Inventors of the original artificial neural networks (ANNs) derived their inspiration from biology [1]. However, today, most ANNs, such as backpropagation-based convolutional deeplearning networks, resemble natural NNs only superficially. Given that, on some tasks, such ANNs achieve human or even superhuman performance, why should one care about such dissimilarity with natural NNs? The algorithms of natural NNs are relevant if one's goal is not just to outperform humans on certain tasks but to develop general-purpose artificial intelligence rivaling that of a human. As contemporary ANNs are far from achieving this goal and natural NNs, by definition, achieve it, natural NNs must contain some "secret sauce" that ANNs lack. This is why we need to understand the algorithms implemented by natural NNs.

中文翻译：

---

受神经科学启发的在线无监督学习算法：人工神经网络

原始人工神经网络 (ANN) 的发明者从生物学 [1] 中获得灵感。然而，今天，大多数 ANN，例如基于反向传播的卷积深度学习网络，仅在表面上类似于自然 NN。鉴于在某些任务上，这样的 ANN 达到了人类甚至超人的表现，为什么要关心与自然 NN 的这种差异呢？如果一个人的目标不仅仅是在某些任务上超越人类，而是开发与人类相媲美的通用人工智能，那么自然 NN 的算法是相关的。由于当代 ANN 远未实现这一目标，而根据定义，自然 NN 可以实现这一目标，因此自然 NN 必须包含一些 ANN 所缺乏的“秘方”。这就是为什么我们需要了解自然神经网络实现的算法。