Distributed computation and reconfiguration in actively dynamic networks

We study here systems of distributed entities that can actively modify their communication network. This gives rise to distributed algorithms that apart from communication can also exploit network reconfiguration to carry out a given task. Also, the distributed task itself may now require a global reconfiguration from a given initial network $$G_s$$ G s to a target network $$G_f$$ G f from a desirable family of networks. To formally capture costs associated with creating and maintaining connections, we define three edge-complexity measures: the total edge activations, the maximum activated edges per round, and the maximum activated degree of a node. We give (poly)log(n) time algorithms for the task of transforming any $$G_s$$ G s into a $$G_f$$ G f of diameter (poly)log(n), while minimizing the edge-complexity. Our main lower bound shows that $$\varOmega (n)$$ Ω ( n ) total edge activations and $$\varOmega (n/\log n)$$ Ω ( n / log n ) activations per round must be paid by any algorithm (even centralized) that achieves an optimum of $$\varTheta (\log n)$$ Θ ( log n ) rounds. We give three distributed algorithms for our general task. The first runs in $$O(\log n)$$ O ( log n ) time, with at most 2n active edges per round, a total of $$O(n\log n)$$ O ( n log n ) edge activations, a maximum degree $$n-1$$ n - 1 , and a target network of diameter 2. The second achieves bounded degree by paying an additional logarithmic factor in time and in total edge activations. It gives a target network of diameter $$O(\log n)$$ O ( log n ) and uses O(n) active edges per round. Our third algorithm shows that if we slightly increase the maximum degree to polylog(n) then we can achieve $$o(\log ^2 n)$$ o ( log 2 n ) running time.

Detection of Anomalous Diffusion with Deep Residual Networks

Identification of the diffusion type of molecules in living cells is crucial to deduct their driving forces and hence to get insight into the characteristics of the cells. In this paper, deep residual networks have been used to classify the trajectories of molecules. We started from the well known ResNet architecture, developed for image classification, and carried out a series of numerical experiments to adapt it to detection of diffusion modes. We managed to find a model that has a better accuracy than the initial network, but contains only a small fraction of its parameters. The reduced size significantly shortened the training time of the model. Moreover, the resulting network has less tendency to overfitting and generalizes better to unseen data.

Incremental Minimum Flow Algorithms

There are various situations in which real-world problems can be modeled and solved as minimum flow problems. Sometimes, in these situations, minor data changes may occur, leading to corresponding changes of the networks in which the practical problems are modeled as flow problems, such as slight variations in capacity or lower bound. For instance, the capacity or the lower bound of an arc may increase or decrease in time, leaving one with no other choice than finding the new minimum network flow. Given both the various ways in which the networks can be changed and the high frequency of these changes, it is desirable to find as fast a computation method for minimum flow as possible. This paper is focused on the cases that concern increasing and decreasing the capacity or the lower bound of an arc. For these cases, both the minimum flow algorithms and the dynamic minimum flow algorithms that are already known are inefficient. Our incremental algorithms for determining minimum flow in the modified network are more efficient than both the above-mentioned types of algorithms. The proposed method starts from the initial network minimum flow and solves the minimum flow problem in a significantly faster way than recalculating the new network minimum flow starting from scratch.

Actin filament alignment causes mechanical hysteresis in cross-linked networks

Cells dynamically control their material properties through remodeling of the actin cytoskeleton, an assembly of cross-linked networks and bundles formed from the biopolymer actin. We recently found that cross-linked networks of actin filaments reconstituted in vitro can exhibit adaptive behavior and thus serve as a model system to understand the underlying mechanisms of mechanical adaptation of the cytoskeleton. In these networks, training, in the form of applied shear stress, can induce asymmetry in the nonlinear elasticity. Here, we explore control over this mechanical hysteresis by tuning the concentration and mechanical properties of cross-linking proteins in both experimental and simulated networks. We find that this effect depends on two conditions: the initial network must exhibit nonlinear strain stiffening, and filaments in the network must be able to reorient during training. Hysteresis depends strongly and non-monotonically on cross-linker concentration, with a peak at moderate concentrations. In contrast, at low concentrations, where the network does not strain stiffen, or at high concentrations, where filaments are less able to rearrange, there is little response to training. Additionally, we investigate the effect of changing cross-linker properties and find that longer or more flexible cross-linkers enhance hysteresis. Remarkably plotting hysteresis against alignment after training yields a single curve regardless of the physical properties or concentration of the cross-linkers.

Automated Classification and Segmentation in Colorectal Images Based on Self-Paced Transfer Network

Colorectal imaging improves on diagnosis of colorectal diseases by providing colorectal images. Manual diagnosis of colorectal disease is labor-intensive and time-consuming. In this paper, we present a method for automatic colorectal disease classification and segmentation. Because of label unbalanced and difficult colorectal data, the classification based on self-paced transfer VGG network (STVGG) is proposed. ImageNet pretraining network parameters are transferred to VGG network with training colorectal data to acquire good initial network performance. And self-paced learning is used to optimize the network so that the classification performance of label unbalanced and difficult samples is improved. In order to assist the colonoscopist to accurately determine whether the polyp needs surgical resection, feature of trained STVGG model is shared to Unet segmentation network as the encoder part and to avoid repeat learning of polyp segmentation model. The experimental results on 3061 colorectal images illustrated that the proposed method obtained higher classification accuracy (96%) and segmentation performance compared with a few other methods. The polyp can be segmented accurately from around tissues by the proposed method. The segmentation results underpin the potential of deep learning methods for assisting colonoscopist in identifying polyps and enabling timely resection of these polyps at an early stage.

An Efficient Topology Discovery Protocol with Node ID Assignment Based on Layered Model for Underwater Acoustic Networks

Underwater acoustic networks are widely used in survey missions and environmental monitoring. When an underwater acoustic network (UAN) is deployed in a marine region or two UANs merge, each node hardly knows the entire network and may not have a unique node ID. Therefore, a network topology discovery protocol that can complete node discovery, link discovery, and node ID assignment are necessary and important. Considering the limited node energy and long propagation delay in UANs, it is challenging to obtain the network topology with reduced overheads and a short delay in this initial network state. In this paper, an efficient topology discovery protocol (ETDP) is proposed to achieve adaptive node ID assignment and topology discovery simultaneously. To avoiding packet collision in this initial network state, ETDP controls the transmission of topology discovery (TD) packets, based on a local timer, and divides the network into different layers to make nodes transmit TD packets orderly. Exploiting the received TD packets, each node could obtain the network topology and assign its node ID independently. Simulation results show that ETDP completes network topology discovery for all nodes in the network with significantly reduced energy consumption and short delay; meanwhile, it assigns the shortest unique IDs to all nodes with reduced overheads.

Complex Network Minority Game Model for the Financial Market Modeling and Simulation

This paper proposes a new financial market model based on the analysis of the minority game model. The agent in this model forms a network through information sharing, and the agent uses the minority game model to realize the evolution of the system. To better describe the financial market, we also adopt a prior connection strategy for the model. The network formed by the agent has the characteristics of a scale-free network, and as the initial network connection probability increases, the growth rate of the corresponding agent’s average connection degree increases and then decreases after reaching the peak.

Pseudo-Darwinian evolution of physical flows in complex networks

The evolution of complex transport networks is investigated under three strategies of link removal: random, intentional attack and “Pseudo-Darwinian” strategy. At each evolution step and regarding the selected strategy, one removes either a randomly chosen link, or the link carrying the strongest flux, or the link with the weakest flux, respectively. We study how the network structure and the total flux between randomly chosen source and drain nodes evolve. We discover a universal power-law decrease of the total flux, followed by an abrupt transport collapse. The time of collapse is shown to be determined by the average number of links per node in the initial network, highlighting the importance of this network property for ensuring safe and robust transport against random failures, intentional attacks and maintenance cost optimizations.

IMPROVING DISPARITY ESTIMATION BASED ON RESIDUAL COST VOLUME AND RECONSTRUCTION ERROR VOLUME

Recently, great progress has been made in formulating dense disparity estimation as a pixel-wise learning task to be solved by deep convolutional neural networks. However, most resulting pixel-wise disparity maps only show little detail for small structures. In this paper, we propose a two-stage architecture: we first learn initial disparities using an initial network, and then employ a disparity refinement network, guided by the initial results, which directly learns disparity corrections. Based on the initial disparities, we construct a residual cost volume between shared left and right feature maps in a potential disparity residual interval, which can capture more detailed context information. Then, the right feature map is warped with the initial disparity and a reconstruction error volume is constructed between the warped right feature map and the original left feature map, which provides a measure of correctness of the initial disparities. The main contribution of this paper is to combine the residual cost volume and the reconstruction error volume to guide training of the refinement network. We use a shallow encoder-decoder module in the refinement network and do learning from coarse to fine, which simplifies the learning problem. We evaluate our method on several challenging stereo datasets. Experimental results demonstrate that our refinement network can significantly improve the overall accuracy by reducing the estimation error by 30% compared with our initial network. Moreover, our network also achieves competitive performance compared with other CNN-based methods.