Approximate Exploitability: Learning a Best Response

Abstract

Recent learning-based image fusion methods have marked numerous progress in pre-registered multi-modality data, but suffered serious ghosts dealing with misaligned multi-modality data, due to the spatial deformation and the difficulty narrowing cross-modality discrepancy. To overcome the obstacles, in this paper, we present a robust cross-modality generation-registration paradigm for unsupervised misaligned infrared and visible image fusion (IVIF). Specifically, we propose a Cross-modality Perceptual Style Transfer Network (CPSTN) to generate a pseudo infrared image taking a visible image as input. Benefiting from the favorable geometry preservation ability of the CPSTN, the generated pseudo infrared image embraces a sharp structure, which is more conducive to transforming cross-modality image alignment into mono-modality registration coupled with the structure-sensitive of the infrared image. In this case, we introduce a Multi-level Refinement Registration Network (MRRN) to predict the displacement vector field between distorted and pseudo infrared images and reconstruct registered infrared image under the mono-modality setting. Moreover, to better fuse the registered infrared images and visible images, we present a feature Interaction Fusion Module (IFM) to adaptively select more meaningful features for fusion in the Dual-path Interaction Fusion Network (DIFN). Extensive experimental results suggest that the proposed method performs superior capability on misaligned cross-modality image fusion.

Computing equilibria by minimizing exploitability with best-response ensembles

“…Unlike extensive-form fictitious play [Heinrich et al, 2015] and counterfactual regret minimization [Zinkevich et al, 2007], their convergence result pertains to the strategies being optimized rather than the time-average strategies. Timbers et al [2022] introduced approximate exploitability, which uses approximate best responses computed through a combination of search and reinforcement learning. It generalizes a domain-specific technique for poker called local best response [Lisý and Bowling, 2017].…”

“…ψ is non-negative and zero precisely at Nash equilibria. It is also known as the NashConv in the literature [Lanctot et al, 2017a;Lockhart et al, 2019;Walton and Lisy, 2021;Timbers et al, 2022], and is the standard measure of closeness to Nash equilibrium. Our goal is to find strategy profiles with low exploitability.…”

Learning Near-Optimal Intrusion Responses Against Dynamic Attackers

“…Convergence metric for T-FP. To estimate the convergence of the sequence of strategy pairs generated by T-FP, we use the approximate exploitability metric δ [102]:…”

The Path to AGI Goes through Embodiment

“…One simple example is chess, where rule-based AI surpassed humans in 1997 (Campbell, Hoane Jr, and Hsu 2002), eventually followed by RL-based AI methods (Silver et al 2018). Nobody would claim that superhuman chess or Go algorithms can do anything other than play the given games, and even simple tweaks to the game rules, and unusual or adversarial strategies can throw the algorithms off (Lan et al 2022;Timbers et al 2020;Wang et al 2022). Even deep RL algorithms that can master multiple Atari games (Mnih et al 2013(Mnih et al , 2015 are still ultimately constrained to a certain subset of game types.…”