Embedded systems were once built with clarity—each line of code grounded in an algorithm, each behavior traceable and explainable. But as AI models rapidly replace classical methods in sensing, scheduling, decision, and control, we have gained accuracy at the cost of trust. Today’s neural networks are black boxes, assembled by intuition or brute force, leaving us unable to explain, debug, or control their behavior. We argue this is not just a tooling issue, but a design flaw: explainability has long been an afterthought, with networks built first and interpreted later. We advocate a principled approach where networks are grounded in algorithms and designed with internal anchors—expected intermediate behaviors, invariants, or interpretable signals—that support debugging and interpretation. This paper presents Algorithm-Informed Neural Networks (AINN)—architectures shaped by algorithms as implicit inductive bias. By decomposing algorithms into logic blocks, we build modular networks that are easier to train, interpret, and debug. As proof of concept, we present two use cases—fall detectionkeyword spotting problems—showing how algorithmic structure improves training efficiency and enables effective debugging.

The emergence of Vision Transformer (ViT) models and their variants (e.g., Swin Transformers) are prevalent in recent years due to their higher accuracy in vision AI applications. However, their efficient execution in edge computing environments (e.g., mobile phones/embedded platforms) remains a challenge due to the heavy computational demands (both GPU cycles and GPU memory) of these large-sized models. To serve these models efficiently at the edge, we introduce a novel approach combining model slicing and smart batching to distribute workloads between resource-constrained client devices and powerful edge servers. Model slicing allows breaking a large model into smaller segments, called slices, and let the client execute a few initial but a variable number of slices (head slices) and a nearby edge server runs the rest of the slices (tail slices), smart batching enables the server to queue several requests from multiple clients batch together for inference leading to better GPU resource utilization. We propose two batching strategies at the server: one runs faster but requires higher GPU memory and the other one demands less memory with slight overhead of internal data movement. Experimental results show that our approach achieves inference time reductions of up to 67 % while maintaining high GPU utilization and demonstrates significant improvements in inference speed, showcasing the viability of this approach for distributed AI systems.

Retrieval-augmented generation (RAG) systems combine the strengths of language generation and information retrieval to power many real-world applications like chatbots. Use of RAG for understanding of videos is appealing but there are two critical limitations. One-time, upfront conversion of all content in large corpus of videos into text descriptions entails high processing times. Also, not all information in the rich video data is typically captured in the text descriptions. Since user queries are not known apriori, developing a system for video to text conversion and interactive querying of video data is challenging.

Edge Intelligence has received attention in the recent times for its potential towards improving responsiveness, reducing the cost of data transmission, enhancing security and privacy, and enabling autonomous decisions by edge devices. However, edge devices lack the power and compute resources necessary to execute most Al models. In this paper, we present FactionFormer, a novel method to deploy resource-intensive deep-learning models, such as vision transformers (ViT), on resource-constrained edge devices. Our method is based on a key observation: edge devices are often deployed in settings where they encounter only a subset of the classes that the resource-intensive Al model is trained to classify, and this subset changes across deployments. Therefore, we automatically identify this subset as a faction, devise on-the fly a bespoke resource-efficient ViT called a modelette for the faction, and set up an efficient processing pipeline consisting of a modelette on the device, a wireless network such as 5G, and the resource-intensive ViT model on an edge server, all of which work collaboratively to do the inference. For several ViT models pre-trained on benchmark datasets, FactionFormer’s modelettes are up to 4× smaller than the corresponding baseline models in terms of the number of parameters, and they can infer up to 2.5× faster than the baseline setup where every input is processed by the resource-intensive ViT on the edge server. Our work is the first of its kind to propose a device-edge collaborative inference framework where bespoke deep learning models for the device are automatically devised on-the-fly for most frequently encountered subset of classes.

Detection-driven real-time video analytics require continuous detection of objects contained in the video frames using deep learning models like YOLOV3, EfficientDet, etc. However, running these detectors on each and every frame in resource-constrained edge devices is computationally intensive. By taking the temporal correlation between consecutive video frames into account, we note that detection outputs tend to be overlapping in successive frames. Elimination of “similar” consecutive frames (the same set of objects with slightly offset bounding boxes) will lead to a negligible drop in performance while offering significant performance benefits by reducing overall computation and communication costs. The key technical questions are, therefore, (a) how to identify which frames to be processed by the object detector, and (b) how many successive frames can be skipped (called skip-length) once a frame is selected to be processed. The overall goal of the process is to keep the error due to skipping frames as small as possible. We introduce a novel error vs processing rate optimization problem with respect to the object detection task that balances between the error rate and the fraction of frames actually passed and processed. Subsequently, we propose an off-line Reinforcement Learning (RL)-based algorithm to determine these skip-lengths as a state-action policy of the RL agent from a recorded video and then deploy the agent online for live video streams. To this end, we develop FrameHopper, an edge-cloud collaborative video analytics framework, that runs a lightweight trained RL agent on the camera and passes filtered frames to the cloud/edge server where the object detection model runs for a set of applications. We have tested our approach on a number of live videos captured from real-life scenarios and show that FrameHopper processes only a handful of frames but produces detection results closer to the "oracle" solution and outperforms recent state-of-the-art solutions in most cases.

Design of multitasking deep learning models has mostly focused on improving the accuracy of the constituent tasks, but the challenges of efficiently deploying such models in a device-edge collaborative setup (that is common in 5G deployments) has not been investigated. Towards this end, in this paper, we propose an approach called Chimera1for training (done Offline) and deployment (done Online) of multitasking deep learning models that are splittable across the device and edge. In the offline phase, we train our multi-tasking setup such that features from a pre-trained model for one of the tasks (called the Primary task) are extracted and task-specific sub-models are trained to generate the other (Secondary) tasks' outputs through a knowledge distillation like training strategy to mimic the outputs of pre-trained models for the tasks. The task-specific sub-models are designed to be significantly lightweight than the original pre-trained models for the Secondary tasks. Once the sub-models are trained, during deployment, for given deployment context, characterized by the configurations, we search for the optimal (in terms of both model performance and cost) deployment strategy for the generated multitasking model, through finding one or multiple suitable layer(s) for splitting the model, so that inference workloads are distributed between the device and the edge server and the inference is done in a collaborative manner. Extensive experiments on benchmark computer vision tasks demonstrate that Chimera generates splittable multitasking models that are at least ~ 3 x parameter efficient than the existing such models, and the end-to-end device-edge collaborative inference becomes ~ 1.35 x faster with our choice of context-aware splitting decisions.

Training a neural network model from scratch is a computationally intensive operation. To alleviate this issue, researchers often employ "transfer learning" that transfers knowledge from a source data distribution to a target data distribution, instead of training the whole model from scratch. Typically, the last few layers of a pretrained convolutional neural network (CNN) are chosen for many transfer learning tasks where the outputs of those selected layers are combined to construct a feature space based on which a task-specific classification n etwork i s t rained o r fi ne-tuned. Th is arbitrary way of selecting layers, however, often fails to achieve the desired accuracy for the target task. What we need is an intelligent way of selecting layers from a pretrained model for a given task so that the additional overhead of successive training remains low. To this end, we propose a novel method, called TransJury, to find t he m ost s ignificant la yers from a pretrained mo del for transfer learning along with preserving the knowledge for the source domain. Through extensive experimentation on several target domain datasets, we show the supremacy of our approach in terms of lower training overhead and improved accuracy. By deploying MobileNet-v2, a lightweight CNN model pretrained on the ImageNet dataset on an edge device, we also discuss the future direction of this research.

In this paper, we address the challenges in supporting reliability and scalability in societal-scale notification systems that aim to reach large populations with customized alerts. We explore fault tolerance (FT) techniques in the context of Big Data Publish-Subscribe systems (BDPS), a scalable hierarchical architecture, that meshes big-data platforms (to store and operate on large volumes of data) with a distributed pub/sub broker network (to manage and communicate with a large number of end subscribers). The role of brokers in this architecture is critical since they serve to mediate interactions between subscribers and the backend big data system. We propose the REAPS (REliable Active Publish Subscribe) framework that can handle different classes of broker failures including randomized failures and geographically-correlated failures (as in a natural disaster). REAPS implements a low overhead fault tolerance service using a primary-backup approach; key features include the ability to exploit subscription similarity among brokers and techniques for quasi-active state replication to support fast recovery and delivery guarantees of notification services. We implement REAPS and conduct measurement studies on a prototype BDPS platform using real world usecases. We further evaluate REAPS under various failure scenarios to explore the scalability and performance of our proposed FT mechanisms via simulation studies.

In this paper, we propose a transfer-learning based model construction technique for the aerial scene classification problem. The core of our technique is a layer selection strategy, named ReLU-Based Feature Fusion (RBFF), that extracts feature maps from a pretrained CNN-based single-object image classification model, namely MobileNetV2, and constructs a model for the aerial scene classification task. RBFF stacks features extracted from the batch normalization layer of a few selected blocks of MobileNetV2, where the candidate blocks are selected based on the characteristics of the ReLU activation layers present in those blocks. The feature vector is then compressed into a low-dimensional feature space using dimension reduction algorithms on which we train a low-cost SVM classifier for the classification of the aerial images. We validate our choice of selected features based on the significance of the extracted features with respect to our classification pipeline. RBFF remarkably does not involve any training of the base CNN model except for a few parameters for the classifier, which makes the technique very cost-effective for practical deployments. The constructed model despite being lightweight outperforms several recently proposed models in terms of accuracy for a number of aerial scene datasets.

Detection-driven video analytics is resource hungry as it depends on running object detectors on video frames. Running an object detection engine (i.e., deep learning models such as YOLO and EfficientDet) for each frame makes video analytics pipelines difficult to achieve real-time processing. In this paper, we leverage selective processing of frames and batching of frames to reduce the overall cost of running detection models on live videos. We discuss several factors that hinder the real-time processing of detection-driven video analytics. We propose a system with configurable knobs and show how to achieve the stability of the system using a Lyapunov-based control strategy. In our setup, heterogeneous edge devices (e.g. mobile phones, cameras) stream videos to a low-resource edge server where frames are selectively processed in batches and the detection results are sent to the cloud or to the edge device for further application-aware processing. Preliminary results on controlling different knobs, such as frame skipping, frame size, and batch size show interesting insights to achieve real-time processing of multi-stream video streams with low overhead and low overall information loss.

The following topics are dealt with: Internet of Things; learning (artificial intelligence); cloud computing; mobile computing; smart cities; data privacy; convolutional neural nets; neural nets; cryptography; and data analysis.

Emerging applications in smart cities and communities require massive IoT deployments using sensors/actuators (things) that can enhance citizens' quality of life and public safety. However, budget constraints often lead to limited instrumentation and/or the use of low-cost sensors that are subject to drift and bias. This raises concerns of robustness and accuracy of the decisions made on uncertain data. To enable effective decision-making while fully exploiting the potential of low-cost sensors, we propose to send mobile units (e.g., trained personnel) equipped with high-quality (more expensive) and freshly-calibrated reference sensors so as to carry out calibration in the field. We design and implement an efficient cooperative approach to solve the calibration planning problem, which aims at minimizing the cost of the recurring calibration of multiple sensor types in the long-term operation. We propose a two-phase solution that consists of a sensor selection phase that minimizes the average cost of recurring calibration, and a path planning phase that minimizes the travel cost of multiple calibrators which have load constraints. We provide fast and effective heuristics for both phases. We further build a prototype that facilitates the mapping of the deployment field and provides navigation guidance to mobile calibrators. Extensive use-case-driven simulations show our proposed approach significantly reduces the average cost compared to naive approaches: up to 30% in a moderate-sized indoor case, and higher in outdoor cases depending on the scale.

Modern IoT devices are equipped with media-rich sensors that generate a heavy burden to local access networks. To improve the efficiency of data collection, we introduce the concept of "IoT zones" as geographically-correlated clusters of local IoT devices with well connected wireless networks that may have limited access to the Internet. We develop techniques to create a cost-effective data collection network using existing transportation fleets with predefined schedules to collect sensor data from IoT zones and upload them at locations with better network connectivity. Specifically, we provide solutions to the upload point placement and upload path planning problems given tradeoffs between collection quality, timing needs (QoS), and installation cost. We evaluate our approaches using a real-world bus network in Orange County, CA and study the applicability and efficiency of the proposed method as compared to several other approaches. The trace-driven simulations reveal that our best-performing algorithm: upload point selection (UPS) algorithm significantly outperforms others, e.g., in one of the scenarios with 160 total cost, it achieves sub-21 sec data transfer time (15+ times improvement), sub 3.2% late delivery ratio (about 12 times improvement), and above 96% data delivery ratio (about 50% improvement). In addition, it achieves the above performance without excessive installation cost: even when a cost limit of 640 is given, UPS algorithm opts for a solution with about 160 total cost (versus 640 from others).

In this paper, we propose a set of caching strategies for big active data (BAD) systems. BAD is a data management paradigm that allows ingestion of massive amount of data from heterogeneous sources, such as sensor data, social networks, web and crowdsourced data in a large data cluster consisting of many computing and storage nodes, and enables a very large number of end users to subscribe to those data items through declarative subscriptions. A set of distributed broker nodes connect these end users to the backend data cluster, manage their subscriptions and deliver the subscription results to the end users. Unlike the most traditional publish-subscribe systems that match subscriptions against a single stream of publications to generate notifications, BAD can match subscriptions across multiple publications (by leveraging storage in the backend) and thus can enrich notifications with a rich set of diverse contents. As the matched results are delivered to the end users through the brokers, the broker node caches the results for a while so that the subscribers can retrieve them with reduced latency. Interesting research questions arise in this context so as to determine which result objects to cache or drop when the cache becomes full (eviction-based caching) or to admit objects with an explicit expiration time indicating how much time they should reside in the cache (TTL based caching). To this end, we propose a set of caching strategies for the brokers and show that the schemes achieve varying degree of efficiency in terms of notification delivery in the BAD system. We evaluate our schemes via a prototype implementation and through detailed simulation studies.

In urban environments, mobile crowdsensing can be used to augment in-situ sensing deployments (e.g. for environmental and community monitoring) in a flexible and cost-efficient manner. The additional participation provided by crowdsensing enables improved data collection coverage and enhances timeliness of data delivery. However, as the number of participating devices/users increases, efficient management is required to handle the increased operational cost of the infrastructure and associated cloud services - exploiting spatiotemporal redundancy in sensing can help cost-efficient utilization of resources. In this paper, we develop solutions to exploit the mobility of the crowd and manage the sensing capability of participating devices to effectively meet application/user demands for hybrid urban sensing applications. Specifically, we address the spatiotemporal scheduling problem to create high-resolution maps (e.g. for pollution sensing) by developing a common framework to capture spatiotemporal impact of multiple sensor types that generate heterogeneous data at different levels of granularity. We develop an online scheduling approach that leverages the knowledge of device location and sensing capability to selectively activate nodes and sensors. We build a multi-sensor platform that enables data collection, data exchange, and node management. Prototype deployments in three different campus/community testbeds were instrumented for measurements. Traces collected from the testbeds are used to drive extensive large scale simulations. Results show that our proposed solution achieves improved data coverage and utility under data constraints with lower costs (30% fewer active nodes) than naive approaches.

We demonstrate a managed edge computing platform for Internet-of-Things (IoT) devices, which supports dynamic deployment of virtualized containers running distributed analytics. We build a model city, and install multiple Raspberry Pis as minions, and a mini PC as the master. Through the web dashboard on the master, we show how users can remotely monitor, manage, and upgrade the IoT analytics and devices. Multiple concrete IoT analytics, namely: (i) air quality monitor, (ii) sound classifier, and (iii) image recognizer are demonstrated. Several sample measurements on deployment speed, Quality-of- Service (QoS) achievements, and event-driven mechanisms are also carried out on the testbed.

In this paper, we propose techniques for dynamically adjusting heartbeat or keep-alive interval of long-lived TCP connections, particularly the ones that are used in push notification service in mobile platforms. When a device connects to a server using TCP, often times the connection is established through some sort of middle-box, such as NAT, proxy, firewall, and so on. When such a connection is idle for a long time, it may get torn down due to binding timeout of the middle-box. To keep the connection alive, the client device needs to send keep-alive packets through the connection when it is otherwise idle. To reduce resource consumption, the keep-alive packet should preferably be sent at the farthest possible time within the binding timeout. Due to varied settings of different network equipments, the binding timeout will not be identical in different networks. Hence, the heartbeat rate used in different networks should be changed dynamically. We propose a set of iterative probing techniques, namely binary, exponential, and composite search, that detect the middle-box binding timeout with varying degree of accuracy; and in the process, keeps improving the keep-alive interval used by the client device. We also analytically derive performance bounds of these techniques. To the best of our knowledge, ours is the first work that systematically studies several techniques to dynamically improve keep-alive interval. To this end, we run experiments in simulation as well as make a real implementation on android to demonstrate the proof-of-concept of the proposed schemes.

Modern IoT analytics are computational and data intensive. Existing analytics are mostly hosted in cloud data centers, and may suffer from high latency, network congestion, and privacy issues. In this paper, we design, implement, and evaluate a fog computing platform that runs analytics in a distributed way on multiple devices, including IoT devices, edge servers, and data-center servers. We focus on the core optimization problem: making deployment decisions to maximize the number of satisfied IoT analytics. We carefully formulate the deployment problem and design an efficient algorithm, named SSE, to solve it. Moreover, we conduct a detailed measurement study to derive system models of the IoT analytics based on diverse QoS levels and heterogeneous devices to facilitate the optimal deployment decisions. We implement a testbed to conduct experiments, which show that the system models achieve reasonably good accuracy. More importantly, 100% of the deployed IoT analytics satisfy the QoS targets. We also conduct extensive simulations for larger-scale scenarios. The simulation results reveal that our SSE algorithm outperforms a state-of-the-art algorithm by up to 89.4% and 168.3% in terms of the number of satisfied IoT analytics and active devices. In addition, our SSE algorithm reduces CPU, RAM, and network resource consumptions by 18.4%, 12.7%, and 898.3%, respectively, and terminates in polynomial time.

Nearly all of today's Big Data systems are passive in nature. We demonstrate our Big Active Data ("BAD") system, a scalable system that continuously and reliably captures Big Data and facilitates the timely and automatic delivery of new information to a large population of interested users as well as supporting analyses of historical information. We built our BAD project by extending an existing scalable, open-source BDMS (AsterixDB [1]) in this active direction. In this demonstration, we allow our audience to participate in an emergency notification application built on top of our BAD platform, and highlight its capabilities.

Underwater acoustic networks hold the promise to support a wide variety of applications; however, many desirable network services still remain unavailable to application designers. One such service is on-line node localization, which is useful both to many routing algorithms as well as to a number of sensing applications. The main contribution of this work is a range-based low-overhead localization technique, LOTUS, for underwater sensor networks. Unlike earlier range-based 3D localization techniques that required distance estimates from at least four reference nodes to localize any node, our scheme can approximate locations with two references. This enables our scheme to work in low density deployment cases and allows more nodes to be localized. Once localized, nodes exchange their locations in regular packets to further improve the estimates. We evaluate LOTUS using simulation and show that it has a good performance vs. overhead trade-off.

In recent years, notification services for social networks, mobile apps, messaging systems and other electronic services have become truly ubiquitous. When a new content becomes available, the service sends an instant notification to the user. When the content is produced in massive quantities, and it includes both large-size media and a lot of meta-information, it gives rise to a major challenge of selecting content to notify about and information to include in such notifications. We tackle three important challenges in realizing rich notification delivery: (1) content and presentation utility modeling, (2) notification selection and (3) scheduling of delivery. We consider a number of progressive presentation levels for the content. Since utility is subjective and hard to model, we rely on real data and user surveys. We model the content utility by learning from large-scale real world data collected from Spotify music streaming service. For the utility of the presentation levels we rely on user surveys. Blending these two techniques together, we derive utility of notifications with different presentation levels. We then model the selection and delivery of rich notifications as an optimization problem with a goal to maximize the utility of notifications under resource budget constraints. We validate our system with large-scale simulations driven by the real-world de-identified traces obtained from Spotify. With the help of several baseline approaches we show that our solution is adaptive and resource efficient.

Safe Community Awareness and Alerting Network (SCALE) is a community government/academic/industry partnership effort that aims to deploy, actuate and evaluate techniques to support multiple heterogeneous IoT technologies in real world communities. SCALE2, an extension of SCALE, engages a multi-tier and multi-network approach to drive data flow from IoT devices to the cloud platforms. While devices are used to gather data, most of the analytics are executed in the cloud. Managing and utilizing these multiple networks, devices and technologies is a big challenge that calls for an integrated management. In this context, we propose to leverage a related effort, MINA (Multi-network INformation Architecture), that aims at integrating operations of multi-networks IoT deployments in a hierarchical manner. This paper discusses the mapping of the SCALE2 heterogeneous platforms in the MINA environment and argues for a hierarchical approach to extending and managing community IoT multi-networks. We discuss resilience methods that can be employed at different tiers in the hierarchical architecture. We illustrate examples of how multiple applications can be supported in this heterogeneous setting; example applications include cooperative seismic event detection, mobile data collections for air quality information and assisted living for elders. Finally, we discuss novel research challenges associated with managing multi-network IoT architecture.

In this paper, we develop effective solutions for enabling mobile sensing/data collection in community IoT deployments where sensing/communication coverage is intermittent and varying. Specifically, we address the optimized upload planning problem, i.e. determine optimal schedules for upload of gathered information to enable timely data collection in dynamic settings. We develop a two-phase approach and associated policies, where an initial upload plan is generated with prior knowledge of upload opportunities and data needs, and a subsequent runtime adaptation phase alters the plan based on dynamic network and data conditions. To validate our approach, we designed and built SCALECycle, a prototype mobile data collection platform and deployed it in real world community settings; measurements from testbeds in Rockville, MD and Irvine, CA are used to drive extensive simulations. Experimental results indicate that a judicious combination of policies in the two phases of upload planning (a balanced delay-opportunity-priority method with Lyapunov-inspired upload adaptation) can result in a 30-60% improvement in overall utility of collected data compared with opportunistic operation along with 30% reduction in collection delays /overheads.

In this paper, we introduce Minerva; an information-centric programming paradigm and toolkit for social sensing. The toolkit is geared for smartphone applications whose main objective is to collect and share information about the physical world. Information-centric programming refers to a publish-subscribe paradigm that maximizes the amount of information delivered. Unlike a traditional publish-subscribe system where publishers are assumed to have independent content, Minerva is geared for social sensing applications where different sources (participants sharing sensor data) often overlap in information they share. For example, through lack of coordination, they might collect redundant pictures of the same scene or redundant speed measurements of the same street. The main contribution of Minerva, therefore, lies in a data prioritization scheme that maximizes information delivery from publishers to subscribers by reducing redundancy, taking into account the non-independent nature of content. The algorithm is implemented on Android phones on top of the recently introduced named data networking framework. Evaluation results from both two smartphone-based experiments and a large-scale real data driven simulation demonstrate that the prioritization algorithm outperforms other candidates in terms of information coverage.

This paper presents a novel multicopy routing protocol for disruption-tolerant networks whose objective is to minimize energy expended on communication. The protocol is designed for disaster-response applications, where power and infrastructure resources are disrupted. Unlike other delay-tolerant networks, energy is a vital resource in post disaster scenarios to ensure availability of (disruption-tolerant) communication until infrastructure is restored. Our approach exploits naturally recurrent mobility and contact patterns in the network, formed by rescue workers, volunteers, survivors, and their (possibly stranded) vehicles to reduce the number of message copies needed to attain an adequate delivery ratio in the face of disconnection and intermittent connectivity. A new notion of intercontact routing is proposed that allows estimating route delays and delivery probabilities, identifying more reliable routes and controlling message replication and forwarding accordingly. In addition, we augment the protocol with a differentiated message delivery service that enables the network to function even in an extremely low energy condition. We simulate the scheme using a mobility model that reflects recurrence inspired by disaster scenarios and compare our results to previous DTN routing techniques. The evaluation shows that the new approach reduces the resource overhead per message over previous approaches while maintaining a comparable delivery ratio at the expense of a small (bounded) increase in latency.

This demonstration illustrates a service for collection and delivery of images, in participatory camera networks, to maximize coverage while removing outliers (i.e., irrelevant images). Images, such as those taken by smart-phone users, represent an important and growing modality in social sensing applications. They can be used, for instance, to document occurrences of interest in participatory sensing campaigns, such as instances of graffiti on campus or invasive species in a park. In applications with a significant number of participants, the number of images collected may be very large. A key problem becomes one of data triage to reduce the number of images delivered to a manageable count, without missing important ones. In prior work, the authors presented a service, called PhotoNet [2], that reduces redundancy among delivered images by maximizing diversity. The current work significantly extends our previous effort by recognizing that diversity maximization often leads to selection of outliers; images that are visually different but not necessarily relevant, which in fact reduces the quality of the delivered image pool. We demonstrate a new prioritization technique that maximizes diversity among delivered pictures, while also reducing outliers.

We propose PhotoNet, a picture delivery service for camera sensor networks. PhotoNet is motivated by the needs of disaster-response applications, where a group of survivors and first responders may survey damage and send images to a rescue center in the absence of a functional communication infrastructure. The protocol runs on mobile devices, handling opportunistic forwarding (when they come in contact) and in-network storage. It assigns priorities to images for forwarding and replacement depending on the degree of similarity (or dissimilarity) among them, such that scarce resources are assigned to delivery of most ``deserving'' content first. Prioritization aims at reducing semantic redundancy such as that between pictures of the same scene at the same location taken from slightly different angles. This is in contrast to redundancy among identical objects and among time series data. PhotoNet delivers more diverse pictures in terms of event coverage suppressing logically redundant content belonging to the same event. We show that, in resource constrained networks, reducing semantic redundancy can significantly improve the utility of the service.

In this paper, we propose countermeasures to mitigate damage caused by spoofing attacks in Delay-Tolerant Networks (DTNs). In our model, an attacker spoofs someone else's address (the victim's) to absorb packets from the network intended for that victim. Address spoofing is arguably a very severe attack in DTNs, compared to other known attacks, such as dropping packets. Without a Public Key Infrastructure in DTNs, providing protection against this attack is challenging. We propose SPREAD (countermeasure against SPoofing by REplica ADjustment), a solution that assesses evidence of spoofing and offers countermeasures designed for quota-based multi-copy routing protocols. Our solution relies on reducing the weight of packet copies, charged to the routing quota, when these packets are given to a node suspected of spoofing. The weight reduction increases as spoofing evidence mounts against a node. The approach is designed to probabilistically maintain the same number of packet copies in the network as would be the case in the absence of attacks, despite the actual occurrence of spoofing. We show that SPREAD makes DTNs robust against spoofing attacks, does not overburden the network, and limits the overall overhead within a certain bound.

At CPSWeek 2011, the authors presented a demonstration of Apollo, a fact-finder for participatory sensing that ranks archived human-centric and sensor data by credibility. The current demonstration significantly extends our previous work by allowing Apollo to operate on live streaming data; in this case, live Twitter feeds. As the role of humans as sensors increases in emerging sensing applications, a principled approach becomes necessary to address the problem of ascertaining the veracity of sources and observations made by them. Participatory and social sensing applications may use potentially unreliable or unverified sources, such as a phone-based sensing application that grows virally in a large un-vetted population, a disaster-response application, where conflicting damage assessment reports may come from large numbers of different volunteers, or a military application, where friendly observers at a remote location may make hard-to-verify claims about local events. Apollo analyzes noisy data that increasingly plagues human-centric sensing to determine which items of information are more likely to be true.

In this paper, we develop a cooperative mechanism, RELICS, to combat selfishness in DTNs. In DTNs, nodes belong to self-interested individuals. A node may be selfish in expending resources, such as energy, on forwarding messages from others, unless offered incentives. We devise a rewarding scheme that provides incentives to nodes in a physically realizable way in that the rewards are reflected into network operation. We call it in-network realization of incentives. We introduce explicit ranking of nodes depending on their transit behavior, and translate those ranks into message priority. Selfishness drives each node to set its energy depletion rate as low as possible while maintaining its own delivery ratio above some threshold. We show that our cooperative mechanism compels nodes to cooperate and also achieves higher energy-economy compared to other previous results.

This paper computes end-to-end delay bounds for prioritized data flows in disruption-tolerant networks (DTNs). DTNs suffer intermittent connectivity among nodes due to node mobility. When deployed in mission-critical applications, such as disaster response, an interesting question becomes to quantify end-to-end packet delays under assumptions on node mobility. In this paper, we answer this question for the special case of DTNs with recurrent mobility patterns. A recurrent pattern refers to one where nodes revisit the same locations repeatedly. We devise a suitable model for recurrent DTNs that captures their timing and mobility properties. We then apply the recently proposed delay composition algebra to the resulting network model in order to determine an upper bound on end-to-end communication delays of network flows. Evaluation results show that the upper bound is moderate in its pessimism and can be used for deployment planning purposes.

Users and administrators of large distributed systems are frequently in need of monitoring and management of its various components, data items and resources. Though there exist several distributed query and aggregation systems, the clustered structure of tele-immersive interactive frameworks and their time-sensitive nature and application requirements represent a new class of systems which poses different challenges on this distributed search. Multi-attribute composite range queries are one of the key features in this class. Queries are given in high level descriptions and then transformed into multi-attribute composite range queries. Designing such a query engine with minimum traffic overhead, low service latency, and with static and dynamic nature of large datasets, is a challenging task. In this paper, we propose a general multi-attribute based range query framework, Q-Tree, that provides efficient support for this class of systems. In order to serve efficient queries, Q-Tree builds a single topology-aware tree overlay by connecting the participating nodes in a bottom-up approach, and assigns range intervals on each node in a hierarchical manner. We show the relative strength of Q-Tree by analytically comparing it against P-Tree, P-Ring, Skip-Graph and Chord. With fine-grained load balancing and overlay maintenance, our simulations with PlanetLab traces show that our approach can answer complex queries within a fraction of a second.

This paper presents a novel multi-copy routing protocol for disruption-tolerant networks whose objective is to minimize energy expended on communication. The protocol is designed for disaster-response applications, where power and infrastructure resources are disrupted. Unlike other delay-tolerant networks, energy is a vital resource in post-disaster scenarios to ensure availability of (disruption-tolerant) communication until infrastructure is restored. Our approach exploits naturally recurrent mobility and contact patterns in the network, formed by rescue workers, volunteers, survivors, and their (possibly stranded) vehicles to reduce the number of message copies needed to attain an adequate delivery ratio in the face of disconnection and intermittent connectivity. A new notion of inter-contact routing is proposed that allows estimating route delays and delivery probabilities, identifying more reliable routes and controlling message replication and forwarding accordingly. We simulate the scheme using a mobility model that reflects recurrence inspired by disaster scenarios, and compare our results to previous DTN routing techniques. The evaluation shows that the new approach reduces the resource overhead per message over previous approaches while maintaining a comparable delivery ratio at the expense of a small (bounded) increase in latency.

This paper introduces the abstraction of energy reserves for sensor networks that virtualizes energy sources. It gives each of several applications sharing a platform the illusion of having its own private energy source. Energy virtualization is the next logical step in embedded systems after visualizing communication links and CPU capacity. Energy virtualization has not been addressed in past sensor network literature because most current wireless sensor networks feature single-user applications. To amortize deployment costs, future sensor networks, deployed in remote or hard- to-access areas, will likely be leveraged by scientists from different disciplines, each having their independent application for their individual research purposes. Platforms, planned for such deployment, will befitted with the union of sensors needed, but independent applications will share the remaining resources such as in-field storage and communication bandwidth, calling for quotas and isolation mechanisms. The most expensive resource shared in sensor networks is energy. This paper provides an energy isolation mechanism, called the virtual battery, that logically divides energy among applications to provide each its private energy reserve. An application can manage its private energy independently as if it were running alone on the platform. The application is terminated when its reserve is depleted. We implement and evaluate this abstraction on MicaZ motes running LiteOS. Our results show that the virtual battery mechanism succeeds at exporting the private reserve abstraction accurately and at a low overhead.