducing the concept of latent domain vectors, which are latent representations for the domains and are used for inferring the models. The latent domain vector for the unseen domain is inferred from the set of the feature vectors in the corresponding domain, which is given in the testing phase. The domain-specific models consist of two components: the first is for extracting a representation of a feature vector to be predicted, and the second is for inferring model parameters given the latent domain vector. The posterior distributions of the latent domain vectors and the domain-specific models are parametrized by neural networks, and are optimized by maximizing the variational lower bound using stochastic gradient descent. The effectiveness of the proposed method was demonstrated through experiments using one regression and two classification tasks.