Intelligent Edge Computing

Permanent URI for this collectionhttps://hdl.handle.net/10125/112562

Browse

Now showing 1 - 9 of 9

Edge-Kalman Filter for Improved Quality of Service in Indoor Spatial Temporal Systems
(2026-01-06) Kalidindi, Vivya; Chowriappa, Pradeep
Indoor Spatial Temporal Systems (ISTS) enable real-time location-aware services in applications like indoor navigation, healthcare monitoring, and emergency response. They rely on technologies such as Bluetooth, Wi-Fi, and Zigbee to estimate distances from Received Signal Strength Indicator (RSSI) signals, which are prone to aleatoric uncertainty. Traditional ISTS mitigate this by applying noise filtering at the cloud layer but introduces transmission latency. Recent work shifts filtering to the edge to reduce transmission latency. However, frequent noise parameter updates in the filtering process lead to a high computational load, which increases communication latency and degrades the Quality of Service (QoS). To address this, we propose a lightweight edge-Kalman Filter (eKF) that updates noise parameters only when a significant change in density ratio is detected using the Kullback-Leibler Importance Estimation Procedure (KLIEP). Experimental results show that eKF reduces computational overhead, enhances distance estimation accuracy, and scales across technologies, demonstrating its potential to improve real-time QoS in ISTS.
Model Evaluation for Radio-Frequency Signal Modulation Classifiers in the Existence of Novel Samples
(2026-01-06) Trott, Adam; Thompson, Henry; Kul, Gokhan
We present an open-set recognition (OSR) pipeline for radio-frequency (RF) signal modulation that extends the recently proposed varMax uncertainty framework to the communications domain. Leveraging a multi-domain convolutional neural network (CNN) that takes in four representations of an electromagnetic signal in the form of raw In-phase and Quadrature (IQ) values and three transformations, Fast Fourier Transform (FFT), Discrete Cosine Transform (DCT), and polar magnitude, we train on eight signal modulations and treat the ninth modulation as unknown. Per fold, 1000 epochs are used to train each classifier on a unique set of eight modulations. Our approach supports the reliable rejection of unseen modulations on most folds while preserving in-distribution classification accuracy. However, the effectiveness is highly dependent on the makeup of the open and closed sets, and even seemingly inconsequential changes have a significant impact on the model's performance. This study demonstrates the strengths and weaknesses of extending the varMax approach to the RF domain and introduces modifications that improve the model's performance when trained on heterogeneous data.
MoCap2Radar: A Spatiotemporal Transformer For Synthesizing Micro‑Doppler Radar Signatures from Motion Capture
(2026-01-06) Chen, Kevin; Parker, Kenneth; Arora, Anish
We present a pure machine learning process for synthesizing radar spectrograms from Motion‑Capture (MoCap) data. We formulate MoCap‑to‑spectrogram translation as a windowed sequence‑to‑sequence task using a transformer‑based model that jointly captures spatial relations among MoCap markers and temporal dynamics across frames. Real‑world experiments show that the proposed approach produces visually and quantitatively plausible doppler radar spectrograms and achieves good generalizability. Ablation experiments show that the learned model includes both the ability to convert multi-part motion into doppler and an understanding of the spatial relation between different parts of the human body. The result is an interesting example of using transformers for time-series signal processing. It is especially applicable to edge computing and Internet of Things (IoT) radars. It also suggests the ability to augment scarce radar datasets using more abundant MoCap data for training higher-level applications. Finally, it requires far less computation than physics-based methods for generating radar data.
Initial Review of Goal-Directed Answer Set Programming for Orchestration of Neuro-Symbolic UAV Autonomy
(2026-01-06) Tudor, Alexis; Kimbrell, Keegan; Gupta, Gopal
Autonomous unmanned aerial vehicles (UAVs) are capable of solving a variety of problems in environments where humans cannot easily go, such as military zones or search and rescue operations. Edge computing explores the trade-off between running software at the edge onboard a UAV (limited space for processing power and memory) versus on a server (increased latency and connectivity problems). The state-of-the-art solution for UAV path planning is the use of deep learning, however these systems are large and unexplainable. Furthermore, when these models need to be adjusted they often cannot be retrained on the edge. We propose the use of a powerful symbolic reasoner that can be run and modified on the UAV for path planning, making explainable and trustworthy edge autonomous systems. We evaluate the feasibility of this approach with a proof-of-concept symbolic UAV system based on the VECSR system for commonsense reasoning, which we call VECSR-A
Edge-Based Fault diagnosis of Autonomous Ground Robots with Simultaneous Actuator and Sensor Faults
(2026-01-06) Janakiraman, Vaishnavi; Zhang, Xiaodong (Frank)
In this paper, an edge-based fault diagnosis scheme is developed for a nonlinear ground vehicle model with possible occurrence of simultaneous actuator faults in the form of loss of effectiveness (LOE) and sensor bias faults. Based on the vehicle and fault models under consideration, the unknown fault parameters are estimated using online adaptive estimation methods. The estimated fault parameters can be utilized to detect, isolate, and accommodate faults in robotic system components, enabling the development of self-diagnostic smart sensors and actuators. Real-time experimental results using a ground robot are shown to illustrate the effectiveness of the proposed algorithm.
Addressing the Lack of Neuromorphic Data in Low SNR Scenarios
(2026-01-06) Perry, Ross; Di Caterina, Gaetano; Bihl, Trevor; Combs, Kara
Neuromorphic systems have shown promising benefits to edge computing systems. However, the scarcity of event-based data is limiting progress in the development of models optimised for asynchronous event-based inputs. Event-based systems have shown promising solutions for addressing low signal-to-noise ratio (SNR) conditions, particularly in scenarios where traditional frame-based computer vision approaches break down. We present a recording methodology that captures low SNR motion using two spatially similar species of blowfly as a proxy for small, erratic objects in a high dynamic range environment. Using this dataset, we evaluate a traditional frame-based Convolutional Neural Network (CNN), ResNet50, and show that despite its rich spatial capabilities, it fails to distinguish between our two classes by achieving a test accuracy of 51%. These results highlight the need for event data and event-native models such as Spiking Neural Networks (SNNs) that can utilise spatial-temporal features for classification tasks.
HALO: A Hierarchical and Adaptive Large Language Model Framework with Centralized Feedback
(2026-01-06) Lee, Yun-Hao; Taaghol, Pouya; Kim, Hyundong; Gupta, Gopal; Li, Bingzhe
The rapid growth of Large Language Models (LLMs) has driven advances across many NLP tasks, but their size and computational demands hinder deployment on resource-constrained client devices, from laptops to small-GPU desktops. At the same time, expanding edge applications and privacy concerns demand on-device intelligence that is efficient, adaptive, and personalized. Existing techniques such as pruning and quantization address parts of this challenge but fall short of balancing accuracy, hardware diversity, user customization, and long-term adaptability. In this paper, we present HALO (Hierarchical and Adaptive LLM Framework with Centralized Feedback), a system for lightweight and continuously improving edge deployment. HALO profiles device hardware, selects and fine-tunes suitable LLM variants with LoRA and quantization, and leverages user feedback to flag underperforming outputs for centralized correction and retraining. Updated adapters are then periodically redistributed, enabling sustained accuracy, efficiency, and privacy across heterogeneous edge devices.
Encoding Time Series on an FPGA, with an Efficient Izhikevich Neuron Implementation
(2026-01-06) Borrill, Thomas; Crockett, Louise; Thomas, Kevin; Di Caterina, Gaetano
Neuromorphic processing is a low power technique that can be used to process different types of data, including time series. ASICs are expensive and custom neuromorphic ICs are not yet widely available, such that another method needs to be considered, to speed up neuromorphic execution in hardware. This paper presents a spike encoding method using an FPGA, a readily available existing technology that can be wielded to process time series data. This approach efficiently encodes and decodes time series data and interfaces into a single layer of Izhikevich neurons. The technique we propose implements a low power and small FPGA design that responds to fast and dynamically changing chaotic time series input with low loss in the encoder and decoder. We explore the differing dynamics of the Izhikevich neuron and the effects on the error of the reconstructed time series data, proposing a set of constraints to minimise the error. Furthermore we investigate the number of neurons required to have an accurate spiking representation of the input with minimum loss.
Introduction to the Minitrack on Intelligent Edge Computing
(2026-01-06) Juric, Radmila; Ronchieri, Elisabetta; Bihl, Trevor; Sanfilippo, Filippo

Browse

Recent Submissions