Edge Computing | Jared Macshane

MANTIS: Modulated Adaptive Neural Compression for Task-Informed Split Computing

Thu, 01 Jan 2026 00:00:00 +0000

MANTIS evaluates task-conditioned split compression on UAVid semantic segmentation, WAID wildlife detection, and Boreal Fire smoke detection. The system improves the low- and mid-rate task-utility frontier, reducing bitrate by up to 62.2% at matched downstream accuracy and improving average normalized task accuracy by up to 9.3% at matched bitrate relative to the best non-conditioned or static baseline at the matched comparison point.

ADAPT: Automated Decision-flow for Adaptive Progressive Inference on Sensor Devices

Mon, 15 Dec 2025 00:00:00 +0000

Published in the Middleware for Autonomous AIoT Systems in the Computing Continuum proceedings, pages 19–24.

MANTIS Neural Compression

Wed, 01 Jan 2025 00:00:00 +0000

MANTIS is a task-informed split-computing system for UAV perception. It compresses visual evidence before transmission by estimating task relevance on the client, modulating the neural analysis transform, entropy-coding the resulting latent, and routing the compact representation to task-specific edge heads.

The system is evaluated across UAV-relevant workloads:

UAVid semantic segmentation for urban scene parsing
WAID wildlife detection for aerial environmental monitoring
Boreal Fire smoke detection for wildfire intelligence

Compared with JPEG, WebP, LADON, learned image codecs, and non-conditioned ablations, MANTIS improves the low- and mid-rate task-utility frontier. The paper reports up to 62.2% bitrate reduction at matched downstream accuracy and up to 9.3% average normalized task-accuracy improvement at matched bitrate.

SHIELD Digital Twin Project

Mon, 01 Jan 2024 00:00:00 +0000

SHIELD is a community-scale digital twin project for disaster resilience. The system connects sensing, simulation, and edge intelligence so resource-constrained perception models can support operational decision-making.

Lead development of a large-scale digital twin platform for disaster resilience
Design progressive inference models for resource-constrained environments
Coordinate cross-functional work across multiple institutions
Build real-time data processing and visualization capabilities

Wildfire Intelligence

Sun, 01 Jan 2023 00:00:00 +0000

Wildfire intelligence work connects perception models with constrained deployment settings where compute, bandwidth, and response time matter.

Developed a generative wildfire-spread model using conditional variational autoencoders
Designed supervised image-compression models that reduced inputs to 4.8 KB while preserving 72.9% wildfire-detection accuracy
Built an edge-computing framework for distributed wildfire detection using early-exit neural networks
Integrated wildfire perception work with broader digital twin and disaster-resilience systems