Search for: All records

The deployment of deep learning models for real-time image classification on resource-constrained sensor devices presents significant challenges. These devices face strict limitations in computational power, energy capacity, and memory resources, making it difficult to achieve both high accuracy and low latency. Current approaches either compromise model performance through compression or incur substantial overhead by offloading computation to remote servers. We introduce a novel distributed progressive inference platform that addresses these limitations by dynamically balancing local and remote computation. Our system employs reinforcement learning to make intelligent decisions about when and where to perform inference. Experimental results across multiple standard datasets demonstrate that our approach achieves up to 3% higher accuracy while reducing network traffic and preserving battery life compared to existing methods. 

Not Federated Learning (FL) enables collaborative model training across decentralized clients without sharing raw data, yet faces significant challenges in real-world settings where client data distributions evolve dynamically over time. This paper tackles the critical problem of covariate and label shifts in streaming FL environments, where non-stationary data distributions degrade model performance and necessitate a middleware layer that adapts FL to distributional shifts. We introduce ShiftEx, a shift-aware mixture of experts framework that dynamically creates and trains specialized global models in response to detected distribution shifts using Maximum Mean Discrepancy for covariate shifts. The framework employs a latent memory mechanism for expert reuse and implements facility location-based optimization to jointly minimize covariate mismatch, expert creation costs, and label imbalance. Through theoretical analysis and comprehensive experiments on benchmark datasets, we demonstrate 5.5-12.9 percentage point accuracy improvements and 22-95 % faster adaptation compared to state-of-the-art FL baselines across diverse shift scenarios. The proposed approach offers a scalable, privacy-preserving middleware solution for FL systems operating in non-stationary, real-world conditions while minimizing communication and computational overhead. 

Developers using LLMs and LLM-based agents in their applications have provided plenty of anecdotal evidence that in-contextlearning (ICL) is fragile. In this paper, we show that in addition to the quantity and quality of examples, the order in which the incontext examples are listed in the prompt affects the output of the LLM and, consequently, their performance. While prior work has explored improving ICL through datasetdependent techniques, we introduce OptiSeq, a purely inference-time, dataset-free optimization method that efficiently determines the best example order. OptiSeq leverages log probabilities of LLM-generated outputs to systematically prune the search space of possible orderings and recommend the best order(s) by distinguishing orderings that yield high levels of accuracy and those that underperform. Extensive empirical evaluation on multiple LLMs, datasets, and prompts demonstrates that OptiSeq improves accuracy by 5.5 - 10.5 percentage points across multiple tasks. 

During disasters, ensuring that critical response resources are efficiently allocated to the most appropriate locations is crucial for minimizing adverse impacts and saving lives. To this end, we present RADAR, a data-driven platform that integrates multisource GIS feeds (e.g., USGS earthquake alerts, Cal Fire wildfire perimeters) with facility and transportation data to support proactive planning and real-time recommendations that can be used by Emergency Operations Centers to guide populations to safety. RADAR uses policy-driven stable matching to optimize routing and resource assignment for evacuation planning and resource delivery. The aggregate model allocates populations in impacted facilities to alternate short-term facilities (e.g., hospitals), and a fine-grained extension for long-term senior-care facilities personalizes allocation using resident preferences, medical profiles, and social constraints. RADAR adapts as conditions evolve by utilizing historical data, live traffic, and changing facility status. We validated RADAR's efficacy in several disaster settings, including real events such as the Palisades wildfire and tabletop drills (earthquake and water-contamination scenarios) involving first responders. 

Data regulations like GDPR require systems to support data erasure but leave the definition of erasure open to interpretation. This ambiguity makes compliance challenging, especially in databases where data dependencies can lead to erased data being inferred from remaining data. We formally define a precise notion of data erasure that ensures any inference about deleted data, through dependencies, remains bounded to what could have been inferred before its insertion. We design erasure mechanisms that enforce this guarantee at minimal cost. Additionally, we explore strategies to balance cost and throughput, batch multiple erasures, and proactively compute data retention times when possible. We demonstrate the practicality and scalability of our algorithms using both real and synthetic datasets. 

Smart city transportation infrastructure will soon demand the development of reliable underground IoT (IoUT) communication. In this paper, we develop a novel analytical model, MAME (Material Aware Measurement Enhanced), to capture signal propagation properties in wireless IoUT networks to achieve reliable data transport. A driving motivation is monitoring underground infrastructure systems (e.g., pipelines and storm drains) for early detection of anomalies and failures to guide human investigation and intervention. We analyze the feasibility of successfully receiving wireless data packets from underground (UG) sensor nodes through multiple material layers and under diverse environmental conditions. Our proposed approach integrates physics-based modeling and empirical studies with small-scale testbeds (in our lab and outdoors) with multiple channel setups and physical layer attributes. We derive a novel MAME approach to model signal propagation in both 802.11-based WiFi and LoRaWAN networks. The resulting MAME model is shown to capture communication behavior in WiFi and LoRaWAN networks accurately. The MAME model is used to augment the popular NS3 simulator to explore scaled-up underground networks and varying channel conditions (e.g., soil moisture level). Such a combined analytical-empirical approach will enable the communication control plane and application layer to better predict channel conditions for improved IoUT network design. 

IoT deployments in smart spaces can enable the development of useful services for their inhabitants. However, the diversity of smart spaces and their sensor infrastructures makes it challenging to develop space-agnostic applications. Moreover, existing schemas addressing interoperability challenges often lack the vocabulary needed to represent the integration of smart space systems and their inhabitants. We present a schema to annotate inhabited smart spaces in support of inhabitant-oriented applica- tions. Our schema integrates well-known ontologies to represent inhabitants, events/activities, and the space itself, along with their interconnections. It also supports the representation of uncertain information from IoT and mobile sensors (e.g., a person’s location or occupancy/attendance at an event). Additionally, we introduce an annotation tool that uses an easy-to-use GUI to describe a smart space based on our schema. We demonstrate the potential of our approach through a series of SPARQL queries and a system deployed at the UCI campus that annotates sensor data to support a space-agnostic occupancy monitoring application. 

The panel was held on 14 November 2023 at Purdue University as part of a Grand Challenges in Resilience Workshop sponsored by the U.S. National Science Foundation and organized by our center, the Center for Resilient Infrastructures, Systems, and Processes (CRISP). 

Next-generation stream processing systems for community scale IoT applications must handle complex nonfunctional needs, e.g. scalability of input, reliability/timeliness of communication and privacy/security of captured data. In many IoT settings, efficiently batching complex workflows remains challenging in resource-constrained environments. High data rates, combined with real-time processing needs for applications, have pointed to the need for efficient edge stream processing techniques. In this work, we focus on designing scalable edge stream processing workflows in real-world IoT deployments where performance and privacy are key concerns. Initial efforts have revealed that privacy policy execution/enforcement at the edge for intensive workloads is prohibitively expensive. Thus, we leverage intelligent batching techniques to enhance the performance and throughput of streaming in IoT smart spaces. We introduce BatchIT, a processing middleware based on a smart batching strategy that optimizes the trade-off between batching delay and the end-to-end delay requirements of IoT applications. Through experiments with a deployed system we demonstrate that BatchIT outperforms several approaches, including micro-batching and EdgeWise, while reducing computation overhead.