Edge Computing and Resource-Aware AI

Research question

Which data should a connected system transmit, where should it compute, and what does AI execution cost? Collaborative research develops vehicular map and perception systems, information-centric networking, offloading methods, and measurements of on-device AI. These studies make latency, information freshness, and energy trade-offs explicit.

Methods & contributions

  1. Collaborative construction of live maps

    Live high-definition map construction divides work between vehicles and edge servers under transmission, computation, and latency constraints. COMAP studies proactive provisioning of crowdsourced maps.

    Zhang et al., IEEE TVT 2024 · COMAP, IEEE ICC 2023

  2. Selecting and compressing shared perception

    AdaMap combines object selection with adaptive point-cloud representation to share perception artifacts under latency constraints and changing network conditions.

    AdaMap, ACM/IEEE SEC 2023

  3. Named data, indexing, and information freshness

    DSORL selects map-data sources in vehicular named-data networking; edge-assisted indexing separates dynamic and static mixed-reality data; age-of-information-aware aggregation accounts for delayed and out-of-sequence updates.

    DSORL, IEEE T-ITS 2023 · Doe et al., Intell. Converged Netw. 2024 · Mallik et al., IEEE ICC 2024

  4. Computation beyond one vehicle

    CAVE studies passing vehicles as computation resources. Related work addresses reinforcement-learning offloading for mobile augmented reality and network-aware task distribution for edge-assisted generative AI.

    CAVE, IEEE GLOBECOM 2024 · REAL, IEEE TMC 2026 · Yan et al., IEEE ICC 2026

  5. Measuring AI before optimizing it

    lm-Meter profiles on-device language-model latency, while related studies measure and predict deep-learning energy across edge devices and examine how prompt properties affect on-device LLM energy.

    lm-Meter, ACM/IEEE SEC 2025 · Tu et al., ACM/IEEE SEC 2023 · Hu et al., EMNLP 2026

  6. Model and inference choices under resource constraints

    Energy predictors guide neural-architecture search in GreenAuto and PlatformX. Video-token sparsification reduces redundant visual input for driving question answering.

    GreenAuto, ACM HotMobile 2025 · PlatformX, ACM/IEEE SEC 2025 · Ma et al., IEEE IV 2025

Visual overview

AdaMap Figure 3: vehicle-side point-cloud localization, perspective transformation, adaptive representation and control-plane optimization connect through object descriptors to server-side decoding, object matching and a global map.

AdaMap’s published vehicle-to-edge architecture, evaluated through emulation with real computation and simulated networking.

Q. Liu et al., AdaMap: High-Scalable Real-Time Cooperative Perception at the Edge, SEC '23 (2023), Figure 3. © 2023 owner/author(s). Publication rights licensed to ACM. Rasterized/cropped from arXiv:2309.13526v2; labels unchanged. Personal use; not for redistribution. Version of Record: DOI 10.1145/3583740.3626612. Source for adamap-published-overview ACM author-reuse policy (not an open license)

Tonic meta-control architecture separating slow vigilance memory from a fast surprise-triggered path to select reactive or deliberative modes.

Tonic's simulation-study architecture separates persistent vigilance from surprise-triggered escalation.

Kyungtae Han and coauthors (2026), existing public Tonic Meta-Control project, architecture figure; unchanged. Source for tonic-architecture

Systems & evidence

Representative publications

A curated selection. Citations above link to each paper; see the full publication list for all work.

ACM/IEEE SEC 2023

AdaMap: High-scalable real-time cooperative perception at the edge

Q. Liu, Y. Xue, Y. Zhang, D. Chen, and K. Han

ACM/IEEE Symposium on Edge Computing (SEC), Wilmington, DE, USA

View paper →
ACM/IEEE SEC 2025

lm-Meter: Unveiling Runtime Inference Latency for On-Device Language Models

H. Wang, X. Tu, H. Ke, H. Chai, D. Chen, and K. Han

10th Symposium on Edge Computing (SEC 2025)

View paper →PDF →
IEEE TVT 2024

Crowdsourcing Live High Definition Map via Collaborative Computation in Automotive Edge Computing

Y. Zhang, Q. Liu, H. Wang, D. Chen, and K. Han

IEEE Transactions on Vehicular Technology, vol. 73, no. 9, pp. 13569–13583

View paper →
IEEE T-ITS 2023

DSORL: Data Source Optimization with Reinforcement Learning Scheme for Vehicular Named Data Networks

D. Doe, D. Chen, K. Han, H. Wang, J. Xie, and Z. Han

IEEE Transactions on Intelligent Transportation Systems, vol. 24, no. 10, pp. 11225–11237

View paper →PDF →
IEEE ICC 2024

Unleashing the true power of age-of-information: Service aggregation in connected and autonomous vehicles

A. Mallik, D. Chen, K. Han, L. Jiang, and Z. Han

IEEE International Conference on Communications (ICC): Mobile and Wireless Networks Symposium, Denver, CO, USA

View paper →PDF →
ACM/IEEE SEC 2023

Unveiling energy efficiency in deep learning: Measurement, prediction, and scoring across edge devices

X. Tu, A. Mallik, D. Chen, K. Han, O. Altintas, H. Wang, and J. Xie

ACM/IEEE Symposium on Edge Computing (SEC), Wilmington, DE, USA

View paper →

Full publication list

Other research areas