Search

As an elementary mesh quality improvement technique, smoothing has been widely used in finite element (FE) analysis. Heuristic smoothing methods and optimization-based smoothing methods are the two main smoothing types. The former is efficient. However, it operates heuristically and may create low-quality elements. In contrast, optimization-based smoothing is very effective at improving mesh quality. However, it suffers from high computational cost since it calculates the optimal position of a free node iteratively. In this paper, we present a new smoothing method. The proposed method imitates the optimization-based smoothing based on a neural network, but it calculates the optimal position of a free node straightforwardly. Hence, the proposed method is more efficient than these optimization-based smoothing methods while being comparable in terms of mesh quality. We present various testing results to illustrate the effectiveness of the proposed method.

Markov state models (MSMs) based on molecular dynamics (MD) simulations are routinely employed to study protein folding, however, their application to functional conformational changes of biomolecules is still limited. In the past few years, the field of computational chemistry has experienced a surge of advancements stemming from machine learning algorithms, and MSMs have not been left out. Unlike global processes, such as protein folding, the application of MSMs to functional conformational changes is challenging because they mostly consist of localized structural transitions. Therefore, it is critical to properly select a subset of structural features that can describe the slowest dynamics of these functional conformational changes. To address this challenge, we recommend several automatic feature selection methods such as Spectral-OASIS. To identify states in MSMs, the chosen features can be subject to dimensionality reduction methods such as TICA or deep learning based VAMPNets to project MD conformations onto a few collective variables for subsequent clustering. Another challenge for the application of MSMs to the study of functional conformational changes is the ability to comprehend their biophysical mechanisms, as MSMs built for these processes often require a large number of states. We recommend the recently developed quasi-MSMs (qMSMs) to address this issue. Compared to MSMs, qMSMs encode the non-Markovian dynamics via the generalized master equation and can significantly reduce the number of states. As a result, qMSMs can be built with a handful of states to facilitate the interpretation of functional conformational changes. In the wake of machine learning, we believe that the rapid advancement in the MSM methodology will lead to their wider application in studying functional conformational changes of biomolecules.

Spiking neural networks (SNNs) possess energy-efficient potential due to event-based computation. However, supervised training of SNNs remains a challenge as spike activities are non-differentiable. Previous SNNs training methods can be generally categorized into two basic classes, i.e., backpropagation-like training methods and plasticity-based learning methods. The former methods are dependent on energy-inefficient real-valued computation and non-local transmission, as also required in artificial neural networks (ANNs), whereas the latter are either considered to be biologically implausible or exhibit poor performance. Hence, biologically plausible (bio-plausible) high-performance supervised learning (SL) methods for SNNs remain deficient. In this paper, we proposed a novel bio-plausible SNN model for SL based on the symmetric spike-timing dependent plasticity (sym-STDP) rule found in neuroscience. By combining the sym-STDP rule with bio-plausible synaptic scaling and intrinsic plasticity of the dynamic threshold, our SNN model implemented SL well and achieved good performance in the benchmark recognition task (MNIST dataset). To reveal the underlying mechanism of our SL model, we visualized both layer-based activities and synaptic weights using the t-distributed stochastic neighbor embedding (t-SNE) method after training and found that they were well clustered, thereby demonstrating excellent classification ability. Furthermore, to verify the robustness of our model, we trained it on another more realistic dataset (Fashion-MNIST), which also showed good performance. As the learning rules were bio-plausible and based purely on local spike events, our model could be easily applied to neuromorphic hardware for online training and may be helpful for understanding SL information processing at the synaptic level in biological neural systems., Comment: 29 pages, 6 figures

Deciphering the free energy landscape of biomolecular structure space is crucial for understanding many complex molecular processes, such as protein-protein interaction, RNA folding, and protein folding. A major source of current dynamic structure data is Molecular Dynamics (MD) simulations. Several methods have been proposed to investigate the free energy landscape from MD data, but all of them rely on the assumption that kinetic similarity is associated with global geometric similarity, which may lead to unsatisfactory results. In this paper, we proposed a new method called Conditional Angle Partition Tree to reveal the hierarchical free energy landscape by correlating local geometric similarity with kinetic similarity. Its application on the benchmark alanine dipeptide MD data showed a much better performance than existing methods in exploring and understanding the free energy landscape. We also applied it to the MD data of Villin HP35. Our results are more reasonable on various aspects than those from other methods and very informative on the hierarchical structure of its energy landscape.

Fragment-based drug design plays an important role in the drug discovery process by reducing the complex small-molecule space into a more manageable fragment space. We leverage the power of deep learning to design ChemPLAN-Net; a model that incorporates the pairwise association of physicochemical features of both the protein drug targets and the inhibitor and learns from thousands of protein co-crystal structures in the PDB database to predict previously unseen inhibitor fragments. Our novel protocol handles the computationally challenging multi-label, multi-class problem, by defining a fragment database and using an iterative featurepair binary classification approach. By training ChemPLAN-Net on available co-crystal structures of the protease protein family, excluding HIV-1 protease as a target, we are able to outperform fragment docking and recover the target’s inhibitor fragments found in co-crystal structures or identified by in-vitro cell assays.

The brain is highly plastic, with synaptic weights changing across a wide range of time scales, from hundreds of milliseconds to days. Changes occurring at different temporal scales are believed to serve different purposes, with long-term changes for learning and memory and short-term changes for adaptation and synaptic computation. By studying the performance of reservoir computing (RC) models in a memory task, we revealed that short-term synaptic plasticity is fundamentally important for long-term synaptic changes in neural networks. Specifically, short-term depression (STD) greatly expands the operational range of a neural network in which it can accommodate long-term synaptic changes while maintaining system performance. This is achieved by dynamically adjusting neural networks close to a critical state. The effects of STD can be further strengthened by synaptic weight heterogeneity, resulting in networks that can tolerate very large, long-term changes in synaptic weights. Our results highlight a potential mechanism used by the brain to organize plasticity at different time scales, thereby maintaining optimal information processing while allowing internal structural changes necessary for learning and memory.

Recently, the transductive graph-based methods have achieved great success in the few-shot classification task. However, most existing methods ignore exploring the class-level knowledge that can be easily learned by humans from just a handful of samples. In this paper, we propose an Explicit Class Knowledge Propagation Network (ECKPN), which is composed of the comparison, squeeze and calibration modules, to address this problem. Specifically, we first employ the comparison module to explore the pairwise sample relations to learn rich sample representations in the instance-level graph. Then, we squeeze the instance-level graph to generate the class-level graph, which can help obtain the class-level visual knowledge and facilitate modeling the relations of different classes. Next, the calibration module is adopted to characterize the relations of the classes explicitly to obtain the more discriminative class-level knowledge representations. Finally, we combine the class-level knowledge with the instance-level sample representations to guide the inference of the query samples. We conduct extensive experiments on four few-shot classification benchmarks, and the experimental results show that the proposed ECKPN significantly outperforms the state-of-the-art methods., Accepted by CVPR2021

Molecular Dynamic (MD) simulations have been extensively used as a powerful tool to investigate dynamics of biological molecules in recent decades. Generally, MD simulations generate high-dimensional data that is very hard to visualize and comprehend. As a result, clustering algorithms have been commonly used to reduce the dimensionality of MD data with the key benefit being their ability to reduce the dimensionality of MD data without prior knowledge of structural details or dynamic mechanisms. In this paper, we propose a new algorithm, the Multi-Level Density-Based Spatial Clustering of Applications with Noise (ML-DBSCAN), which combines the clustering results at different resolution of density levels to obtain the hierarchical structure of the free energy landscape and the metastable state assignment. At relatively low resolutions, the ML-DBSCAN can efficiently detect high population regions that contain all metastable states, while at higher resolutions, the ML-DBSCAN can find all metastable states and structural details of the free energy landscape. We demonstrate the powerfulness of the ML-DBSCAN in generating metastable states with a particle moving in a Mexican hat-like potential, and four peptide and protein examples are used to demonstrate how hierarchical structures of free energy landscapes can be found. Furthermore, we developed a GPU implementation of the ML-DBSCAN, which allows the algorithm to handle larger MD datasets and be up to two orders of magnitude faster than the CPU implementation. We demonstrate the power of the ML-DBSCAN on MD simulation datasets of five systems: a 2D-potential, alanine dipeptide, β-hairpin Tryptophan Zipper 2 (Trpzip2), Human Islet Amyloid Polypeptide (hIAPP), and Maltose Binding Protein (MBP). Our code is available at https://github.com/liusong299/ML-DBSCAN.

Cryogenic Electron Microscopy (cryo-EM) preserves the ensemble of protein conformations in solution and thus provide a promising way to characterize conformational changes underlying protein functions. However, it remains challenging for existing software to elucidate distributions of multiple conformations from a heterogeneous cryo-EM dataset. We developed a new algorithm: Linear Combinations of Template Conformations (LCTC) to obtain distributions of multiple conformations from cryo-EM datasets. LCTC assigns 2D images to the template 3D structures obtained by Multi-body Re-finement of RELION via a novel two-stage matching algorithm. Specifically, an initial rapid assignment of experimental 2D images to template 2D images was applied based on auto-correlation functions of image contours that can efficiently remove the majority of irrelevant 2D images. This is followed by pixel-pixel matching of images with fewer number of 2D images, which can accurately assign the 2D images to the template images. We validate the LCTC method by demonstrating that it can accurately reproduce the distributions of 3Thermus aquaticus(Taq) RNA polymerase (RNAP) structures with different degrees of clamp opening from a simulated cryo-EM dataset, in which the correct distributions are known. For this dataset, we also show that LCTC greatly outperforms clustering-based Manifold Embedding and Maximum Likelihood-based Multi-body Re-finement algorithms in terms of reproducing the structural distributions. Lastly, we also successfully applied LCTC to reveal the populations of various clamp-opening conformations from an experimentalEscherichia coliRNAP cryo-EM dataset. Source code is available athttps://github.com/ghl1995/LCTC.

Cross-laminated timber (CLT) is well known as an interesting technical and economical product for modern wood structures. The use of CLT for modern construction industry has become increasingly popular in particular for residential timber buildings. Analyzing the CLT behavior in high thermal environment has attracted scholars’ attention. Thermal environment greatly influences the CLT properties and load bearing capacity of CLT, and the investigation can form the basis for predicting the structural response of such CLT-based structures. In the present work, the finite element method (FEM) is employed to analyze the thermal influence on the deformation of CLT. Furthermore, several factors were taken into consideration, including board layer number, hole conformation, and hole position, respectively. In order to determine the influence, several numerical models for different calculation were established. The calculation process was validated by comparing with published data. The performance is quantified by demonstrating the temperature distribution and structural deformation.

The triangular mesh is widely used in computer graphics. The advancing-front-Delaunay method is a mainstream method to generate the triangular mesh. However, it generates interior nodes on the basis of the segment front and needs to manage and update the generation segment front set carefully. This paper describes an improved advancing-front-Delaunay method that generates interior nodes based on the node front. The idea of node front can be implemented easily by our disk packing algorithm and does not need a complicated management strategy. Besides, unlike the traditional advancing-front-Delaunay method that generates interior node and the mesh at the same time, the method generates all the nodes firstly by the disk packing method, then generates the mesh. Hence, the method can be more efficient using these more efficient algorithms for a given fixed node-set to generate Delaunay triangular meshes or these more efficient algorithms with a carefully designed insertion sequence to insert the interior nodes. Four examples are given to show the effectiveness and robustness of the improved advancing-front-Delaunay method.

Functional conformational changes of proteins can facilitate numerous biological events in cells. The Markov state model (MSM) built from molecular dynamics simulations provide a powerful approach to study them. We here introduce a protocol that is tailor-made for constructing MSMs to study the functional conformational changes of proteins. In this protocol, one of the important steps is to select proper molecular features that can collectively describe the slowest timescales of conformational changes of interest. We recommend spectral oASIS, the modified version of oASIS, as a promising approach for automatic feature selection. Recently developed deep learning methods could also serve efficient approaches for selecting features and finding collective variables. Using DNA repair enzymes and RNA polymerases as examples, we review recent applications of MSMs to elucidate molecular mechanisms of functional conformational changes. Finally, we discuss remaining challenges and future perspectives for constructing MSMs to study functional conformational changes of proteins.

Understanding and analyzing multi-view data is a fundamental research topic of feature learning for a wide range of practical applications such as image classification. Canonical correlation analysis (CCA) is a popular unsupervised method of analyzing multi-view data, which captures common subspace of two groups of variable sets by maximizing the correlations between them. However, traditional CCA ignores the underlying geometric structure within dataset, which shows great power in describing data distribution, and thus fails in some tasks such as classification. To handle this limitation, this paper proposes an improved CCA variant of Nonnegative Constrained Graph regularized CCA (NCGCCA). Specifically, we improve CCA to NCGCCA with the following two contributions. Firstly, we develop a nonnegative constrained graph based self-representation to explore the underlying group-wise structure within dataset. Secondly, based on the proposed informative representation, we offer a graph embedding schema to incorporate the underlying structure into CCA. Experiments of image classification on four face datasets including Yale, ORL, UMIST, and YaleB demonstrate the efficacy of NCGCCA compared with existing baseline CCA methods.

Time warping is the popular technique of temporally aligning two sequences and has successfully applied in temporal alignment tasks such as activity recognition. However, existing time warping methods suffer from limited representation ability because aligning process is performed on either raw sequences or the projected lower-dimensional features. In this paper, we propose a stacked time warping framework (STW) to learn layer-wise representation for temporal alignment in a stacked structure. By using this structure, STW gives higher flexibility than existing methods meanwhile unifies them into a deep architecture. Based on the proposed STW framework, we explore a stacked marginal time warping (SMTW) method by using marginal stacked denoising autoencoder (mSDA) as the regularization term which enables SMTW to marginalize out noises and learn layer-wise non-linear representations with the effective closed-form solution. Benefitting from the incorporation of mSDA, SMTW achieves better alignment performance and keeps comparable time efficiency with regular time warping methods. Experiments on both synthetic data and practical human activity recognition datasets demonstrate that SMTW is superior to the state-of-the-art time warping methods in quantity.

In molecular self-assembly, hundreds of thousands of freely-diffusing molecules associate to form ordered and functional architectures in the absence of an actuator. This intriguing phenomenon plays a critical role in biology and has become a powerful tool for the fabrication of advanced nanomaterials. Due to the limited spatial and temporal resolutions of current experimental techniques, computer simulations offer a complementary strategy to explore self-assembly with atomic resolution. Here, we review recent computational studies focusing on both thermodynamic and kinetic aspects. As we shall see, thermodynamic approaches based on modeling and statistical mechanics offer initial guidelines to design nanostructures with modest computational effort. Computationally more intensive analyses based on molecular dynamics simulations and kinetic network models (KNMs) reach beyond it, opening the door to the rational design of self-assembly pathways. Current limitations of these methodologies are discussed. We anticipate that the synergistic use of thermodynamic and kinetic analyses based on computer simulations will provide an important contribution to the de novo design of self-assembly.

Locating the minimum free energy paths (MFEPs) between two conformational states is among the most important tasks of biomolecular simulations. For example, knowledge of the MFEP is critical for focusing the effort of unbiased simulations that are used for the construction of Markov state models to the biologically relevant regions of the system. Typically, existing path searching methods perform local sampling around the path nodes in a pre-selected collective variable (CV) space to allow a gradual downhill evolution of the path toward the MFEP. Despite the wide application of such a strategy, the gradual path evolution and the non-trivial a priori choice of CVs are also limiting its overall efficiency and automation. Here we demonstrate that non-local perpendicular sampling can be pursued to accelerate the search, provided that all nodes are reordered thereafter via a traveling-salesman scheme. Moreover, path-CVs can be computed on-the-fly and used as a coordinate system, minimizing the necessary prior knowledge about the system. Our traveling-salesman based automated path searching method achieves a 5-8 times speedup over the string method with swarms-of-trajectories for two peptide systems in vacuum and solution, making it a promising method for obtaining initial pathways when investigating functional conformational changes between a pair of structures.

The prefrontal cortex (PFC), which plays key roles in many higher cognitive processes, is a hierarchical system consisting of multi-scale organizations. Optimizing the working state at each scale is essential for PFC's information processing. Typical optimal working states at different scales have been separately reported, including the dopamine-mediated inverted-U profile of the working memory (WM) at the system level, critical dynamics at the network level, and detailed balance of excitatory and inhibitory currents (E/I balance) at the cellular level. However, it remains unclear whether these states are scale-specific expressions of the same optimal state and, if so, what is the underlying mechanism for its regulation traversing across scales. Here, by studying a neural network model, we show that the optimal performance of WM co-occurs with the critical dynamics at the network level and the E/I balance at the level of individual neurons, suggesting the existence of a unified, multi-scale optimal state for the PFC. Importantly, such a state could be modulated by dopamine at the synaptic level through a series of U or inverted-U profiles. These results suggest that seemingly different optimal states for specific scales are multi-scale expressions of one condition regulated by dopamine. Our work suggests a cross-scale perspective to understand the PFC function and its modulation.

Semantic segmentation is a pixel-level image dense labeling task and plays a core role in autonomous driving. In this regard, how to balance between precision and speed is a frequently-studied issue. In this paper, we propose an alternative attentional residual dense factorized network (AttRDFNet) to address this issue. Specifically, we design a residual dense factorized convolution block (RDFB), which reaps the benefits of low-level and high-level layer-wise features through dense connection to boost segmentation precision whilst enjoying efficient computation by factorizing large convolution kernel into the product of two smaller kernels. This reduces computational burdens and makes real time become possible. To further leverage layer-wise features, we explore the graininess-aware channel and spatial attention modules to model different levels of salient features of interest. As a result, AttRDFNet can run with the inputs of the resolution 512 \( \times \) 1024 at the speed of 55.6 frames per second on a single Titan X GPU with solid 68.5% Mean IOU on the test set of Cityscapes. Experiments on the Cityscapes dataset show that AttRDFNet has real-time inference whilst achieving competitive precision against well-behaved counterparts.

We present an efficient density-based adaptive-resolution clustering method APLoD for analyzing large-scale molecular dynamics (MD) trajectories. APLoD performs the k-nearest-neighbors search to estimate the density of MD conformations in a local fashion, which can group MD conformations in the same high-density region into a cluster. APLoD greatly improves the popular density peaks algorithm by reducing the running time and the memory usage by 2-3 orders of magnitude for systems ranging from alanine dipeptide to a 370-residue Maltose-binding protein. In addition, we demonstrate that APLoD can produce clusters with various sizes that are adaptive to the underlying density (i.e., larger clusters at low-density regions, while smaller clusters at high-density regions), which is a clear advantage over other popular clustering algorithms including k-centers and k-medoids. We anticipate that APLoD can be widely applied to split ultra-large MD datasets containing millions of conformations for subsequent construction of Markov State Models. © 2016 Wiley Periodicals, Inc.

Biomolecular dynamics play an important role in numerous biological processes. Markov State Models (MSMs) provide a powerful approach to study these dynamic processes by predicting long time scale dynamics based on many short molecular dynamics (MD) simulations. In an MSM, protein dynamics are modeled as a kinetic process consisting of a series of Markovian transitions between different conformational states at discrete time intervals (called "lag time"). To achieve this, a master equation must be constructed with a sufficiently long lag time to allow interstate transitions to become truly Markovian. This imposes a major challenge for MSM studies of proteins since the lag time is bound by the length of relatively short MD simulations available to estimate the frequency of transitions. Here, we show how one can employ the generalized master equation formalism to obtain an exact description of protein conformational dynamics both at short and long time scales without the time resolution restrictions imposed by the MSM lag time. Using a simple kinetic model, alanine dipeptide, and WW domain, we demonstrate that it is possible to construct these quasi-Markov State Models (qMSMs) using MD simulations that are 5-10 times shorter than those required by MSMs. These qMSMs only contain a handful of metastable states and, thus, can greatly facilitate the interpretation of mechanisms associated with protein dynamics. A qMSM opens the door to the study of conformational changes of complex biomolecules where a Markovian model with a few states is often difficult to construct due to the limited length of available MD simulations.

Spatially separated brain areas interact with each other to form networks with coordinated activities, supporting various brain functions. Interaction structures among brain areas have been widely investigated through pairwise measures. However, interactions among multiple (e.g., triple and quadruple) areas cannot be reduced to pairwise interactions. Such higher order interactions (HOIs), e.g., exclusive-or (XOR) operation, are widely implemented in computation systems and are crucial for effective information processing. However, it is currently unclear whether any HOIs are present in large-scale brain functional networks when subjects are executing specific tasks. Here we analyzed functional magnetic resonance imaging (fMRI) data collected from human subjects executing various perceptual, motor, and cognitive tasks. We found that HOI strength in the macroscopic functional networks was very weak for all tasks, suggesting that major brain activities do not rely on HOIs on the macroscopic level at the timescale of hundreds of milliseconds. These weak HOIs during tasks were further investigated with a neural network model activated by external inputs, which suggested that weak pairwise interactions among brain areas organized the system without involving HOIs. Taken together, these results demonstrated the dominance of pairwise interactions in organizing coordinated activities among different brain areas to support various brain functions.

The cryo-electron microscope (cryo-EM) is increasingly popular these years. It helps to uncover the biological structures and functions of macromolecules. In this paper, we address image denoising problem in cryo-EM. Denoising the cryo-EM images can help to distinguish different molecular conformations and improve three dimensional reconstruction resolution. We introduce the use of data-driven tight frame (DDTF) algorithm for cryo-EM image denoising. The DDTF algorithm is closely related to the dictionary learning. The advantage of DDTF algorithm is that it is computationally efficient, and can well identify the texture and shape of images without using large data samples. Experimental results on cryo-EM image denoising and conformational classification demonstrate the power of DDTF algorithm for cryo-EM image denoising and classification., 2018 IEEE Global Signal and Information Processing

Canonical correlation analysis (CCA) is a classical subspace learning method of capturing the common semantic information underlying multi-view data. It has been used in person re-identification (re-ID) task by treating the task of matching identical individuals across non-overlapping multi-cameras as a multi-view learning problem. However, CCA-based reID methods still achieve unsatisfactory results because few jointly consider discriminative margin information and selecting importantly relevant features. To address this issue, we propose a novel $l_{2,1}$ -norm regularized margin-embedding CCA ( $l_{2,1}$ -MCCA), which learns a generalized discriminative subspace by employing more discriminative margin information. Moreover, the new method enforces the $l_{2,1}$ -norm regularization term over the learned subspace to identify the relevant features. Both lightweight and effective schemes can benefit from each other and endeavor to enlarge the interclass variations whilst reducing the intra-class variations. Experiments on three popular datasets show the efficacy of $l_{2,1}$ -MCCA as compared with recently representative re-ID methods.

Subspace clustering seeks to segment a given unlabeled data into clusters with the hope of each cluster corresponding to a union of low-dimensional subspaces. Among them, low-rank representation (LRR) is a promising potential method which intends to build a good affinity matrix by using the self-expression of inputs. However, it completely ignores the important data locality. Although several works in this regard have considered the local geometric structure through the Laplacian regularizer, they also neglect the correlation of the data. In this paper, we propose a graph-Laplacian correlated low-rank representation model (GCLRR) to address such an issue. Particularly, GCLRR factorizes the self-expression as the product of two low-dimensional matrices, of which one is the latent representation of the self-expression. On the basis of the latent representation, the Laplacian regularizer is integrated with the orthogonal constraint together and behaves like the spectral clustering. Moreover, we devise a Frobenius norm based trace loss and use it to constrain both the latent representation and the self-expression to capture the correlation of the data. Our improved trace loss is more efficient than the original one. More importantly, GCLRR provides an effective unified framework to seamlessly integrate both aspects above. Then, we optimize GCLRR in the frame of alternating direction method (ADM) and fortunately derive the analytical solution to each subproblem. Experiments of motion segmentation and image clustering confirm the efficacy of the proposed GCLRR.

Person re-identification (re-ID) seeks to match the identical individuals across different cameras and is still a challenging visual task due to substantial variances of person appearance in complex scenarios. Different from most of conventional person re-ID methods, which generally reduce person re-ID task to either a multi-view learning problem or a multi- domain learning problem alone, this paper treats such a task as a multi-view multi-domain (MVMD) learning problem to exploit the both benefits by refreshing canonical correlation analysis (CCA) with two improvements, termed as ranking-embedded transfer CCA (RTCCA). Specifically, to bridge the semantic gap between different views, we first embed a ranking weight matrix into CCA to strength the correlations among the multi-view images of the same identity and simultaneously to weaken that of different identities. Furthermore, we utilize the well-known distribution metric maximum mean discrepancy (MMD) as a regularization term to reduce the domain shift between training set and testing set. More importantly, the two improvements benefit from each other and the joint merit can further boost the re-ID performance. Experiments on three benchmarks verify the efficacy of the proposed RTCCA when compared with the recently representative baseline person re-ID methods.

In unsupervised graph-based hashing for large-scale image retrieval, many efforts have been made to bridge the gap between the learned graph embedding and the corresponding binary codes. Relatively, few studies focus on the issue of the discrimination of graph embedding. In this paper, we firstly devise a discrete graph hashing model (DGH) that smooths graph embedding and simultaneously solving binary codes under the balanced discrete constraint, which equals a novel method of jointly learning graph embedding and spectral rotation, theoretically. To further induce discriminant graph embedding, we substitute affine transformation for spectral rotation in our DGH (abbreviated as ADGH). This is because affine transformation can accommodate both rotational angle and distance of graph embedding, while respecting the neighborhood structure among most samples. Besides, each subproblem of ADGH can yield the closed-form solution. Experiments of image retrieval on three benchmark datasets show that ADGH outperforms the representative hashing methods in quantity.

Speed and accuracy are the two most important focuses for many visual tracking methods. Recently, siamese networks based trackers have shown very promising potentials in both aspects, which develop a twin network to measure the responses between target and hypotheses with a fully convolutional operation. However, the learned response maps are vulnerable to background clutters and scale changes as they ignore priori knowledge such as the object salience and multi-granularity cues. To explore the benefits of priori, this paper devises a multi-granularity hierarchical attention siamese network tracker (MHA-Siam) to further enhance the tracking stability without sacrificing real-time speed. Particularly, the channel-wise attention mechanism is exploited here to filter out the background while remain the salient object region; then, the response maps of the coarse-to-finer multi-layer features are fused to capture multi-granularity location information helpful for improvement in tracking stability. To make full use of them, MHA-Siam imposes the element-wise max-and-sum operation on them to induce a reliable response map for accurate location. Experiments of visual tracking on OTB benchmark shows the superiority of MHA-Siam with the competitive efficiency to its counterpart trackers.

Projective non-negative matrix factorization (PNMF) learns a non-negative projection matrix to project highdimensional examples onto a lower-dimensional space spanned by the transpose of the learned projection matrix. Since PNMF can learn parts-based representation, it has attracted ample attention from computer vision community. However, existing PNMF methods either completely ignore labels of the dataset or endure the slow convergent optimization algorithm. In this paper, we propose a box-constrained discriminant PNMF (BDPNMF) method to address these issues. Specifically, BDPNMF jointly exploits the Fisher’s criterion and the augmented Lagrangian multiplier (ALM) method into PNMF to boost discriminative capacity of the learned subspace and its efficiency. Experimental results on four popular face image datasets confirm the efficacy of BDPNMF compared to previous PNMF methods in quantity.

In most trackers for visual tracking, Siamese network based trackers construct a pair of twin structures to learn a similarity metric between tracked object and search region to predict the position of the object in the coming frame. They have achieved impressive performance in both speed and accuracy. However, semantic features from different layers are not fully explored in most current Siamese network based tracker. To this, we propose a cross-layer convolutional Siamese network tracker (Siam-CC) which attempts to explore more semantic features of different layers from two aspects. Firstly, we combine the shallow-to-deep cross-layer convolutional response maps to capture various semantic-aware features and meanwhile enforce Siam-CC to only focus on the most interesting location, because much more semantic information is able to reduce negative effect of background. Secondly, to further boost the discrimination of responses, an adaptive contrastive loss is additionally developed together with traditional logistical loss, which, to some extent, assists in filtering out some noisy responses. Experiments on a large-scale benchmark dataset show the effectiveness of Siam-CC as compared to the state-of-the-art trackers.

Constructing Markov state models from large-scale molecular dynamics simulation trajectories is a promising approach to dissect the kinetic mechanisms of complex chemical and biological processes. Combined with transition path theory, Markov state models can be applied to identify all pathways connecting any conformational states of interest. However, the identified pathways can be too complex to comprehend, especially for multi-body processes where numerous parallel pathways with comparable flux probability often coexist. Here, we have developed a path lumping method to group these parallel pathways into metastable path channels for analysis. We define the similarity between two pathways as the intercrossing flux between them and then apply the spectral clustering algorithm to lump these pathways into groups. We demonstrate the power of our method by applying it to two systems: a 2D-potential consisting of four metastable energy channels and the hydrophobic collapse process of two hydrophobic molecules. In both cases, our algorithm successfully reveals the metastable path channels. We expect this path lumping algorithm to be a promising tool for revealing unprecedented insights into the kinetic mechanisms of complex multi-body processes.

Clustering has long been a popular topic in machine learning and is the basic task of many vision applications. Graph regularized NMF (GNMF) and its variants as extensions of NMF decompose the whole dataset as the product of two low-rank matrices which respectively indicate centroids of clusters and cluster memberships for each sample. Although they utilize graph structure to reveal the geometrical structure within datasets, these methods completely ignore the robustness of graph structure. To address the issue above, this paper jointly incorporates a novel Robust Graph and Reconstruction-based Graph regularization into NMF (RG\(^2\)NMF) to promote the gain in clustering performance. Particularly, RG\(^2\)NMF stabilizes the objective of GNMF through the reconstruction regularization, and meanwhile exploits a learning procedure to derive the robust graph. Experiments of image clustering on two popular datasets illustrate the effectiveness of RG\(^2\)NMF compared with the baseline methods in quantities.

Existing approaches for still image based action recognition rely heavily on bounding boxes and could be restricted to specific applications with bounding boxes available. Thus, exploring the boxless action recognition in still images is very challenging for lack of any supervised knowledge. To address this issue, we propose an attention focused spatial pyramid pooling (SPP) network (AttSPP-net) free from the bounding boxes by jointly integrating the soft attention mechanism and SPP into a convolutional neural network. Particularly, soft attention mechanism automatically indicates relevant image regions to be an action. Besides, AttSPP-net further exploits SPP to boost the robustness to action deformation by capturing spatial structures among image pixels. Experiments on two public action recognition benchmark datasets including PASCAL VOC 2012 and Stanford-40 demonstrate that AttSPP-net can achieve promising results and even outweighs some methods based on ground-truth bounding boxes, and provides an alternative way towards practical applications.

Boxless action recognition in still images means recognizing human actions in the absence of ground-truth bounding boxes. Since no ground-truth bounding boxes are provided, boxless action recognition is more challenging than traditional action recognition tasks. Towards this end, AttSPP-net jointly integrates soft attention and spatial pyramid pooling into a convolutional neural network, and achieves comparable recognition accuracies even with some bounding box based approaches. However, the soft attention of AttSPP-net concentrates on only one fixation, rather than combining information from different fixations over time, which is the mechanism of human visual attention. In this paper, we take inspiration from this mechanism and propose a ReAttSPP-net for boxless action recognition. ReAttSPP-net utilizes a recurrent neural network model of visual attention in order to extract information from a sequence of fixations. Experiments on three public action recognition benchmark datasets including PASCAL VOC 2012, Willow and Sports demonstrate that ReAttSPP-net can achieve promising results and obtains higher recognition performance than AttSPP-net.

An improved anti-noise morphology vision navigation algorithm is proposed for intelligent tractor tillage in a complex agricultural field environment. At first, the two key steps of guided filtering and improved anti-noise morphology navigation line extraction were addressed in detail. Then, the experiments were carried out in order to verify the effectiveness and advancement of the presented algorithm. Finally, the optimal template and its application condition were studied for improving the image-processing speed. The comparison experiment results show that the YCbCr color space has minimum time consumption of 0.094 &nbsp, s in comparison with HSV, HIS, and 2R-G-B color spaces. The guided filtering method can effectively distinguish the boundary between the tillage soil compared to other competing vanilla methods such as Tarel, multi-scale retinex, wavelet-based retinex, and homomorphic filtering in spite of having the fastest processing speed of 0.113 &nbsp, s . The extracted soil boundary line of the improved anti-noise morphology algorithm has the best precision and speed compared to other operators such as Sobel, Roberts, Prewitt, and Log. After comparing different sizes of image templates, the optimal template with the size of 140 &nbsp, &times, &nbsp, 260 pixels could achieve high-precision vision navigation while the course deviation angle was not more than 7.5 &deg, The maximum tractor speed of the optimal template and global template were 51.41 &nbsp, km / h and 27.47 &nbsp, km / h , respectively, which can meet the real-time vision navigation requirement of the smart tractor tillage operation in the field. The experimental vision navigation results demonstrated the feasibility of the autonomous vision navigation for tractor tillage operation in the field using the tillage soil boundary line extracted by the proposed improved anti-noise morphology algorithm, which has broad application prospect.

Wearable device with an ego-centric camera would be the next generation device for human-computer interaction such as robot control. Hand gesture is a natural way of ego-centric human-computer interaction. In this paper, we present an ego-centric multi-stage hand gesture analysis pipeline for robot control which works robustly in the unconstrained environment with varying luminance. In particular, we first propose an adaptive color and contour based hand segmentation method to segment hand region from the ego-centric viewpoint. We then propose a convex U-shaped curve detection algorithm to precisely detect positions of fingertips. And parallelly, we utilize the convolutional neural networks to recognize hand gestures. Based on these techniques, we combine most information of hand to control the robot and develop a hand gesture analysis system on an iPhone and a robot arm platform to validate its effectiveness. Experimental result demonstrates that our method works perfectly on controlling the robot arm by hand gesture in real time.

Constructing Markov State Models (MSMs) based on short molecular dynamics simulations is a powerful computational technique to complement experiments in predicting long-time kinetics of biomolecular processes at atomic resolution. Even though the MSM approach has been widely applied to study one-body processes such as protein folding and enzyme conformational changes, the majority of biological processes, e.g. protein–ligand recognition, signal transduction, and protein aggregation, essentially involve multiple entities. Here we review the attempts at constructing MSMs for multi-body systems, point out the challenges therein and discuss recent algorithmic progresses that alleviate these challenges. In particular, we describe an automatic kinetics based partitioning method that achieves optimal definition of the conformational states in a multi-body system, and discuss a novel maximum-likelihood approach that efficiently estimates the slow uphill kinetics utilizing pre-computed equilibrium populations of all states. We expect that these new algorithms and their combinations may boost investigations of important multi-body biological processes via the efficient construction of MSMs.

Abnormal ST segment is an important parameter for the diagnosis of myocardial ischemia and other heart diseases. As most abnormal ST segments sustain for only a few seconds, it is impractical for the doctors to detect and classify abnormal ones manually on time. Even though many ST segment classification algorithms are proposed to meet the rising demand of automatic myocardial ischemia diagnosis, they are often with lower recognition rate. The aim of this study is to detect abnormal ST segments precisely and classify them into more categories, and thus provide more detailed category information to help the clinicians make decisions. This study sums up ten common abnormal ST segments according to the clinical ECG records and proposes a morphological classification algorithm of ST segment based on multi-features. This algorithm consists of two parts: Feature points extraction and ST segment classification. In the first part, R wave is detected by using the 2B-spline wavelet transform, and mode-filtering method and morphological characteristics are used for other feature points extraction. In the ST segment classification process, ST segment level, variance, slope value, number of convex/concave points and other feature parameters are employed to classify the ST segment. This algorithm can classify abnormal ST segments into ten categories above. We evaluated the performance of the proposed algorithm based on ECG data in the European ST-T database. The global recognition rate of 92.7% and the best accuracy of 97% demonstrated the effectiveness of the proposed solution.

Speech recognition (SR) has been improved significantly by artificial neural networks (ANNs), but ANNs have the drawbacks of biologically implausibility and excessive power consumption because of the nonlocal transfer of real-valued errors and weights. While spiking neural networks (SNNs) have the potential to solve these drawbacks of ANNs due to their efficient spike communication and their natural way to utilize kinds of synaptic plasticity rules found in brain for weight modification. However, existing SNN models for SR either had bad performance, or were trained in biologically implausible ways. In this paper, we present a biologically inspired convolutional SNN model for SR. The network adopts the time-to-first-spike coding scheme for fast and efficient information processing. A biological learning rule, spike-timing-dependent plasticity (STDP), is used to adjust the synaptic weights of convolutional neurons to form receptive fields in an unsupervised way. In the convolutional structure, the strategy of local weight sharing is introduced and could lead to better feature extraction of speech signals than global weight sharing. We first evaluated the SNN model with a linear support vector machine (SVM) on the TIDIGITS dataset and it got the performance of 97.5%, comparable to the best results of ANNs. Deep analysis on network outputs showed that, not only are the output data more linearly separable, but they also have fewer dimensions and become sparse. To further confirm the validity of our model, we trained it on a more difficult recognition task based on the TIMIT dataset, and it got a high performance of 93.8%. Moreover, a linear spike-based classifier-tempotron-can also achieve high accuracies very close to that of SVM on both the two tasks. These demonstrate that an STDP-based convolutional SNN model equipped with local weight sharing and temporal coding is capable of solving the SR task accurately and efficiently.

Markov State Model (MSM) has become a popular approach to study the conformational dynamics of complex biological systems in recent years. Built upon a large number of short molecular dynamics simulation trajectories, MSM is able to predict the long time scale dynamics of complex systems. However, to achieve Markovianity, an MSM often contains hundreds or thousands of states (microstates), hindering human interpretation of the underlying system mechanism. One way to reduce the number of states is to lump kinetically similar states together and thus coarse-grain the microstates into macrostates. In this work, we introduce a probabilistic lumping algorithm, the Gibbs lumping algorithm, to assign a probability to any given kinetic lumping using the Bayesian inference. In our algorithm, the transitions among kinetically distinct macrostates are modeled by Poisson processes, which will well reflect the separation of time scales in the underlying free energy landscape of biomolecules. Furthermore, to facilitate the search for the optimal kinetic lumping (i.e., the lumped model with the highest probability), a Gibbs sampling algorithm is introduced. To demonstrate the power of our new method, we apply it to three systems: a 2D potential, alanine dipeptide, and a WW protein domain. In comparison with six other popular lumping algorithms, we show that our method can persistently produce the lumped macrostate model with the highest probability as well as the largest metastability. We anticipate that our Gibbs lumping algorithm holds great promise to be widely applied to investigate conformational changes in biological macromolecules.

Part of understanding a molecule’s conformational dynamics is mapping out the dominant metastable, or long lived, states that it occupies. Once identified, the rates for transitioning between these states may then be determined in order to create a complete model of the system’s conformational dynamics. Here we describe the use of the MSMBuilder package (now available at https://simtk.org/home/msmbuilder/) to build Markov State Models (MSMs) to identify the metastable states from Generalized Ensemble (GE) simulations, as well as other simulation datasets. Besides building MSMs, the code also includes tools for model evaluation and visualization.

The conformational dynamics of multibody systems plays crucial roles in many important problems. Markov state models (MSMs) are powerful kinetic network models that can predict long-time-scale dynamics using many short molecular dynamics simulations. Although MSMs have been successfully applied to conformational changes of individual proteins, the analysis of multibody systems is still a challenge because of the complexity of the dynamics that occur on a mixture of drastically different time scales. In this work, we have developed a new algorithm, automatic state partitioning for multibody systems (APM), for constructing MSMs to elucidate the conformational dynamics of multibody systems. The APM algorithm effectively addresses different time scales in the multibody systems by directly incorporating dynamics into geometric clustering when identifying the metastable conformational states. We have applied the APM algorithm to a 2D potential that can mimic a protein-ligand binding system and the aggregation of two hydrophobic particles in water and have shown that it can yield tremendous enhancements in the computational efficiency of MSM construction and the accuracy of the models.

© Springer International Publishing Switzerland 2015. Non-negative matrix factorization (NMF) has been a popular data analysis tool and has been widely applied in computer vision. However, conventional NMF methods cannot adaptively learn grouping structure froma dataset.This paper proposes a non-negative low-rank and group-sparse matrix factorization (NLRGS) method to overcome this deficiency. Particularly, NLRGS captures the relationships among examples by constraining rank of the coefficients meanwhile identifies the grouping structure via group sparsity regularization. By both constraints, NLRGS boosts NMF in both classification and clustering. However, NLRGS is difficult to be optimized because it needs to deal with the low-rank constraint. To relax such hard constraint, we approximate the low-rank constraint with the nuclear norm and then develop an optimization algorithm for NLRGS in the frame of augmented Lagrangian method(ALM). Experimental results of both face recognition and clustering on four popular face datasets demonstrate the effectiveness of NLRGS in quantities.

Principal component analysis (PCA) projects data on the directions with maximal variances. Since PCA is quite effective in dimension reduction, it has been widely used in computer vision. However, conventional PCA suffers from following deficiencies: 1) it spends much computational costs to handle high-dimensional data, and 2) it cannot reveal the nonlinear relationship among different features of data. To overcome these deficiencies, this paper proposes an efficient two-dimensional Euler PCA (2D-ePCA) algorithm. Particularly, 2D-ePCA learns projection matrix on the 2D pixel matrix of each image without reshaping it into 1D long vector, and uncovers nonlinear relationships among features by mapping data onto complex representation. Since such 2D complex representation induces much smaller kernel matrix and principal subspaces, 2D-ePCA costs much less computational overheads than Euler PCA on large-scale dataset. Experimental results on popular face datasets show that 2D-ePCA outperforms the representative algorithms in terms of accuracy, computational overhead, and robustness.

Molecular recognition, the process by which biological macromolecules selectively bind, plays an important role in many biological processes. Molecular simulations hold great potential to reveal the chemical details of molecular recognition and to complement experiments. However, it is challenging to reconstruct the binding process for two-body systems like protein-ligand complexes because the system’s dynamics occurs on significantly different timescales due to several physical processes involved, such as diffusion, local interactions and conformational changes. In this chapter, we review some recent progress on applying Markov state models (MSMs) to two-body systems. Emphasis is placed on the value of projecting dynamics onto collective reaction coordinates and treating the ligand dynamics with different resolution models depending on the proximity of the protein and ligand. We also discuss some future directions on constructing MSMs to investigate molecular recognition processes.

Conformational changes of proteins are an*Author contributed equally with all other contributors. essential part of many biological processes such as: protein folding, ligand binding, signal transduction, allostery, and enzymatic catalysis. Molecular dynamics (MD) simulations can describe the dynamics of molecules at atomic detail, therefore providing a much higher temporal and spatial resolution than most experimental techniques. Although MD simulations have been widely applied to study protein dynamics, the timescales accessible by conventional MD methods are usually limited to timescales that are orders of magnitude shorter than the conformational changes relevant for most biological functions. During the past decades great effort has been devoted to the development of theoretical methods that may enhance the conformational sampling. In recent years, it has been shown that the statistical mechanics framework provided by discrete-state and -time Markov State Models (MSMs) can predict long timescale dynamics from a pool of short MD simulations. In this chapter we provide the readers an account of the basic theory and selected applications of MSMs. We will first introduce the general concepts behind MSMs, and then describe the existing procedures for the construction of MSMs. This will be followed by the discussions of the challenges of constructing and validating MSMs, Finally, we will employ two biologically-relevant systems, the RNA polymerase and the LAO-protein, to illustrate the application of Markov State Models to elucidate the molecular mechanisms of complex conformational changes at biologically relevant timescales.

Stimulus information is encoded in the spatial-temporal structures of external inputs to the neural system. The ability to extract the temporal information of inputs is fundamental to brain function. It has been found that the neural system can memorize temporal intervals of visual inputs in the order of seconds. Here we investigate whether the intrinsic dynamics of a large-size neural circuit alone can achieve this goal. The network models we consider have scale-free topology and the property that hub neurons are difficult to be activated. The latter is implemented by either including abundant electrical synapses between neurons or considering chemical synapses whose efficacy decreases with the connectivity of the postsynaptic neuron. We find that hub neurons trigger synchronous firing across the network, loops formed by low-degree neurons determine the rhythm of synchronous firing, and the hardness of exciting hub neurons avoids epileptic firing of the network. Our model successfully reproduces the experimentally observed rhythmic synchronous firing with long periods and supports the notion that the neural system can process temporal information through the dynamics of local circuits in a distributed way.

Markov models and master equations are a powerful means of modeling dynamic processes like protein conformational changes. However, these models are often difficult to understand because of the enormous number of components and connections between them. Therefore, a variety of methods have been developed to facilitate understanding by coarse-graining these complex models. Here, we employ Bayesian model comparison to determine which of these coarse-graining methods provides the models that are most faithful to the original set of states. We find that the Bayesian agglomerative clustering engine and the hierarchical Nystrom expansion graph (HNEG) typically provide the best performance. Surprisingly, the original Perron cluster cluster analysis (PCCA) method often provides the next best results, outperforming the newer PCCA+ method and the most probable paths algorithm. We also show that the differences between the models are qualitatively significant, rather than being minor shifts in the boundaries between states. The performance of the methods correlates well with the entropy of the resulting coarse-grainings, suggesting that finding states with more similar populations (i.e., avoiding low population states that may just be noise) gives better results.

Markov state models (MSMs) have become a popular approach for investigating the conformational dynamics of proteins and other biomolecules. MSMs are typically built from numerous molecular dynamics simulations by dividing the sampled configurations into a large number of microstates based on geometric criteria. The resulting microstate model can then be coarse-grained into a more understandable macrostate model by lumping together rapidly mixing microstates into larger, metastable aggregates. However, finite sampling often results in the creation of many poorly sampled microstates. During coarse-graining, these states are mistakenly identified as being kinetically important because transitions to/from them appear to be slow. In this paper, we propose a formalism based on an algebraic principle for matrix approximation, i.e., the Nystrom method, to deal with such poorly sampled microstates. Our scheme builds a hierarchy of microstates from high to low populations and progressively applies spectral clustering on sets of microstates within each level of the hierarchy. It helps spectral clustering identify metastable aggregates with highly populated microstates rather than being distracted by lowly populated states. We demonstrate the ability of this algorithm to discover the major metastable states on two model systems, the alanine dipeptide and trpzip2 peptide.

Non-negative matrix factorization (NMF) learns the latent semantic space more direct and reliable than the latent semantic indexing (LSI) and the spectral clustering methods, thus performs well in document clustering. Recently, semi-supervised NMF such as N2S2L, CNMF and unsupervised method such as GNMF significantly improve the face recognition performance, but they are designed for classification. In this paper, we combine both geometric structure and label information with NMF under the non-negative patch alignment framework (NPAF) to form SS-NPAF. Due to this combination, it greatly improves the clustering performance. To optimize SS-NPAF, we apply the well-known projected gradient method to overcome the slow convergence problem of the mostly used multiplicative update rule. Experimental results on two popular document datasets, i.e., Reuters21578 and TDT-2, show that SS-NPAF outperforms the representative SS-NMF algorithms.