神经科学与心理学2019cell.com
大脑中误差反向传播的几种理论
Theories of Error Back-Propagation in the Brain
进阶 · 约 90 分钟读完前置知识：反向传播算法、神经可塑性基础、预测编码
导读
这篇综述探讨了大脑是否以及如何实现类似反向传播（backpropagation）的误差信号传递机制，梳理了生物神经系统中可能支持梯度式学习的各种理论方案（如反馈连接、预测编码等）。文章试图弥合机器学习中的 BP 算法与真实神经可塑性之间的差距。
HN 关注点：HN 读者长期关注'大脑是否做反向传播'这一神经科学与深度学习的交叉问题，讨论焦点在于生物可行性以及 BP 是否被高估为 AI 的万能解法。
HN 评论精选
这几条评论里：质疑 3 亲历 1
亲历作者声称自己的实验表明反向传播会自然产生 STDP 行为，并提供了开源实现。
针对'脉冲时序依赖可塑性（STDP）与其他模型的关系需要进一步澄清'这一点，我认为我发现了反向传播会催生 STDP 的现象，并开源了相关项目（用 PyTorch 实现脉冲神经网络，展示 BP 引出 STDP）。
GChevalier查看英文原文
> "The relationship of spike-time-dependent plasticity to other models requires further clarifying work" - I got something for that, in fact I think I discovered that backpropagation engenders STDP: https://github.com/guillaume-chevalier/Spiking-Neural-Network-SNN-with-PyTorch-where-Backpropagation-engenders-STDP
质疑认为生物学智能的关键在于演化塑造的神经架构而非学习算法，BP 的作用被高估了。
有趣的是，大多数 AI 模型关注学习，但生物学中繁重的工作其实是在发育演化和神经发生阶段完成的。你不可能用反向传播教会你的狗相对论——然而我们却把 BP 当作能解决 AI 所有问题的方式。在更多关注神经架构之前，我不担心 AI 统治世界。
rdlecler1查看英文原文
It’s interesting that most AI models focus on learning, but in biology the heavy lifting was all done in the developmental evolution and neurogenesis. You’re not going to teach your dog relativity with back propagation—-yet we treat BP like it was going to solve all the problems in AI. Until we start focusing more on neural architecture I have no worries about AI taking over the world.
质疑质疑大脑与 BP 的类比缺乏实际科学价值，指出许多生物神经现象是该框架无法解释的。
我也做过类似的类比尝试，但从未得出有用的结果——这类研究如何推动科学前进？我唱个反调：怎么解释抑制性神经元？紧张性神经元？大脑听一次名字就能回忆的能力？程序性记忆？你指出的这些关系能帮助理解这些现象吗？
mjpuser查看英文原文
I have been on similar kicks for drawing similarities between the two, but it never pans out to anything useful. I.e how does this move the science forward? I can play devil's advocate (aka classic hacker news' commenter) and say what about inhibiting neurons? Tonic neurons? What about the ability for the brain to do recall after hearing a name once? Procedural memory? Does the relationship you point out help you understand any of these things?
质疑神经生物学家从生物物理角度质疑：信号在胞体整合后来源信息丢失，误差如何精确归因到具体突触是 BP 生物实现的根本难题。
作为神经生物学家，我对此很感兴趣。先说我的担忧：设想一个网络，3000 个输入神经元各自与下游神经元形成约 10 个突触连接，某个下游神经元的树突上有 3 万个突触。这些输入有的强有的弱，有的来自远端树突（信号到达胞体时已大幅衰减），但它们都在胞体（轴丘）整合，一旦达到阈值就发放动作电位——此时信号来源已无从区分。如果这个动作电位在网络下游某处导致了误差，按监督学习反向传播的规则，引发该误差脉冲的那些突触应该……（评论未完）
subroutine查看英文原文
As a neurobiologist, this interests me. I'm leaving this comment because I don't have time to read the article rn, but what to come back and see what HN thinks about it. Briefly, and naive to the content of the article, here is my concern. Consider this network architecture... You have 3000 input neurons that each makes ~10 synaptic connections with many downstream neurons, but let's focus on just one of those downstream neurons. This downstream neuron thus has 30k synaptic connections along its dendrite. At some time t its cell body (specifically the axon hillock area) receives enough graded electrical potential from some number of those synapses to fire an 'action potential' (a single electrical impulse). Some of those inputs contributed bursts of weak inputs, some may have contributed several strong inputs, some of the inputs were from synapses on very distal regions of the dendrite (therefore the graded signal was significantly weakened by the time it reached the cell body) some were from synapses very near the cell body. But they all integrate at the cell body, at which point, it doesn't matter where they came from - they've arrived is all that matters now - and together their strength is enough to evoke an action potential. Let's say this action potential impulse resulted in an error of some sort, somewhere downstream in the network. If you're playing by the rules of supervised learning with backprop, the synapses that evoked the signal producing the error-impulse should be made weaker. How? In biological NNs this is the impossible question. Signals that could target individual synapses do not propagate up axons through cell bodies up dendrites and back to synapses (you do have slow signaling that is for homeostatic scaling, but this is thought to simply scale network input up or down across all synapses). This means you will need another group of neurons to send error signals back. Maybe these recurrent projections exist, and say they did, these recurrent projections would need to (1) form a 3rd party connection at every synapse and (2) know which synapses were to blame for the error. Those don't seem like trivial phenomena from a neurophysiology perspective. I've read theories on local molecular tags or peptide synthesis, but they all seemed very much hand-wavy. Anything that has gained traction on those hypotheses hasn't stood up to scrutiny. The simplest theory to me, is that cortical networks on the size of human brains don't explicitly 'forget'. That is, they only learn. When you call someone Bob, and they correct you saying their name is Mike, your neural nets don't erase Bob and replace it with Mike. They remember the old name, the new name, the embarrassing situation, all of it. Biological neural nets don't need to explicitly forget because forgetting is (unfortunately?) a byproduct of being a biological entity. Also, neurons have finite resources (they are essentially zero-sum), so you simply will not see infinite run-up of synaptic strengths (which you might in some artificial NN without setting certain hyperparameters). Together, this suggests that recency is a fairly dominant factor when it comes to associative learning (and what gets forgotten) in biological NN.
相关资源
从这篇论文所有 HN 讨论帖的全部评论里整理出来的链接。
历次讨论
- 2019Hacker News 讨论帖160 分 · 33 评
同领域还值得读
神经科学家能理解微处理器吗？（2017）
Could a Neuroscientist Understand a Microprocessor? (2017)
提出一个思想实验：把神经科学常用的分析方法（损伤实验、电生理、功能定位等）应用到一个我们完全了解其原理的工程系统——6502 微处理器（运行游戏）上，检验这些方法能否真正“理解”…
神经科学与心理学2017 · 411 分 · 135 评进阶 · 约 60 分钟编辑精选大脑衰老呈现非线性转变，提示中年“关键窗口”
Brain aging shows nonlinear transitions, suggesting a midlife "critical window"
这篇 PNAS 论文研究大脑衰老的非线性变化，发现大脑老化并非匀速进行，而是在中年（约 40-60 岁）存在一个关键的"临界窗口期"，此时干预可能有效。
神经科学与心理学2025 · 295 分 · 159 评进阶 · 约 75 分钟编辑精选社交网络中大规模情绪传染的证据（2014）
Evidence of massive-scale emotional contagion through social networks (2014)
这篇 2014 年发表于 PNAS 的论文研究了 Facebook 上约 70 万用户的信息流中，减少正面或负面内容是否会传染性地影响用户自己发布内容的情绪。
神经科学与心理学2020 · 285 分 · 72 评入门 · 约 45 分钟编辑精选裸盖菇素使人类大脑去同步化
Psilocybin desynchronizes the human brain
这篇 Nature 研究用个体化的 fMRI 精密成像追踪受试者服用裸盖菇素（psilocybin）前后的脑功能连接变化，发现药物会造成大范围的脑网络「去同步化」，尤其是默认模式…
神经科学与心理学2024 · 273 分 · 259 评进阶 · 约 60 分钟编辑精选对智力的赞扬会削弱儿童的动机与表现（1998）
Praise for Intelligence Can Undermine Children's Motivation (1998)
这篇 1998 年的经典论文（Dweck 等人）通过实验研究表扬方式对儿童动机的影响：夸孩子“聪明”（表扬智力）会让他们回避挑战、害怕失败，而表扬努力则更能培养坚持性。
神经科学与心理学2016 · 253 分 · 155 评入门 · 约 45 分钟编辑精选