查看更多>>摘要：We assess whether deep convolutional networks (DCN) can account for a most fundamental property of human vision: detection/discrimination of elementary image elements (bars) at different contrast levels. The human visual process can be characterized to varying degrees of "depth,'' ranging from percentage of correct detection to detailed tuning and operating characteristics of the underlying perceptual mechanism. We challenge deep networks with the same stimuli/tasks used with human observers and apply equivalent characterization of the stimulus-response coupling. In general, we find that popular DCN architectures do not account for signature properties of the human process. For shallow depth of characterization, some variants of network-architecture/training-protocol produce human-like trends; however, more articulate empirical descriptors expose glaring discrepancies. Networks can be coaxed into learning those richer descriptors by shadowing a human surrogate in the form of a tailored circuit perturbed by unstructured input, thus ruling out the possibility that human-model misalignment in standard protocols may be attributable to insufficient representational power. These results urge caution in assessing whether neural networks do or do not capture human behavior: ultimately, our ability to assess "success "in this area can only be as good as afforded by the depth of behavioral characterization against which the network is evaluated. We propose a novel set of metrics/protocols that impose stringent constraints on the evaluation of DCN behavior as an adequate approximation to biological processes. (C) & nbsp;2022 Elsevier Ltd. All rights reserved.
查看更多>>摘要：Deep learning (DL) and reinforcement learning (RL) methods seem to be a part of indispensable factors to achieve human-level or super-human AI systems. On the other hand, both DL and RL have strong connections with our brain functions and with neuroscientific findings. In this review, we summarize talks and discussions in the "Deep Learning and Reinforcement Learning"session of the symposium, International Symposium on Artificial Intelligence and Brain Science. In this session, we discussed whether we can achieve comprehensive understanding of human intelligence based on the recent advances of deep learning and reinforcement learning algorithms. Speakers contributed to provide talks about their recent studies that can be key technologies to achieve human-level intelligence. (c) 2022 Published by Elsevier Ltd.
查看更多>>摘要：Recent years deep learning-based methods incorporating facial prior knowledge for face super resolution (FSR) are advancing and have gained impressive performance. However, some important priors such as facial landmarks are not fully exploited in existing methods, leading to noticeable artifacts in the resultant SR face images especially under large magnification. In this paper, we propose a novel multi-level landmark-guided deep network (MLGDN) for FSR. More specifically, to fully exploit the dependencies between low and high resolution images and to reduce network parameters as well as capture more reliable feature representation, we introduce a recursive back-projection network with a particular feedback mechanism for coarse-to-fine FSR. Furthermore, we incorporate an attention fusion module in the front of backbone network to strengthen face components and a feature modulation module to refine features in the middle of backbone network. By this way, the facial landmarks extracted from face images can be fully shared by the modules in different levels, which benefit to produce more faithful facial details. Both quantitative and qualitative performance evaluations on two benchmark databases demonstrate that the proposed MLGDN can achieve more impressive SR results than other state-of-the-art competitors. Code will be available at https://github. com/zhuangcheng31/MLG_Face.git/(C) 2022 Elsevier Ltd. All rights reserved.
查看更多>>摘要：Accurately predicting Polyadenylation (Poly(A)) signals is the key to understand the mechanism of translation regulation and mRNA metabolism. However, existing computational algorithms fail to work well for predicting Poly(A) signals due to the vanishing gradient problem when simply increasing the number of layers. In this work, we devise a spatiotemporal context-aware neural model called ACNet for Poly(A) signal prediction based on co-occurrence embedding. Specifically, genomic sequences of Poly(A) signals are first split into k-mer sequences, and k-mer embeddings are pre-trained based on the co-occurrence matrix information; Then, gated residual networks are devised to fully extract spatial information, which has an excellent ability to control the information flow and ease the problem of vanishing gradients. The gated mechanism generates channel weights by a dilated convolution and aggregates local features by identity connections which are obtained by multi-scale dilated convolutions. Experimental results indicate that our ACNet model outperforms the state-of-the-art prediction methods on various Poly(A) signal data, and an ablation study shows the effectiveness of the design strategy. (C) 2022 Elsevier Ltd. All rights reserved.
查看更多>>摘要：Multivariate time series forecasting remains a challenging task because of its nonlinear, non-stationary, high-dimensional, and spatial-temporal characteristics, along with the dependence between variables. To address this limitation, we propose a novel method for multivariate time series forecasting based on nonlinear spiking neural P (NSNP) systems and non-subsampled shearlet transform (NSST). A multivariate time series is first converted into the NSST domain, and then NSNP systems are automatically constructed, trained, and predicted in the NSST domain. Because NSNP systems are used as nonlinear prediction models and work in the NSST domain, the proposed prediction method is essentially a multiscale transform (MST)-based prediction method. Therefore, the proposed prediction method can process nonlinear and non-stationary time series, and the dependence between variables can be characterized by the multiresolution features of the NSST transform. Five real-life multivariate time series were used to compare the proposed prediction method with five state-of-the-art and 28 baseline prediction methods. The comparison results demonstrate the effectiveness of the proposed method for multivariate time-series forecasting.
查看更多>>摘要：Suspended sediment is a significant threat to the Great Barrier Reef (GBR) ecosystem. This catchment pollutant stems primarily from terrestrial soil erosion. Bulk masses of sediments have potential to propagate from river plumes into the mid-shelf and outer-shelf regions. Existing sediment forecasting methods suffer from the problem of low-resolution predictions, making them unsuitable for wide area coverage. In this paper, a novel sediment distribution prediction model is proposed to augment existing water quality management programs for the GBR. This model is based on the state-of-theart Transformer network in conjunction with the well-known finite element analysis. For model training, the emerging physics-informed neural network is employed to incorporate both simulated and measured sediment data. Our proposed Finite Element Transformer (FE-Transformer) model offers accurate predictions of sediment across the entire GBR. It provides unblurred outputs, which cannot be achieved with previous next-frame prediction models. This paves a way for accurate forecasting of sediment, which in turn may lead to improved water quality management for the GBR. (C) 2022 Elsevier Ltd. All rights reserved.
查看更多>>摘要：Learning efficient graph representation is the key to favorably addressing downstream tasks on graphs, such as node or graph property prediction. Given the non-Euclidean structural property of graphs, preserving the original graph data's similarity relationship in the embedded space needs specific tools and a similarity metric. This paper develops a new graph representation learning scheme, namely EGG, which embeds approximated second-order graph characteristics into a Grassmann manifold. The proposed strategy leverages graph convolutions to learn hidden representations of the corresponding subspace of the graph, which is then mapped to a Grassmann point of a low dimensional manifold through truncated singular value decomposition (SVD). The established graph embedding approximates denoised correlationship of node attributes, as implemented in the form of a symmetric matrix space for Euclidean calculation. The effectiveness of EGG is demonstrated using both clustering and classification tasks at the node level and graph level. It outperforms baseline models on various benchmarks. (C) 2022 Elsevier Ltd. All rights reserved.
Fakhoury, DanieleFakhoury, EmanueleSpeleers, Hendrik
15页
查看更多>>摘要：In this paper we present ExSpliNet, an interpretable and expressive neural network model. The model combines ideas of Kolmogorov neural networks, ensembles of probabilistic trees, and multivariate B-spline representations. We give a probabilistic interpretation of the model and show its universal approximation properties. We also discuss how it can be efficiently encoded by exploiting B-spline properties. Finally, we test the effectiveness of the proposed model on synthetic approximation problems and classical machine learning benchmark datasets. (C) 2022 Elsevier Ltd. All rights reserved.
查看更多>>摘要：The kernel principal component analysis (KPCA) serves as an efficient approach for dimensionality reduction. However, the KPCA method is sensitive to the outliers since the large square errors tend to dominate the loss of KPCA. To strengthen the robustness of KPCA method, we propose a novel robust kernel principal component analysis with optimal mean (RKPCA-OM) method. RKPCA-OM not only possesses stronger robustness for outliers than the conventional KPCA method, but also can eliminate the optimal mean automatically. What is more, the theoretical proof proves the convergence of the algorithm to guarantee that the optimal subspaces and means are obtained. Lastly, exhaustive experimental results verify the superiority of our method. (C) 2022 Elsevier Ltd. All rights reserved.
查看更多>>摘要：A recent paper (Mhaskar (2020)) introduces a straightforward and simple kernel based approximation for manifold learning that does not require the knowledge of anything about the manifold, except for its dimension. In this paper, we examine how the pointwise error in approximation using least squares optimization based on similarly localized kernels depends upon the data characteristics and deteriorates as one goes away from the training data. The theory is presented with an abstract localized kernel, which can utilize any prior knowledge about the data being located on an unknown sub-manifold of a known manifold. We demonstrate the performance of our approach using a publicly available micro-Doppler data set, and investigate the use of different preprocessing measures, kernels, and manifold dimensions. Specifically, it is shown that the localized kernel introduced in the above mentioned paper when used with PCA components leads to a near-competitive performance to deep neural networks, and offers significant improvements in training speed and memory requirements. To demonstrate the fact that our methods are agnostic to the domain knowledge, we examine the classification problem in a simple video data set. (C) 2022 Elsevier Ltd. All rights reserved.
NSTL
Elsevier