Recently, there have been increasing demands to construct compact deep architectures remove unnecessary redundancy and improve the inference speed. While many recent works focus on reducing by eliminating unneeded weight parameters, it is not possible apply a single network for multiple devices with different resources. When new device or circumstantial condition requires architecture, necessary train from scratch. In this work, we propose novel learning framework, called nested sparse...

There is a growing interest in smart homes and predicting behaviors of inhabitants key element for the success home services. In this paper, we propose two algorithms, DBN-ANN DBN-R, based on deep learning framework various activities home. We also address drawbacks contrastive divergence, widely used method restricted Boltzmann machines, an efficient online algorithm bootstrapping. From experiments using activity datasets, show that our proposed prediction algorithms outperform existing...

Learning a low-dimensional structure plays an important role in computer vision. Recently, new family of methods, such as l1 minimization and robust principal component analysis, has been proposed for low-rank matrix approximation problems shown to be against outliers missing data. But these methods often require heavy computational load can fail find solution when highly corrupted data are presented. In this paper, elastic-net regularization based factorization method subspace learning is...

In this paper, we propose a novel Gaussian process motion controller that can navigate through crowded dynamic environment. The proposed predicts future trajectories of pedestrians using an autoregressive model (AR-GPMM) from the partially-observable egocentric view robot and controls (AR-GPMC) based on predicted pedestrian trajectories. performance method is extensively evaluated in simulation validated experimentally Pioneer 3DX mobile with Microsoft Kinect sensor. particular, shows over...

In this work, we consider the problem of instance-wise dynamic network model selection for multi-task learning. To end, propose an efficient approach to exploit a compact but accurate in backbone architecture each instance all tasks. The proposed method consists estimator and selector. is based on structured hierarchically. It can produce multiple different models configurations hierarchical structure. selector chooses dynamically from pool candidate given input instance. relatively...

Low-rank matrix approximation plays an important role in the area of computer vision and image processing. Most conventional low-rank methods are based on l2 -norm (Frobenius norm) with principal component analysis (PCA) being most popular among them. However, this can give a poor for data contaminated by outliers (including missing data), because exaggerates negative effect outliers. Recently, to overcome problem, various l1 -norm, such as robust PCA methods, have been proposed...

In stereopsis, excessive screen disparity is known as one of the main causes visual fatigue. Although fatigue caused by stereoscopic depth perception has been reported in a large number subjective assessments, there little effort to measure objective indicators for this study, we investigated relationship between and amount binocular using functional magnetic resonance imaging (fMRI). As result, detected strong neuronal activities frontal eye field (FEF) when presented stimuli had...

Recently, finding the low-dimensional structure of high-dimensional data has gained much attention. Given a set points sampled from single subspace or union subspaces, goal is to learn capture underlying set. In this paper, we propose elastic-net representation, new representation framework using regularization singular values. Due strong convexity enforced by elastic-net, proposed method more stable and robust in presence heavy corruptions compared with existing lasso-type rank minimization...

ABSTRACT The aim of this study is to evaluate brain regions related with excessive binocular disparity that may be linked stereoscopic visual fatigue. In displays, generate blurring or double vision in the stereovision and induce unnatural oscillations accommodation vergence. These phenomena lead fatigue activation (or deactivation) human sensory eye movement functions. A functional magnetic resonance imaging (fMRI) method used investigate effect on brain. Subjective assessments are also...

Deep networks consume a large amount of memory by their nature. A natural question arises can we reduce that requirement whilst maintaining performance. In particular, in this work address the problem efficient learning for multiple tasks. To end, propose novel network architecture producing different configurations, termed deep virtual (DVNs), Each DVN is specialized single task and structured hierarchically. The hierarchical structure, which contains levels hierarchy corresponding to...

In this work, a new stacked encoder–decoder transformer (SEDT) model is proposed for action segmentation. SEDT composed of series modules, each which consists an encoder with self-attention layers and decoder cross-attention layers. By adding before every decoder, it preserves local information along global information. The pair also prevents the accumulation errors that occur when features are propagated through decoders. Moreover, approach performs boundary smoothing in order to handle...

Despite the continuous development of convolutional neural networks, it remains a challenge to achieve performance improvement with fewer parameters and floating point operations (FLOPs) as light-weight model. In particular, excessive expressive power on module is crucial cause skyrocketing computational cost entire network. We argue that necessary optimize network by optimizing single modules or blocks Therefore, we propose GhostNeXt, promising alternative GhostNet, adjusting configuration...

In this paper, we propose a novel regression method that can incorporate both positive and negative training data into single framework. detail, leveraged kernel function for non-stationary Gaussian process is proposed. With new function, vary the correlation betwen two inputs in directions by adjusting leverage parameters. By using property, resulting anchor regressor to while avoiding data. We first prove semi-definiteness of Bochner's theorem. Then, apply real-time motion control problem....

While human intelligence can easily recognize some characteristics of classes with one or few examples, learning from examples is a challenging task in machine learning. Recently emerging deep generally requires hundreds thousands samples to achieve generalization ability. Despite recent advances learning, it not easy generalize new little supervision. Few-shot (FSL) aims learn how per class. However, makes the model difficult and susceptible overfitting. To overcome difficulty, data...

This paper considers the problem of modeling complex motions pedestrians in a crowded environment. A number methods have been proposed to predict motion pedestrian or an object. However, it is still difficult make good prediction due challenges, such as complexity and outliers training set. addresses these issues by proposing robust autoregressive model based on Gaussian process regression using l <sub xmlns:mml="http://www.w3.org/1998/Math/MathML"...

Abstract Recent attention has been focused on detecting interregional connectivity in a resting state of the brain, general, described terms functional based magnetic resonance imaging (fMRI) data. The fMRI data are given form multivariate time‐series. authors have proposed model for effective brain regions autoregressive (MAR) model. MAR modeling allows identification by combining graphical methods with concept Granger causality. In our current model, time‐series were performed only when...

This paper considers the problem of approximating a kernel matrix in an autoregressive Gaussian process regression (AR-GP) presence measurement noises or natural errors for modeling complex motions pedestrians crowded environment. While number methods have been proposed to robustly predict future humans, it still remains as difficult noises. addresses this issue by proposing structured low-rank approximation method using nuclear-norm regularized l <sub...

We present a computationally as well statistically efficient method of inferring causal networks for the brain regions. It is based on James‐Stein‐type shrinkage estimation covariance matrix, suggested by (Opgen‐Rhein and Strimmer, BMC Syst Biol 1 ( ), 37‐40), among different regions interest functional magnetic resonance imaging (fMRI) experiment, that enhance accuracy vector autoregressive (VAR) model coefficient estimates. have shown this approach suited small number samples in time large...

High Dynamic Range (HDR) imaging seeks to enhance image quality by combining multiple Low (LDR) images captured at varying exposure levels. Traditional deep learning approaches often employ reconstruction loss, but this method can lead ambiguities in feature space during training. To address issue, we present a new loss function, termed Gravitated Latent Space (GLS) that leverages metric tensor introduce form of virtual gravity within the latent space. This helps model overcoming saddle...

Neural Architecture Search (NAS) has emerged as a promising tool in the field of AutoML for designing more accurate and efficient architectures. The majority NAS works employ weight-sharing technique to reduce search cost by sharing weights supernet, which is composite all architectures produced from space. Nonetheless, this method significant drawback that negative interference may arise when candidate share same weights. This issue becomes even severe multi-task searches, where supernet...