Farahmandi Arefeh, Abedi Khoozani Parisa, Blohm Gunnar
Centre for Neuroscience Studies, Queen's University, Kingston, Ontario, Canada, K7L 3N6.
J Neurosci. 2025 Aug 29. doi: 10.1523/JNEUROSCI.0104-25.2025.
The integration of multiple sensory inputs is essential for human perception and action in uncertain environments. This process includes reference frame transformations as different sensory signals are encoded in different coordinate systems. Studies have shown multisensory integration in humans is consistent with Bayesian optimal inference. However, neural mechanisms underlying this process are still debated. Different population coding models have been proposed to implement probabilistic inference. This includes a recent suggestion that explicit divisive normalization accounts for empirical principles of multisensory integration. However, whether and how divisive operations are implemented in the brain is not well understood. Indeed, all existing models suffer from the curse of dimensionality and thus fail to scale to real-world problems. Here, we propose an alternative model for multisensory integration that approximates Bayesian inference: a multilayer-feedforward neural network of multisensory integration (MSI) across different reference frames trained on the analytical Bayesian solution. This model displays all empirical principles of multisensory integration and produces similar behavior to that reported in Ventral Intraparietal (VIP) neurons in the brain. The model achieved this without a neatly organized and regular connectivity structure between contributing neurons, such as required by explicit divisive normalization. Overall, we show that simple feedforward networks of purely additive units can approximate optimal inference across different reference frames through parallel computing principles. This suggests that it is not necessary for the brain to use explicit divisive normalization to achieve multisensory integration. This research presents an alternative model to divisive normalization models of multisensory integration in the brain. Our study demonstrates that a feed-forward neural network can achieve optimal multisensory integration across different reference frames without explicitly implementing divisive operations, challenging the long-held assumption that such operations are necessary for multisensory integration. The model displays all the empirical principles of multisensory integration, producing similar behavior to that reported in Ventral Intraparietal (VIP) neurons in the brain. This work offers profound insights into the putative neural computations underlying multisensory processing.
在不确定的环境中，多种感官输入的整合对于人类的感知和行动至关重要。这一过程包括参考系转换，因为不同的感官信号是在不同的坐标系中编码的。研究表明，人类的多感官整合与贝叶斯最优推理一致。然而，这一过程背后的神经机制仍存在争议。人们提出了不同的群体编码模型来实现概率推理。这包括最近的一种观点，即明确的除法归一化解释了多感官整合的经验原则。然而，大脑中是否以及如何实现除法运算尚不清楚。事实上，所有现有的模型都受到维度诅咒的困扰，因此无法扩展到现实世界的问题。在这里，我们提出了一种用于多感官整合的替代模型，该模型近似贝叶斯推理：一个跨不同参考系的多感官整合（MSI）多层前馈神经网络，它是在解析贝叶斯解的基础上进行训练的。该模型展示了多感官整合的所有经验原则，并产生了与大脑中腹侧顶内（VIP）神经元所报告的类似行为。该模型在没有像明确的除法归一化所要求的那样，在参与的神经元之间具有整齐组织和规则连接结构的情况下实现了这一点。总体而言，我们表明，纯粹加法单元的简单前馈网络可以通过并行计算原则在不同参考系中近似最优推理。这表明大脑没有必要使用明确的除法归一化来实现多感官整合。这项研究提出了一种替代大脑中多感官整合除法归一化模型的模型。我们的研究表明，前馈神经网络可以在不明确实施除法运算的情况下，在不同参考系中实现最优的多感官整合，这挑战了长期以来认为此类运算对于多感官整合是必要的假设。该模型展示了多感官整合的所有经验原则，并产生了与大脑中腹侧顶内（VIP）神经元所报告的类似行为。这项工作为多感官处理背后的假定神经计算提供了深刻的见解。
