A low-pass filter is essential for stabilizing strictly proper repetitive-control systems, but it inevitably degrades steady-state tracking accuracy due to gain attenuation and phase lag. This article presents a new filter design and optimization method that improves both transient response and steady-state accuracy in continuous-time repetitive-control systems. First, the gain and phase characteristics of conventional filter-based repetitive controllers are rigorously analyzed to reveal the relationship between filter parameters and tracking performance. Based on this analysis, a new filter structure is designed to precisely compensate for gain attenuation and phase delay, specifically at the fundamental frequency, by minimizing the error term without increasing the filter bandwidth. A guideline for selecting the filter parameters for varying periodic trajectories is also provided. In addition, according to the one-to-one mapping between control and learning behaviors and their respective gains, dual performance indices are constructed to account for tracking error and control effort across multiple learning cycles. A multiobjective optimization framework is then developed to directly tune these gains subject to stability constraints, achieving an optimal balance between rapid transient convergence and control energy efficiency. Experimental results validate the effectiveness and superiority of the design.

This article presents a two-degree-of-freedom structure of a fuzzy repetitive-control system that accurately tracks periodic signals and effectively suppresses aperiodic disturbances based on an enhanced equivalent-input-disturbance (EEID) method. To make full use of the learning characteristic, a 2-D model of the fuzzy repetitive-control system is established to independently regulate the control and learning behaviors. The EEID approach decouples the design of the observer and the estimator by adding a flexible controller to the estimator. This makes it possible to simultaneously reduce the estimated and filtered errors, which is a contradictory problem in the conventional EID method. The two errors cannot be eliminated due to the lack of an internal model of the external disturbance. To actively cancel the negative influence of the two errors, a new observation-error compensator is added to the observer input to improve the disturbance-suppression performance. Two low-conservative linear-matrix-inequality stability conditions are derived using the Lyapunov–Krasovskii functionals and zero equations with free-weighting matrices. The nondominated sorting genetic algorithm II and the particle swarm optimization algorithm are used to optimize the parameters of the controllers for the tracking and disturbance-suppression systems based on the stability conditions and performance indexes. The effectiveness and superiority of the design are verified by numerical simulation and comparison results.

In practice, repetitive control (RC) is a type of learning control that exhibits good tracking performance. However, existing nonlinear RC methods lack analysis and design of the learning property, which results in limited performance. This study addresses the learning-enhanced issue. First, a new fuzzy Lyapunov candidate is constructed for stability analysis, which contains an integral term associated with the membership function and a double integral term. The design of the learning-dependent term integrates the nonlinear membership function information, which enhances the learning ability. Second, an additional first-order low-pass filter is incorporated into the conventional equivalent input disturbance estimator. The new filter acts as an integrator that adjusts the bandwidth of the disturbance compensation and gradually eliminates the disturbances in the output error. Third, a recursive optimization algorithm is used to design the controllers. Experimental comparisons on motor drive systems demonstrate the effectiveness and superiority of the method.

Focal cortical dysplasia (FCD) is a significant cause of intractable epilepsy, with its identification and segmentation being crucial for diagnosis. While fully-supervised learning models can detect and segment FCD, they are limited by the need for extensive labeled data, which is time-consuming and requires expert annotation. To address this, we introduce a semi-supervised model for FCD segmentation, leveraging a small set of labeled data alongside unlabeled data. Our method estimates uncertainty by measuring voxel-wise deviations between predictions and ground truth, using both ground truth and pseudo-labels to refine segmentation masks. For unlabeled data, multi-scale predictions capture different frequency components, allowing for selective consistency regularization based on region reliability. This approach is the first application of semi-supervised learning in epilepsy segmentation and demonstrates superior performance over state-of-the-art methods in experiments on clinical and public datasets.

Medical imaging is now a widely used test for the preoperative evaluation of focal cortical dysplasia (FCD). Deep learning-based methods can learn lesion features from image data to automatically recognize and segment FCD in epilepsy treatment. However, the existing FCD segmentation networks lack the ability to fully extract the FCD lesion information and automatically aggregate the salient features of the lesions, the segmentation accuracy needs to be improved. To this end, we propose an end-to-end 3D Convolutional Neural Network segmentation model, Multi Attention Two-Path Residual UNet (MATPR-UNet). Specifically, we propose two modules: (1) Two-Path Residual Attention module, which can extract both local and global information, and suppress invalid information by fully fusing the features in channel and space; (2) Spatially Gated Attention module, which enables the model to automatically focus on the FCD lesion region, highlighting its salient features. We combine the proposed two modules with the 3D UNet to construct the MATPR-UNet. Extensive experiments on the private FCD dataset and the public EPISURG dataset demonstrate that our method outperforms other state-of-the-art methods, and is robust.

This study presents a two-dimensional (2-D) repetitive control method to address the issues of periodic tracking and disturbance suppression in uncertain Takagi–Sugeno systems. The disturbance and uncertainty are treated as an equivalent-input-disturbance (EID). However, the conventional EID estimators typically suppress the EID through high gain. Meanwhile, the low-pass filter associated with EID causes a certain degree of phase lag. A proportional–integral (PI) filter is integrated with an EID estimator to develop a PI-EID structure to improve the estimation accuracy. Based on the self-learning mechanism of repetitive control, the 2-D repetitive controller is used to achieve a high level of tracking. Unlike the conventional nonlinear repetitive control methods, the state observer and the PI-EID estimator are membership function dependent. The gains of both controllers switch in line with the signs of the time derivative of the normalized premise variables, and this framework takes full account of the information of the nonlinear membership functions. The controller design procedures and the stability conditions are detailedly presented. Finally, a rotation speed control experiment is conducted to validate the developed PI-EID method.

The success of supervised deep learning heavily depends on large labeled datasets whose construction is often challenging in medical image analysis. Contrastive learning, a variant of self-supervised learning, is a potential solution to alleviate the strong demand for data annotation. In this work, we extend the contrastive learning framework to 3D volumetric medical imaging. Specifically, we propose (1) multiview contrasting strategy to maximize the mutual information between three views of 3D image to learn global representations and (2) long-short spatial contrasting strategy to learn local representations by matching a short spatial clip to a long spatial clip in the latent space. To combine these two strategies, we propose multiview long-short spatial contrastive learning (MLSSCL) framework, which can effectively learn generic 3D representations. Our extensive experiments on two brain Magnetic Resonance Imaging (MRI) datasets demonstrate that MLSSCL significantly outperforms learning from scratch and other self-supervised learning methods on both classification and segmentation tasks.

Existing few-shot learning (FSL) methods assume that there exist sufficient training samples from source classes for knowledge transfer to target classes with few training samples. However, this assumption is often invalid, especially when it comes to fine-grained recognition. In this work, we define a new FSL setting termed scarce-source few-shot learning (SSFSL), under which both the source and target classes have limited training samples. To overcome the source class data scarcity problem, a natural option is to crawl images from the web with class names as search keywords. However, the crawled images are inevitably corrupted by large amount of noise (irrelevant images) and thus may harm the performance. To address this problem, we propose a graph convolutional network (GCN)-based label denoising (LDN) method to remove the irrelevant images. Further, with the cleaned web images as well as the original clean training images, we propose a GCN-based FSL method. For both the LDN and FSL tasks, a novel adaptive aggregation GCN (AdarGCN) model is proposed, which differs from existing GCN models in that adaptive aggregation is performed based on a multi-head multi-level aggregation module. With AdarGCN, how much and how far information carried by each graph node is propagated in the graph structure can be determined automatically, therefore alleviating the effects of both noisy and outlying training samples. Extensive experiments demonstrate the superior performance of our AdarGCN under both the new SSFSL and the conventional FSL settings.

Visual dialog is a challenging vision-language task, which requires the agent to answer multi-round questions about an image. It typically needs to address two major problems: (1) How to answer visually-grounded questions, which is the core challenge in visual question answering (VQA); (2) How to infer the co-reference between questions and the dialog history. An example of visual co-reference is: pronouns (e.g., “they”) in the question (e.g., “Are they on or off?”) are linked with nouns (e.g., “lamps”) appearing in the dialog history (e.g., “How many lamps are there?”) and the object grounded in the image. In this work, to resolve the visual co-reference for visual dialog, we propose a novel attention mechanism called Recursive Visual Attention (RvA). Specifically, our dialog agent browses the dialog history until the agent has sufficient confidence in the visual co-reference resolution, and refines the visual attention recursively. The quantitative and qualitative experimental results on the large-scale VisDial v0.9 and v1.0 datasets demonstrate that the proposed RvA not only outperforms the state-of-the-art methods, but also achieves reasonable recursion and interpretable attention maps without additional annotations. The code is available at https://github.com/yuleiniu/rva.

Zero-shot learning (ZSL) is originally designed to address the small sample size problem often encountered in computer vision by recognizing unseen object classes without any training samples. Existing ZSL models (particularly deep ones) often assume that hundreds of labelled samples are collected from each seen class. In real-world applications, this assumption tends to become invalid. Therefore, a new ZSL setting is concerned in this paper: each seen class only has few labelled samples, while each unseen class still has no samples. This is more challenging yet more useful/practical than the conventional ZSL setting. To overcome the extreme label scarcity, we choose to obtain more training samples from image search engine for data augmentation: the name of each seen class is used as the query of Google, and the top returned images can be viewed as the noisy labelled samples for this seen class. With the augmented but noisy labelled training data, a novel inductive ZSL model is proposed by formulating label noise reduction (LNR) and semantic projection learning (SPL) within a unified framework: (1) LNR aims to refine the noisy labelled samples for projection learning; (2) SPL aims to learn the projection function with the refined training data. Extensive experiments show that our ZSL model outperforms the state-of-the-art alternatives.

Zero-shot learning (ZSL) typically learns a projection function between a feature space and a semantic space (e.g. attribute space) to recognize unseen classes without any training samples. However, when the projection function learned only with seen class training samples is applied to unseen class test samples, most existing ZSL models suffer from projection domain shift. In this paper, we propose a new ZSL model, termed joint inductive learning (JIL) of bidirectional projections and shared subspace, to learn more generalizable projections and thus algin the seen and unseen class domains. By inducing the two tasks of bidirectional projection learning (BPL) and shared subspace learning (SSL), we integrate three forms of projection learning into a unified framework, which makes our JIL model become more generalizable across the two domains. Moreover, since a shared subspace is actually learned between the feature and semantic spaces, our JIL model can be naturally applied to cross-modal retrieval (CMR). Extensive experiments demonstrate that our JIL model generally outperforms the state-of-the-art alternatives in the two challenging tasks (i.e. ZSL and CMR).