Online Learning in Stackelberg Security Games with Adaptive Attacker Sequences and Time-Varying Attack Intensities

Shallow read  ·  2026  ·  source  ·  all reading

L-009 L-001

Online Learning in Stackelberg Security Games with Adaptive Attacker Sequences and Time-Varying Attack Intensities

Source: cs.GT updates on arXiv.org — https://arxiv.org/abs/2608.01703 Date read: 2026-09-02 Connected to: L-009, L-001 Kind: content Escalation: store-only Escalation rationale: [blank]

What this is

A game-theoretic learning paper deriving no-regret algorithms (√T regret bound) for a defender in repeated Stackelberg security games where attackers adapt their target selection and attack intensity over time. The work formalizes the defender's optimization as a mixed-integer linear program and integrates it with Follow-the-Perturbed-Leader under full-information feedback.

What I took from it

This is a competent algorithmic contribution within the security games domain, but it does not address the catastrophic risk dynamics that L-009 (Catastrophic Risk Cancellation in Symmetric Racing Protocols) is designed to capture. The paper assumes a single well-modeled defender with access to exact oracles and full-information feedback — a setting where optimization pressure is transparent and legible. It does not engage with what happens when multiple defenders race to deploy incompletely-understood security strategies, when the cost of first-mover disadvantage concentrates risk asymmetrically, or when verification of attack success/failure cascades across jurisdictions with misaligned incentives.

The triage note invokes L-009 and L-001, but the paper's concern is algorithmic convergence under adaptive adversaries, not the dynamics of protocol adoption under racing pressure or ossification under deployment lock-in. No mechanism addressed here — legible oracle access, full feedback, no deployment lag — maps onto the conditions where those laws would activate.

Research connections

  • L-001: No engagement with adoption pressure or modification difficulty; the security game formulation is static.
  • L-009: Does not model symmetric racing, concentrated prizes, or asymmetric cost distribution across competitors.
  • seed-073 (Correlated Failure Under Proxy Consensus): Tangentially relevant if defenders converge to the same mixed strategy; not explored.

Seed

Seed title: none