AI 中文总结
该研究提出基于传递熵的信息论框架，量化感官与行为的双向信息流，可预测生物与人工系统的导航效率，揭示了驱动导航的通用行为-环境反馈回路。
AI 中文摘要
当生物在环境中导航以定位关键资源时，其行为动作必须与感官输入紧密耦合。本文提出一种基于信息论的框架，利用传递熵量化这种耦合，传递熵用于测量感官输入与行为输出之间的信息流。从感官输入到行为的信息流定义了导航策略的“反应性”组件，而从行为到感官输入的信息流则定义了“主动性”组件，即动作会塑造后续的感官体验。分析这种双向信息流既能预测导航性能，又能从轨迹中剖析导航策略。通过一个捕捉主动和反应组件的最小模型，我们将宏观性能与微观信息流关联起来。随后，我们将该框架应用于细菌、蠕虫、苍蝇的实验测量轨迹，以及在感官环境中导航的机器学习智能体轨迹。在所有系统中，双向信息流均可可靠预测导航效率，揭示了一种通用的行为-环境反馈回路。分解主动和反应信息流还进一步揭示了细菌趋化性背后的不同策略、苍蝇嗅觉导航中导航策略的空间依赖性，以及强化学习智能体学习到的策略。总之，这些结果确立了双向信息流作为理解生物和人工系统导航的统一原则。
英文摘要
As organisms navigate the environment to locate critical resources, their behavioral actions must be tightly coupled to their sensory inputs. Here, we introduce an information-theoretic framework that quantifies this coupling using transfer entropy, which measures information flow between sensory inputs and behavioral outputs. Information flow from sensory inputs to behavior defines a "reactive" component of a navigational strategy, whereas information flow from behavior to sensory inputs defines an "active" component, whereby actions shape subsequent sensory experiences. Analyzing these bidirectional information flows enables us to both predict navigational performance and dissect navigation strategies from trajectories. Using a minimal model that captures the active and reactive components, we connect macroscopic performance to microscopic information flows. We then apply the framework to experimentally measured trajectories of bacteria, worms, and flies, as well as to machine learning agents navigating sensory landscapes. Across systems, bidirectional information flow reliably predicts navigation efficiency, revealing a common behavioral-environment feedback loop. Decomposing active and reactive information flows further exposes distinct strategies underlying bacterial chemotaxis, the spatial dependency of the navigation strategy in fly olfactory navigation, and the learned policies of a reinforcement-trained agent. Together, these results establish bidirectional information flow as a unifying principle for understanding navigation in biological and artificial systems.
Comments14 pages, 6 figures; supplementary information included
