Skip to main content
Usami Laboratory
Research explainers

Building IoT Prediction with Sparse Graphs and Physical Constraints

Spatiotemporal prediction that limits graph neighbors and incorporates bandwidth and connection constraints into training.

IEICE Technical Report · 2026

  1. 01

    Neighbor Limit K

  2. 02

    Bandwidth Optimization

  3. 03

    Spatiotemporal Prediction

Conceptual overview of the research workflow

The problem

Handling many sensors creates bottlenecks in computational cost and communication bandwidth depletion.

The research idea

Each node has at most K neighbors, giving O(VK) spatial computation, linear in V when K is fixed. Bandwidth and connection constraints are incorporated into the loss.

What was evaluated

Campus Wi-Fi AP experiments (V=81, V=50) achieved 0% bandwidth violations with comparable regression error; also evaluated on BDG2 electricity data.

Considering applications

A research approach for predicting building communication or electricity data while considering computation and operational constraints.

Original research

O(V) Computational Optimization and Physics-informed STGNN for Building IoT Simulation

Tajima, Koichiro, Usami, Hiroyasu

IEICE Technical Report, 2026 · EST2025-95, 2026-01-23

Publication recordPublic source