Abstract

Modern stream processing systems must handle increasingly complex, real-time data flows across diverse domains such as finance, IoT, and multimedia analytics. These systems often operate within service-based architectures, where microservices, serverless computing, and distributed workflows manage the continuous ingestion, processing, and analysis of high-velocity data. Service computing enhances the flexibility of stream systems by enabling dynamic service composition, workload distribution, and adaptive resource management. However, these advantages come with significant challenges, including balancing latency and throughput, ensuring real-time QoS-aware scheduling, and maintaining system reliability under evolving workloads. Traditional approaches, which rely on static rules or predefined heuristics, struggle to adapt to dynamic changes, such as fluctuating network conditions, evolving user demands, and heterogeneous service capabilities. These rigid methodologies create inefficiencies in critical tasks like workload offloading, task scheduling, and anomaly detection, where contextual reasoning and real-time adaptability are crucial.


To address these challenges, Agentic AI introduces a paradigm shift by combining autonomous decision-making, adaptive learning, and real-time reasoning. Agentic AI systems leverage advanced techniques—including LLM-based agents—to interpret unstructured user inputs, dynamically optimize service orchestration, and provide explainable decisions in natural language. Unlike conventional stream systems that apply fixed threshold-based policies, Agentic AI can assess trade-offs dynamically, such as prioritizing low-latency event processing for security alerts while delaying non-urgent analytical tasks. In anomaly detection, these agents do not simply flag outliers but also generate human-interpretable justifications, aiding both incident response and compliance auditing. Service computing further amplifies the capabilities of Agentic AI by providing a structured environment for modular and scalable stream processing. Agentic AI can autonomously compose, coordinate, and optimize microservices across edge/cloud infrastructures—offloading compute-heavy tasks to cloud-based functions while executing time-sensitive operations at the edge. Additionally, these intelligent agents can integrate third-party APIs (e.g., fraud detection, geospatial analytics) into streaming workflows, making real-time data pipelines more adaptable to domain-specific constraints.


However, integrating Agentic AI into high-performance stream-service architectures presents several challenges, such as 1) Balancing inference latency with real-time throughput, especially in large-scale event-driven systems, 2) Ensuring deterministic behavior in mission-critical applications, where autonomous decisions must be predictable and reliable, and 3) Maintaining privacy and security when AI agents process sensitive data in distributed service environments.
This track seeks to explore novel solutions at the intersection of Agentic AI, stream processing, and service computing to enable intelligent, autonomous, and context-aware systems for real-time data processing. We welcome research contributions in the following areas: 1) Agent-driven optimization techniques for stream processing, including reasoning-based service composition and QoS-aware scheduling. 2) Architectures integrating Agentic AI with service-based stream systems, including microservices, serverless platforms, and federated learning.

Topics Covered: Submissions are invited on (but not limited to):

  • Stream tasks offloading, scheduling in edge-cloud environments
  • Agents for autonomic recovery and adaptive scheduling of streams systems
  • Self-improving optimization strategies
  • LLM agents as stream controllers (architectures, prompt engineering).
  • Real-time anomaly detection and explainability (few-shot learning, justification of outliers, Natural language justification), and causal reasoning for root cause analysis
  • LLM-agent architecture for efficiency optimizations.
  • Reasoning-augmented optimization.
  • In context learning and Lightweight fine-tuning for domain-specific efficiency.
  • Agentic AI case studies in IoT, finance, healthcare, smart cities, etc.
  • Agentic AI for service computing management (e.g., composition, orchestration, etc.)

The submission and review process for this Special Track will be managed through the EasyChair conference system.

Relevance to AICCSA

This track perfectly aligns with AICCSA’s focus on applied computing by:

  • Bringing cutting-edge AI reasoning techniques to practical stream processing challenges
  • Addressing challenges at the intersection of 3 Tracks of AICCSA, such as Track 4(AI & Cognitive Systems) and Track 1 (Ubiquitous, Parallel and Distributed Computing)
  • Showcasing real-world applications that demonstrate measurable improvements

Target attendees include:

  • AI researchers working in: Agentic AI, LLM agents, offloading and scheduling stream applications, management of stream service systems, optimization of edge and cloud computing infrastructure, outlier detection in stream data, and optimization problems
  • Industry professionals implementing intelligent stream processing solutions

Potential keynote speakers:

Track co-chairs

Technical program committee

  • Faiza Loukil, Université de Savoie Mont Blanc, France
  • Imen Suiden, Sousse University, Tunisia
  • Sabrine Amri, University of Montreal, Canada
  • Sabeur Lajili, Police College of Qatar, Qatar
  • Biswaranjan Acharya, Marwadi University, India
  • Adel Abusitta, Polytechnique Montreal, Canada
  • Rui Zhu, University of Alberta, Canada
  • Lei Cao, University of Arizona, USA