Advanced Topics in Computer Networks

 

  • Time/Place: Sun/Wed 10:00-11:30, B9, Room 3120
  • Required textbook: None
  • Reference Materials: Research papers.
  • Course description: The course aims to train students in conducting major research in relevant aspects of Agentic AI, 6G and NTNs. In particular, scalability, technologies, standards, and protocols for various communication technologies, edge vs. fabirc, virtualization, dynamic networks , PHY-layer, and MAC-layer.
  • Prerequisites: Solid computer networks background, excellent skills in Python, Linux programming.
  • Topic: AI Agents in 6G None-Terrestrial Networks

 Student must obtain the passing grade (70%) in each task and the final project to pass the course.
The course will include a major research project component that requires performing several paper reviews, presentations, and implementation. 

  • Grading: 
    • Class participation (attendance, paper reviews, and interaction in discussions): 10%
    • Presentations: 10%
    • Homework: 30%
    • Final Project: 50%
  • Homework policy: all assignments, including contributions to discussion, submitted by students in the course of this class should be work written by themselves specifically for this class. Students must clearly cite and reference each and every source that was used in their development. Where students use the actual words of a source, they must put those words inside quotation marks.
  • Calendar: 

30/08 Couse Introduction 

02/09 What is AI Agents 

  • Key Topics: course overview; from LLM chatbots to AI agents; Agent–Environment interaction; Agent = Model + Context + Tools; reasoning, acting, observation, trajectories; observation and action spaces.
  • Required Readings:
    • S. Yao et al., "ReAct: Synergizing Reasoning and Acting in Language Models," in Proceedings of the International Conference on Learning Representations, 2023.
    • T. Schick et al., "Toolformer: Language models can teach themselves to use tools," in Advances in Neural Information Processing Systems, vol. 36, pp. 68539–68551, 2023.
  • Reference: AI Agent Book, Chapter 1 — https://github.com/bojieli/ai-agent-book/blob/main/book-en/chapter1.md

06/09 Building an Agent

  • Key Topics: from an LLM API call to an agent loop; tool/function calling; context and state; ReAct; stopping conditions; error handling and retries; agent harness and interfaces.

  • Required Readings:

    • J. Yang et al., "Swe-agent: Agent-computer interfaces enable automated software engineering," in Advances in Neural Information Processing Systems, vol. 37, pp. 50528–50652, 2024.

    • Y. Shen et al., "HuggingGPT: Solving AI Tasks with ChatGPT and its Friends in Hugging Face," in Advances in Neural Information Processing Systems, vol. 36, pp. 38154–38180, 2023.

  • Implementation References:

    - Hello-Agents — https://github.com/datawhalechina/hello-agents

    - AI Agents From Scratch — https://github.com/pguso/ai-agents-from-scratch

  • Assignment 1 — Build a Minimal Agent, Due 27/9

    Goal: build a minimal conversational AI agent from first principles and understand the core execution loop rather than rely on a high-level framework.

     

    Minimum requirements: LLM interaction; system instruction; at least two tools; function/tool calling; multi-step agent loop; task state; stopping condition; basic error handling; trajectory logging; a small evaluation set.

09/09 Context Engineering I: What Should the Agent See?

  • Key Topics: context window vs. usable context; system and user prompts; tool definitions; observations and history; context selection and ordering; context budgeting; lost-in-the-middle effects; prompt and context compression.

  • Required Readings:

    • N. F. Liu et al., "Lost in the Middle: How Language Models Use Long Contexts," in Transactions of the Association for Computational Linguistics, vol. 12, pp. 157–173, 2024.

    • H. Jiang et al., "LongLLMLingua: Accelerating and Enhancing LLMs in Long Context Scenarios via Prompt Compression," in Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pp. 1658–1677, 2024.

  • Implementation Reference:

    - LLMLingua — https://github.com/microsoft/LLMLingua

     

    Reference: AI Agent Book, Chapter 2 — https://github.com/bojieli/ai-agent-book/blob/main/book-en/chapter2.md

     

13/09 Context Engineering II: Managing Long-Horizon Agents

  • Key Topics: KV cache and prompt caching; context reuse; structured state; scratchpads and working memory; context pruning and summarization; Agent Skills; dynamic loading of tools and instructions; long-horizon state management.

  • Required Readings:

    • I. Gim et al., "Prompt Cache: Modular Attention Reuse for Low-Latency Inference," in Proceedings of Machine Learning and Systems, vol. 6, 2024.

    • J. Yao et al., "CacheBlend: Fast Large Language Model Serving for RAG with Cached Knowledge Fusion," in Proceedings of the European Conference on Computer Systems (EuroSys), pp. 94–109, 2025.

  • Implementation References:

    - LMCache — https://github.com/LMCache/LMCache

    - Anthropic Skills — https://github.com/anthropics/skills

     

    Reference: AI Agent Book, Chapter 2 — https://github.com/bojieli/ai-agent-book/blob/main/book-en/chapter2.md

16/09 Agent Memory: Remembering Users and Experiences

  • Key Topics: conversation history vs. memory; working, episodic, semantic, and user memory; memory write, retrieval, and update; personalization; forgetting; persistent agent state.
  • Required Readings:
    • C. Packer et al., "MemGPT: Towards LLMs as Operating Systems," arXiv preprint arXiv:2310.08560, 2023.
    • J. S. Park et al., "Generative Agents: Interactive Simulacra of Human Behavior," in Proceedings of the ACM Symposium on User Interface Software and Technology, pp. 1–22, 2023.
  • Implementation References:

          - Letta — https://github.com/letta-ai/letta

          - Hello-Agents — https://github.com/datawhalechina/hello-agents

         Reference: AI Agent Book, Chapter 3 — https://github.com/bojieli/ai-agent-book/blob/main/book-en/chapter3.md

20/09 Knowledge for Agents: RAG, Search & Structured Knowledge + Assignment 1 Studio

  • Key Topics: external knowledge vs. agent memory; RAG pipelines; embeddings and semantic search; reranking; structured retrieval; metadata filters; knowledge graphs; retrieval as an agent action; grounding and citation

  • Required Readings:

    • A. Asai et al., "Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection," in Proceedings of the International Conference on Learning Representations, 2024

    • D. Edge et al., "From Local to Global: A Graph RAG Approach to Query-Focused Summarization," arXiv preprint arXiv:2404.16130, 2024.

      Implementation References:

      - RAG From Scratch — https://github.com/pguso/rag-from-scratch

      - Microsoft GraphRAG — https://github.com/microsoft/graphrag

      Reference: AI Agent Book, Chapter 3 — https://github.com/bojieli/ai-agent-book/blob/main/book-en/chapter3.md

       

      Assignment 1 Studio: Debugging agent loops; tool design; trajectory inspection; failure analysis; designing the Assignment 1 evaluation set; common implementation mistakes.

23/09 No Class

27/09 Assignment 1 Results

  • Each to present 5-8 slides on the assignment results

    Suggested presentation: task & motivation; agent architecture; tools/environment; execution loop; evaluation setup; results; failure cases; lessons learned.

  • Assignment 2 — Evaluate a SOTA Agent, Due 14/10

    Goal: design and conduct a reproducible evaluation of a state-of-the-art agent on a task domain you care about. Build a small benchmark rather than only describing product experience.

    Minimum requirements: define a task domain and evaluation set; define success criteria and rubric/metrics before running the agent; evaluate a current SOTA agent; include at least one meaningful baseline; record cost/token usage and/or latency where possible; analyze failures; make the evaluation reproducible.

30/09 Evaluating Agents I: What Does “Good” Mean?

  • Key Topics: evaluating agents vs. LLMs; model vs. agent vs. harness; task distributions; evaluation environments; success criteria and metrics; benchmark design; capability vs. reliability; final-answer vs. trajectory evaluation

  • Required Readings:

04/10 Evaluating Agents II: Reliability, Rubrics & Benchmark Design

  • Key Topics: reproducible evaluation; test cases and datasets; rubric-based evaluation; LLM-as-a-Judge; repeated trials; Pass@k and reliability; cost, latency, token usage; failure taxonomy; model-swap and ablation experiments.
  • Required Readings:
    • S. Yao et al., "τ-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains," in arXiv preprint, 2024.
    • G. Mialon et al., "GAIA: a benchmark for General AI Assistants," in Proceedings of the International Conference on Learning Representations, 2024.

          Implementation References:

           - τ-bench / tau2-bench — https://github.com/sierra-research/tau2-bench

            - GAIA — https://huggingface.co/gaia-benchmark

           Reference: AI Agent Book, Chapter 6 — https://github.com/bojieli/ai-agent-book/blob/main/book-en/chapter6.md

07/10 Tools: Connecting Agents to the World 

  • Key Topics: tool/function calling; tool schemas; observation vs. action tools; tool routing and selection; retries and timeouts; permissions and sandboxing; MCP; proactive and asynchronous tool use.
  • Required Readings:
    • S. G. Patil et al., "Gorilla: Large Language Model Connected with Massive APIs," in Advances in Neural Information Processing Systems, vol. 37, 2024.
    • Y. Qin et al., "ToolLLM: Facilitating Large Language Models to Master 16000+ Real-world APIs," in Proceedings of the International Conference on Learning Representations, 2024.

            Implementation References:

            - Model Context Protocol Servers — https://github.com/modelcontextprotocol/servers

            - ToolBench — https://github.com/OpenBMB/ToolBench

             - Gorilla — https://github.com/ShishirPatil/gorilla

            Reference: AI Agent Book, Chapter 4 — https://github.com/bojieli/ai-agent-book/blob/main/book-en/chapter4.md

11/10 From Large AI Models to Agentic AI in 6G: Architectures, Protocols, and Applications 

  • Key Topics: LAMs to Agentic AI evolution; LAM components and classifications; Agentic AI architecture (reasoning, planner, RAG/KG, tools, memory); single- vs. multi-agent collaboration; wireless/semantic comms/network automation applications; scalability and trust challenges

  • Required Readings:

    F. Jiang, et al., "From Large AI Models to Agentic AI: A Tutorial on Future Intelligent Communications," in IEEE Journal on Selected Areas in Communications, vol. 44, pp. 3507-3540, 2026

14/10 Assignment 2 Results

  • Each to present 5-8 slides on the assignment results

    Suggested presentation: research question; task domain and evaluated agent; evaluation dataset; metrics/rubric; baseline; quantitative results; failure analysis; conclusions.

    Assignment 3 — Multi-Agent Communication, Due 4/11

    Goal: design and implement a small multi-agent system in which multiple agents communicate and collaborate to solve a task. The focus is on how information is exchanged between agents and whether communication provides measurable value over a single-agent or non-communicating baseline.

    Minimum requirements: at least three agents with clearly defined roles; an explicit agent-to-agent communication mechanism; a defined communication topology (e.g., peer-to-peer, coordinator, or shared workspace); communication/trajectory logging; at least one meaningful baseline; a small evaluation set; and analysis of task performance, communication cost, and failure cases.

    Students may use multi-agent frameworks such as AutoGen, CAMEL, or similar systems, but should be able to explain the communication mechanism and information flow between agents.

18-21/10  Mid-Semester Break 

25/10 Multi-Agent Systems: Architecture & Coordination

  • Key Topics: role specialization; shared vs. isolated context; agent-to-agent communication; tool calls, shared artifacts, and message passing; peer collaboration; manager/orchestration; decentralized collaboration; cost and complexity.

  • Required Readings:

28/10 The Future of Agentic AI in 6G

  • Key Topics: standardization of agent interfaces; multi-vendor coordination; the future of autonomous network assurance; from AGI to ASI in telecommunications.

  • Required Readings

    • M. A. Ferrag et al., "6G Needs Agents: Toward Agentic AI-Native Networks for Autonomous Intelligence," in IEEE Open Journal of the Communications Society, vol. 7, pp. 7254-7282, 2026.

    • H. Feng et al., "Agentic AI Security in 6G Networks: A Survey of Emerging Attack Vectors, Vulnerabilities, and Defenses," in IEEE Open Journal of the Communications Society, vol. 7, pp. 6334-6368, 2026

28/10 Project Proposal

01/11 Agentic AI for Autonomous Space–Air–Ground Integrated Networks (SAGIN)

  • Key Topics: SAGIN resource management; LLM-enabled agents; hierarchical multi-agent architectures; agent–RL collaboration; distributed satellite coordination; on-orbit edge intelligence; partial observability and trust

  • Required Readings:

    L. Zhang, H. Zhao, B. Xu, H. Zhu and X. Wang, "Agentic AI for SAGIN Resource Management: Semantic Awareness, Orchestration, and Optimization," in IEEE Network, pp. 1-8, 2026

04/11 Assignment 3 Results

  • Each to present 5-8 slides on the assignment results

08/11 Agentic AI Takes Flight: Autonomous UAVs

  • Key Topics: Agentic UAV autonomy; aerial perception, planning, and decision-making; multidomain UAV applications;

  • Required Readings:

    • R. Sapkota et. al., "UAVs meet agentic AI: A multidomain survey of autonomous aerial intelligence and agentic UAVs." arXiv preprint arXiv:2506.08045, 2025.

    • F. Jiang et al., "Agentic AI-Empowered Multi-UAV Trajectory Optimization in Low-Altitude Economy Networks," in IEEE Transactions on Cognitive Communications and Networking, vol. 12, pp. 7461-7473, 2026

11/11 Do We Really Need Multiple Agents? From Collective Intelligence to ASI

  • Key Topics: collective intelligence; multi-agent collectives as a scaling path; when adding agents helps; new information and external feedback; inference-compute control; performance–cost trade-offs; when a single agent is preferable.

  • Required Readings:

15/11 Agentification of Edge Intelligence and Multi-Agent Edge–Terminal Collaboration  

  • Key Topics: Agent AI in edge intelligence, compact model compression, energy-aware computing strategies, robust connectivity frameworks, LLM multi-agent systems; dual-loop edge–terminal collaboration; task decomposition and parallel tool calling; resource-constrained agent deployment

  • Required Readings:

    • R. Zhang et al., "Toward Edge General Intelligence With Agentic AI and Agentification: Concepts, Technologies, and Future Directions," in IEEE Communications Surveys & Tutorials, vol. 28, pp. 4285-4318, 2026

    • Z. Qu et. al., "LLM Enabled Multi-Agent System for 6G Networks: Framework and Method of Dual-Loop Edge-Terminal Collaboration," in IEEE Communications Magazine, vol. 64, no. 3, pp. 54-60, 2026

18/11 From Evaluation to Agent Improvement

  • Key Topics: trajectories as learning signals; evaluation-to-feedback loops; execution feedback; skill acquisition; replay and reflection; deciding whether to update memory, prompts/skills, code, or model parameters.

  • Required Readings:

22/11 Agent Continuous Evolution: Updating Knowledge, Instructions, Code & Parameters

  • Key Topics: memory consolidation; prompt/skill evolution; automatic prompt optimization; tool/skill creation; code/self-modification; model adaptation; agentic RL; evaluation-driven iteration; when to update the harness vs. the model.
  • Required Readings:
    • L. Agrawal et al., "GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning," in International Conference on Learning Representations, 2026.
    • H. Zhang et al., "AgentRL: Scaling Agentic Reinforcement Learning with a Multi-Turn, Multi-Task Framework," in arXiv preprint arXiv:2510.04206, 2025

Implementation References:

- GEPA — https://github.com/gepa-ai/gepa

- AgentRL — https://github.com/THUDM/AgentRL

Reference: AI Agent Book, Chapter 8 — https://github.com/bojieli/ai-agent-book/blob/main/book-en/chapter8.md

25/11 Agentic AI for Network Security and Autonomous Intrusion Detection

  • Key Topics: Multi-agent IDS; LLM-based telemetry analysis; DDoS mitigation and zero-day detection; XAI for root-cause analysis; agentic AI security threats (prompt injection, tool hijacking)
  • Required Readings:
    • P. Zambare et al., "NetMoniAI: An Agentic AI Framework for Network Security & Monitoring," in Proc. IEEE 3rd Int. Conf. Artificial Intelligence, Blockchain, and Internet of Things (AIBThings), 2025.
    • M. Islam et al., "MA-IDS: Multi-Agent RAG Framework for IoT Network Intrusion Detection with an Experience Library." arXiv preprint arXiv:2604.05458, 2026.

29/11 Agentic AI for Digital Twin Networks of Next-Gen Wireless Networks

  • Key Topics: DTN, Multi-agent orchestration, domain-specific applications, closed-loop automation
  • Required Readings:
    • R. Khaldi, et. al., "Agentic-AI Framework for Integrated Design, Implementation, Testing, and Operation of Digital Twin Networks," in IEEE Open Journal of the Communications Society, vol. 7, pp. 4352-4375, 2026

02/12 Agentic AI for QoE-Aware Network Slicing

  • Key Topics: QoE-aware network slicing for IIoT; natural language intent inference with RAG; DRL-based slice orchestration with LLM reward shaping; vector database-assisted cloud-edge LLM QoS; heterogeneous edge LLM routing
  • Required Readings:
    • X. Wang et al., "LLM-Empowered Agentic AI for QoE-Aware Network Slicing Management in Industrial IoT," in IEEE Network, vol. 40, no. 3, pp. 202-210, May 2026

06 - 09/12 Final Project Results