量子储备池中时间信息处理的能量代价
Energetic Cost of Temporal Information Processing in Quantum Reservoirs
AI总结：
研究量子储备池中时间信息处理的能量代价，推导弱相互作用下平均切换功的解析表达式，明确能量代价与性能的关联，阐明能量效率与计算性能的兼容条件。
AI中文摘要：
量子储备池计算通过以最小的训练开销处理时间信息，为高能效机器学习提供了一条有前景的途径。然而，将其能量代价与计算性能关联起来的物理原理在很大程度上仍未被探索。本文表明，在相互作用的自旋储备池中，信息编码和信息处理受不同的物理机制支配。在弱相互作用区域，我们推导了平均（切换）功的解析表达式，表明编码新输入的能量代价由储备池单元的局域响应决定。相反，相互作用主要重新分配编码的信息，产生记忆和非线性特征，同时仅对功产生微弱影响。这种分离在代表性线性和非线性基准任务中，使能量代价与性能之间产生了相反的关联。我们的结果确定了切换功是信息编码的能量特征，并阐明了能量效率与计算性能何时兼容。
英文摘要：
Quantum reservoir computing offers a promising route toward energy-efficient machine learning by processing temporal information with minimal training overhead. Yet, the physical principles linking its energetic cost to computational performance remain largely unexplored. Here we show that, in an interacting spin reservoir, information encoding and information processing are governed by distinct physical mechanisms. In the weak interacting regime, we derive an analytical expression for the average (switching) work, showing that the energetic cost of encoding new inputs is determined by the local response of the reservoir units. In contrast, interactions primarily redistribute the encoded information, generating memory and nonlinear features while only weakly affecting the work. This separation produces opposite correlations between energetic cost and performance for representative linear and nonlinear benchmark tasks. Our results identify the switching work as the energetic signature of information encoding and clarify when energetic efficiency and computational performance are compatible.
