The softmax function needs to be frequently used in the multi-head attention layer of Transformer networks. Compared to DNNs and other networks, Transformers have higher computational complexity, requiring higher accuracy and hardware performance for softmax function calculations. Therefore, we propose mixed-base softmax (MBS) for the first time for the approximation of the softmax function. This method combines exponential functions with bases of 2 and 4, which is advantageous for hardware implementation. MBS has a high similarity to the softmax function and demonstrates advanced performance during inference in Transformer network. Through algorithm transformation and hardware optimization, we have designed a low-complexity and highly parallel hardware architecture, which only occupies few additional hardware resources compared to base-2 softmax but achieves higher accuracy. Experimental results show that, under TSMC 90nm CMOS technology at the frequency of 0.5 GHz, our design can achieve the efficiency of 236.18 Gps/(mm2⋅mW) with the area of 4234 μm2. Furthermore, MBS exhibits higher computational accuracy and inference precision compared with base-2 softmax.

This paper explores efficient 3G/4G budget utilization in mobile sensing applications. Distinct from previous research work that either relied on limited WiFi access points or assumed the availability of unlimited 3G/4G communication capability, we offer a more practical mobile sensing system that leverages potential 3G/4G budgets that participants contribute at will, and uses it efficiently customized for the needs of multiple mobile sensing applications with heterogeneous sensitivity to environmental changes. We address the challenge that the information of data generation and WiFi encounters is not a priori knowledge, and propose an online decision making algorithm that takes advantage of participants' historical data. Three typical mobile sensing applications, vehicular application, mobile health and video sharing application are explored. Experimental results demonstrate that our proposed algorithms lead to significantly better system performance compared to alternative solutions for both applications.

This paper explores direct phone-to-phone communication (via WiFi interface) among vehicles to support mobile sensing applications. Direct communication among drivers' phones is important in improving data collection efficiency and sharing participatory sensing information in an inexpensive manner. We design a practical and optimized communication mechanism for direct phone-to-phone data transfer among drivers' phones that strategically enables phone-to-phone and/or phone-to-WiFiAP communications by optimally toggles the phone between the normal client and the hotspot modes. We take advantage of the WiFi hotspot functionality on smartphones, and hence require neither involvement of participants nor changes to existing wireless infrastructure and protocols. An analytical model is established to optimize toggling between client and hotspot modes for optimal system efficiency. We fully implement this system on off-the-shelf Google Galaxy Nexus and Nexus S phones. Through a 35-vehicle 2-month deployment study, as well as simulation experiments using the real-world T-drive 9,211-taxicab dataset, we show that our solution significantly reduces data transfer delay time and maintains over 80% efficiency under varying system parameters. We even achieve 90% for parameter settings of the latest smartphones.

The primary goal of breadcrumb trail sensor networks is to transmit in real-time users' physiological parameters that measure life-critical functions to an incident commander through reliable multihop communication. In applications using breadcrumb solutions, there are often many users working together, and this creates a well-known body shadowing effect (BSE). In this article, we first measure the characteristics of body shadowing for 2.4GHz sensor nodes. Our empirical results show that the body shadowing effect leads to severe packet loss and consequently very poor real-time performance. Then we develop a novel Intentional Forwarding solution. This solution accurately detects the shadowing mode and enables selected neighbors to forward data packets. Experimental results from a fully implemented testbed demonstrate that Intentional Forwarding is able to improve the end-to-end average packet delivery ratio (PDR) from 58% to 93% and worst-case PDR from 45% to 85%, and is able to meet soft real-time requirements even under severe body shadowing problems.

Breadcrumb systems (BCS) aid first responders by communicating their physiological parameters to remotely located base stations. In this paper, we describe the design, implementation, and evaluation of an automatic and robust multi-user breadcrumb system for indoor first response applications. Our solution includes a breadcrumb dispenser with a link estimator that is used to decide when to deploy breadcrumbs to maintain reliable wireless connectivity. The solution includes accounting for realities of buildings and dispensing such as the height difference between where the dispenser is worn and the floor where the dispensed nodes are found. We also include adaptive power management to maintain link quality over time. Moreover, we propose UF, a distributed cooperative deployment algorithm, to achieve longer breadcrumb chain lengths while maintaining fairness and high system reliability via selecting appropriate benefit and cost functions. We deployed and evaluated our system in real buildings with several different first responder mobility patterns. Experimental results from our study show that compared to the state of the art solution , our breadcrumb system achieves 200 percent link redundancy with only 23 percent additional deployed nodes. Our deployed breadcrumb chain can achieve 90 percent PRR when one node fails in the chain. In addition, by applying the UF coordination algorithm, the system can maintain connectivity for up to 87 percent longer distances than baseline greedy coordination approach while maintaining 96 percent packet delivery ratio.

This paper explores efficient 3G budget utilization in mobile participatory sensing applications. 1 Distinct from previous research work that either rely on limited WiFi access points or assume the availability of unlimited 3G communication capability, we offer a more practical participatory sensing system that leverages potential 3G budgets that participants contribute at will, and uses it efficiently customized for the needs of multiple participatory sensing applications with heterogeneous sensitivity to environmental changes. We address the challenge that the information of data generation and WiFi encounters is not a priori knowledge, and propose an online decision making algorithm that takes advantage of participants' historical data. We also develop a heuristic algorithm to consume less energy and reduce the storage overhead while maintaining efficient 3G budget utilization. Experimental results from a 30-participant deployment demonstrate that, even when the budget is as small as 2.5% of a popular data plan, these two algorithms achieve higher utility of uploaded data compared to the baseline solution, especially, they increase the utility of received data by 151.4% and 137.8% for those sensitive applications.

In this demo, a learning system, called Metis, is presented that extrapolates missing pieces in participatory sensing data. The work addresses the challenge of incomplete coverage in participatory sensing applications, where lack of complete control over participant mobility and sensing patterns may create coverage gaps in space and in time. Metis learns the underlying spatiotemporal patterns of the measured phenomenon from available incomplete observations, and uses these patterns to infer missing data. We describe the overall system design and demonstrate the system using data collected during the New York City gas crisis in the aftermath of Hurricane Sandy.

The primary goal of breadcrumb trail sensor networks is to transmit in real-time users' physiological parameters that measure life critical functions to an incident commander through reliable multihop communication. In applications using breadcrumb solutions, there are often many users working together, and this creates a well-known body shadowing effect (BSE). In this paper, we first measure the characteristics of body shadowing for 2.4 GHz sensor nodes. Our empirical results show that the body shadowing effect leads to severe packet loss, and consequently very poor real-time performance. Then we develop a novel Intentional Forwarding solution. This solution accurately detects the shadowing mode and enables selected neighbors to forward data packets. Experimental results from a fully implemented testbed demonstrate that Intentional Forwarding is able to improve the end-to-end average packet delivery ratio (PDR) from 58% to 93% and worst-case PDR from 45% to 85%, and meet soft real-time requirements even under severe body shadowing problems.

This paper describes a novel group-based programming abstraction called a “Bundle” for cyber-physical systems (CPS). Similar to other programming abstractions, a Bundle creates logical collections of sensing devices. However, previous abstractions were focused on wireless sensor networks (WSNs) and did not address key aspects of CPS. Bundles elevate the programming domain from a single WSN to complex systems of systems by allowing the programming of applications involving multiple CPSs that are controlled by different administrative domains and support mobility both within and across CPSs. Bundles can seamlessly group not only sensors, but also actuators which constitute an important part of CPS. They enable programming in a multiuser environment with fine grained access right control and conflict resolution mechanism. Bundles support heterogeneous devices, such as motes, PDAs, laptops, and actuators according to the applications' requirements. They allow different applications to simultaneously use the same sensors and actuators. Bundles facilitate feedback control mechanisms by dynamic membership update and requirements reconfiguration based on feedback from the current members. The Bundle abstraction is implemented in Java which ensures ease and conciseness of programming. We present the design and implementation details of Bundles as well as a performance evaluation using 32 applications written with Bundles. This set includes across-network applications that have sophisticated sensing and actuation logic, mobile nodes that are heterogeneous, and feedback control mechanisms. Each of these applications is programmed in less than 60 lines of code.

Previous localization solutions in wireless sensor networks mainly focus on using various techniques to estimate node positions. In this paper, we argue that quantifying the uncertainty of these estimates is equally important in practice. By using the quantitative uncertainty of measurements and estimates, we can derive more accurate estimates by better fusing the measurements, provide confidence information for confidence-based applications, and know how to select the best anchor nodes so as to minimize the total mean square errors of the whole network. This paper quantifies the estimation uncertainty as an error covariance matrix, and presents an efficient incremental centralized algorithm---INOVA and a decentralized algorithm---OSE-COV for calculating the error covariance matrix. Furthermore, we present how to use the error covariance matrix to infer the confidence region of each node's estimate, and provide an optimal strategy for the anchor selection problem. Extensive simulation results show that INOVA significantly improves the computation efficiency when the network changes dynamically; the confidence region inference is accurate when the measurement number to node number ratio is more than 2; and the optimal anchor selection strategy reduces the total mean square error by four times as much as the variation-based algorithm in best case.

Breadcrumb systems (BCS) aid first responders by communicating their physiological parameters to remotely located base stations. However, state-of-the-art research only focuses on deploying breadcrumb systems on the assumption of uncoordinated users, which is inefficient. In this paper, we present the first design, implementation, and evaluation of reliable multiuser breadcrumb systems (MUBCS) which exploits efficient and automatic coordination among system users to achieve better utilization of limited breadcrumbs. We propose UF, a distributed cooperative deployment algorithm, to achieve longer breadcrumb chain length while maintaining fairness and high system reliability via selecting appropriate benefit and cost functions. UF also requires no prior assumptions about users' mobility models, making the design practical for real applications. We deployed and evaluated our system in real buildings with several different first responder mobility patterns. Experimental results indicate that this approach can maintain connectivity for up to 87% longer distances than baseline greedy coordination approach while maintaining 96% packet delivery ratio.

Breadcrumb systems (BCS) have been proposed to aid firefighters inside buildings by communicating their physiological parameters to base stations outside the buildings. In this paper, we describe the design, implementation and evaluation of an automatic and robust breadcrumb system for firefighter applications. Our solution includes a breadcrumb dispenser with an optimized link estimator that is used to decide when to deploy breadcrumbs to maintain reliable wireless connectivity. The solution includes accounting for realities of buildings and dispensing such as the height difference between where the dispenser is worn and the floor where the dispensed nodes are found. We also include adaptive power management to maintain link quality over time.

This paper describes the design and implementation of a pervasive computing framework, named Physicalnet. Essentially, Physicalnet is a generic paradigm for managing and programming world-wide distributed heterogeneous sensor and actuator resources in a multi-user and multi-network environment. Using a four-tier light-weight service oriented architecture, Physicalnet enables global uniform access to heterogeneous resources and decouples applications from particular resources, locations and networks. Through a negotiator module, it allows a large number of applications to concurrently execute on the same resources and to span multiple physical networks and logical administrative domains. By providing a fine-grained use-based access rights control and conflict resolution mechanism, Physicalnet not only ensures owners having total control of sharing and protecting their resources, but also dramatically increases the number of applications that can concurrently execute on the devices. Furthermore, Physicalnet supports resource dynamic location-aware mobility, application run-time reconfigurability and on-the-fly access rights specification. To quantify the performance, we evaluate Physicalnet based on memory usage, the number of concurrent applications, and dynamic responsiveness. The results show Physicalnet has excellent performance, but low overheads.

Physicalnet is a fully implemented middleware for wireless networks of sensors and actuators (WNSAs). Based on a lightweight service oriented architecture (SOA), Physicalnet allows the users to create administrative domains, in which owners of the nodes can share and protect their resources so that multiple applications can simultaneously use the same sensors and actuators with conflict resolution mechanisms. By using a general programming abstraction, named bundle, Physicalnet allows programming across networks and administrative domains, and has a good support for node mobility. Other features of Physicalnet include using the Java language, supporting heterogeneous devices, and providing a chain of user tools.