What We Work On

Research

LAN Lab (Learning and Automation for Networks) studies how learning systems can make next-generation networks more autonomous, robust, and efficient. Our work sits at the intersection of causal inference, reinforcement learning, and game theory, applied to the design and control of 6G, O-RAN, and optical access networks. A common thread runs through all of it: we want decision-making systems that don't just fit patterns in data but reason about cause and effect, so they generalize when conditions shift, stay interpretable, and come with guarantees wherever possible.

Causal AI for Networks

Modern networks are saturated with confounding. Traffic, interference, user behaviour, and control actions all influence one another, and a policy trained to exploit correlations often fails the moment conditions change. We build decision-making systems on explicit causal models, using structural causal models, counterfactual reasoning, and causal credit assignment so that what a system learns reflects the underlying mechanism rather than a spurious association. This makes policies more robust to distribution shift, more sample-efficient, and easier to interpret and audit.

Representative directions

  • causal and counterfactual methods for sequential decision-making
  • structural causal models as a basis for network control
  • robustness of learned policies and models under changing conditions
  • causal perspectives on credit assignment and attribution

Reinforcement Learning & Control

Many network problems, such as scheduling, resource allocation, and RAN control, are sequential decision problems under uncertainty, and reinforcement learning is a natural fit. Off-the-shelf RL, however, is sample-hungry and offers few guarantees, which is a poor match for systems with hard delay budgets and little tolerance for exploration. We develop RL methods tailored to these constraints: federated and multi-agent formulations for distributed network control, variance-reduction techniques that make credit assignment more reliable, and approaches informed by network calculus so that delay and backlog bounds remain provable.

Representative directions

  • reinforcement learning for scheduling and resource allocation under delay constraints
  • federated and multi-agent reinforcement learning for distributed control
  • learning-based control for Open RAN
  • RL methods informed by network calculus for provable performance guarantees

Game Theory & Mechanism Design

Next-generation networks are shared by many self-interested parties, including operators, tenants, slices, and users, each optimizing for itself. Treating them as a single controllable system is unrealistic; instead, we design the rules of interaction so that good system-wide outcomes emerge from strategic behaviour. Drawing on mechanism design and algorithmic game theory, we build incentive-compatible mechanisms for allocating network resources, with an emphasis on truthfulness, efficiency, and fairness in multi-tenant settings.

Representative directions

  • mechanism design for allocating network resources
  • auction-based spectrum and resource allocation
  • incentive-compatible and truthful protocols for multi-tenant networks
  • the interplay of incentives and learning in multi-agent systems

6G / O-RAN & Optical Access

Underlying every learning method we build is a real networking substrate. We work on the architecture and performance of 6G access and backhaul, including Open RAN and its intelligent controllers, optical access networks, and the delivery of demanding immersive applications. This thrust grounds the lab's algorithmic work in concrete systems. It is where models meet measured latency, dynamic bandwidth allocation, and standardized interfaces.

Representative directions

  • application-aware MAC scheduling for immersive media over optical access and 6G backhaul
  • intelligent control of Open RAN via the Non-RT and Near-RT RIC
  • QoS-aware network slicing
  • delay-constrained transport for latency-sensitive applications

For the full list of the lab's papers, see the Publications page.