[论文解读] Abrupt and spontaneous strategy switches emerge in simple regularised neural networks

Anika T. Löwe, Léo Touzo+4 more|arXiv (Cornell University)|Feb 22, 2023

Neural dynamics and brain function被引用 6

一句话总结

该论文表明，通过渐进式随机梯度下降，极简且正则化的神经网络中可自发涌现出突然、具有洞察感的策略切换——其特征为性能突然提升、选择性出现以及延迟可变。该机制源于权重更新中的噪声与L1正则化门控，后者抑制了‘沉默知识’，直至突然释放后才实现快速性能提升，从而在无需专门认知机制的情况下模拟人类洞察，其机制与人类洞察行为高度相似。

ABSTRACT

Humans sometimes have an insight that leads to a sudden and drastic performance improvement on the task they are working on. Sudden strategy adaptations are often linked to insights, considered to be a unique aspect of human cognition tied to complex processes such as creativity or meta-cognitive reasoning. Here, we take a learning perspective and ask whether insight-like behaviour can occur in simple artificial neural networks, even when the models only learn to form input-output associations through gradual gradient descent. We compared learning dynamics in humans and regularised neural networks in a perceptual decision task that included a hidden regularity to solve the task more efficiently. Our results show that only some humans discover this regularity, whose behaviour was marked by a sudden and abrupt strategy switch that reflects an aha-moment. Notably, we find that simple neural networks with a gradual learning rule and a constant learning rate closely mimicked behavioural characteristics of human insight-like switches, exhibiting delay of insight, suddenness and selective occurrence in only some networks. Analyses of network architectures and learning dynamics revealed that insight-like behaviour crucially depended on a regularised gating mechanism and noise added to gradient updates, which allowed the networks to accumulate "silent knowledge" that is initially suppressed by regularised (attentional) gating. This suggests that insight-like behaviour can arise naturally from gradual learning in simple neural networks, where it reflects the combined influences of noise, gating and regularisation.

研究动机与目标

• 探究类洞察行为（通常被视为一种独特的认知现象）是否可从纯粹渐进式学习中在人工神经网络中涌现。
• 确定极简且正则化的神经网络是否能再现人类洞察的核心行为特征：突然性、选择性与可变延迟。
• 探讨噪声、正则化与门控机制在实现无显式重构或元认知过程的突发策略切换中的作用。
• 检验‘沉默知识’——即最初被注意力门控抑制但功能完整的表征——在延迟释放后是否可导致性能突然提升。

提出的方法

• 使用带有两个输入节点、一个输出节点以及每个输入对应乘法门控的极小神经网络，通过在门控上施加L1正则化的随机梯度下降进行训练。
• 在梯度更新中注入噪声以模拟随机性，从而探索潜在表征的多样性。
• 网络通过隐藏规律学习输入与输出的关联，其中最优性能依赖于利用基于颜色的规则。
• 通过将S型函数拟合到性能轨迹并识别陡峭拐点来检测类洞察切换，这些拐点指示了策略的突然采用。
• 利用参与者自述报告与基于模型的分类方法验证人类中的洞察检测，并将相同方法应用于神经网络。
• 使用贝叶斯信息准则（BIC）与保护性超越概率比较模型拟合（线性、阶跃、S型），S型模型在捕捉突发转变方面表现更优。

Figure 1: Stimuli, task design and insight classification procedure (A) Stimuli and stimulus-response mapping: dot clouds were either coloured in orange or purple and moved to one of the four directions NW, NE, SE, SW with varying coherence. A left response key, "X", corresponded to the NW/SE motion

实验结果

研究问题

• RQ1 在通过标准随机梯度下降训练的简单神经网络中，是否可涌现出类洞察行为（定义为突然性、选择性与延迟）？
• RQ2 噪声、正则化与门控机制在实现无显式重构的突发策略切换中扮演何种角色？
• RQ3 是否存在‘沉默知识’——即被抑制但功能完整的表征——在延迟后被激活的证据？
• RQ4 极简神经网络的学习动态在定性与定量上与人类在感知决策任务中的洞察行为如何比较？

主要发现

• 采用极简架构（两个输入、一个输出、门控权重）的L1正则化神经网络表现出类洞察的策略切换，其特征为突然性、选择性与延迟，与人类行为模式高度一致。
• S型模型在人类与网络的性能轨迹中均提供了最佳拟合，保护性超越概率表明其对线性与阶跃模型具有显著偏好。
• 类洞察切换之前存在‘沉默知识’——一种功能完整但被抑制的表征——仅在梯度更新的噪声导致的随机释放后才被激活。
• 梯度更新中的噪声至关重要：即使存在正则化，若无噪声则不会发生类洞察切换。
• 表现出类洞察行为的网络比例具有选择性（并非所有网络均切换），且切换时间各不相同，反映了人类中观察到的可变延迟。
• 模型识别出的类洞察受试者与人类参与者自报使用颜色规则的重叠率达79.6%，验证了检测方法的有效性。

Figure 2: Humans: task performance and insight-like strategy switches (A) Accuracy (% correct) during the motion phase increases with increasing motion coherence. N = 99, error bars signify standard error of the mean (SEM). (B) Accuracy (% correct) over the course of the experiment for all motion co