Precise quantification and localization of tracer-labeled neurons are essential for unraveling brain connectivity patterns and constructing a mesoscopic brain connectome atlas in macaques. However, methodological challenges and limitations in dataset development have impeded this scientific progress. This work introduced the Macaque Fluorescently Labeled Neurons (MFN) dataset, derived from retrograde tracing on three rhesus macaques. The dataset, meticulously annotated by six specialists, includes 1,600 images and 33,411 high-quality neuron annotations. Leveraging this dataset, we developed a Dense Convolutional Attention U-Net (DAUNet) cell counting model. By integrating Dense Convolutional blocks and a multi-scale attention module, the model exhibits robust feature extraction and representation capabilities while maintaining low complexity. On the MFN dataset, DAUNet achieved a Mean Absolute Error of 0.97 for cell counting and an F1-score of 96.29% for cell localization, outperforming several benchmark models. Extensive validation across four additional public datasets demonstrated the robust generalization ability of the model. Furthermore, the trained model was applied to quantify labeled neurons of a macaque brain, mapping the input connectivity patterns of two adjacent subregions in the lateral prefrontal cortex. This work provides a training dataset and algorithmic resource that advances mesoscopic brain connectivity research in macaques. The MFN dataset and source code are available at https://github.com/Gendwar/DAUnet.

We investigate a transport layer protocol design that integrates 3G and WiFi networks, specifically targeting vehicular mobility. The goal is to move load from the expensive 3G network to the less expensive WiFi network without hurting the user experience. As the test platform we choose a nationwide 3G network and a commercially operated metro-scale WiFi network. We exploit the often complementary characteristics of these networks for a hybrid design at the transport layer. To this end, we modify the stock Linux SCTP implementation to support `striping' across multiple interfaces and the ability to handle frequent path failures and recovery in a seamless fashion. Instead of simply striping data over two network connections, we develop a utility and cost-based formulation that decides the right amount of load that can be put on the 3G network to maximize the user's benefit. We develop and experiment with a transport level scheduler to do this. We call the new SCTP design as oSCTP, meaning `SCTP to be used for offloading.' We demonstrate the effectiveness of oSCTP and show that it is able to deliver superior network throughput and user experience, while significantly reducing the load on the 3G network.

We perform a head-to-head comparison of the performance characteristics of a 3G network operated by a nation-wide provider and a metro-scale WiFi network operated by a commercial ISP, from the perspective of vehicular network access. Our experience shows that over a wide geographic region and under vehicular mobility, these networks exhibit very different throughput and coverage characteristics. WiFi has frequent disconnections even in a commercially operated, metro-scale deployment; but when connected, indeed delivers high throughputs even in a mobile scenario. The 3G network offers similar or lower throughputs in general, but provides excellent coverage and less throughput variability. The two network characteristics are often complementary. It is conceivable that these properties can be judiciously exploited for a hybrid network design where 3G data can be offloaded to WiFi for better performance and to reduce 3G network congestion and to lower costs.

In this paper, we address the problem of preserving generated data in a sensor network in case of node failures. We focus on the type of node failures that have explicit spatial shapes such as circles or rectangles (e.g., modeling a bomb attack or a river overflow). We consider two different schemes for introducing redundancy in the network, by simply replicating data or by using erasure codes, with the objective to minimize the communication cost incurred to build such data redundancy. We prove that the problem is NP-hard using either replication or coding. We design Oα-approximation centralized and distributed algorithms for the two redundancy schemes, where α is the "fatness" of the potential node failure events. Using erasure codes, data distribution can be handled in an efficient distributed manner. Simulation results show that by exploiting the spatial properties of the node failure patterns, one can substantially reduce the communication cost compared to the resilient data storage schemes in the prior literature.