Our paper "Tackling the Runtime Context Fluctuation: Steady Scheduling in Data Stream Analytics for Industrial Internet of Things" is Accepted by IEEE ICDCS 2026!

This paper proposes a steady scheduling framework based on the concept of side-effect for IIoT data stream analytics, which updates only configurations with small side-effects to shrink the search space, thereby maintaining already-met QoS objectives while re-meeting violated ones under runtime context fluctuations.

Dayu system Version 1.2 is released!

Our paper "Tackling the Imbalance in Video Analytics Pipelines with Hierarchical Embodied Intelligence" is Accepted by IEEE INFOCOM 2026!

This paper proposes Hier-EI (Hierahical Embodied Intelligence), an efficient scheduling framework that integrates a two-phase hierarchical design with embodied intelligence, which simultaneously considers spatial and temporal imbalance and enhance long-term QoE performance in cloud-edge collaborative video analytics pipelines.

Refer to homepage: https://dayu-autostreamer.github.io/hier-ei/.

Dayu system Version 1.1 is released!

Our paper "Adaptive Region-aware Video Encoding for Real-time Cloud-edge Collaborative Object Detection" is Accepted by IEEE ICDCSW 2025!

This paper proposes CRAVE (Collaborative Region-aware Adaptive Video Encoding), which achieves efficient candidate region extraction through lightweight foreground detection and historical inference result analysis.

Dayu system Version 1.0 is released!

Welcome to Dayu, an auto scheduling system for stream data processing on distributed cloud-edge platforms.

We provide infrastructure for cloud-edge collaborative stream data analytics and focus on scheduling tasks like configuration optimization, model evolution, task offloading, and so on.