标题： 学习解决信用分配问题 显示英文标题

标题： Learning to solve the credit assignment problem

摘要： 反向传播正在驱动当今的人工神经网络（ANNs）。 然而，尽管进行了大量研究，仍然不清楚大脑是否实现了这个算法。 在神经科学家中，强化学习（RL）算法常被视为一种现实的替代方案：神经元可以随机引入变化，并使用非特定的反馈信号来观察其对成本的影响，从而近似其梯度。 然而，这种学习的收敛速度随着参与神经元数量的增加而显著下降。 在此，我们提出了一种混合学习方法。 每个神经元使用一种类似RL的策略，学习如何近似反向传播所提供的梯度。 我们提供了证明，说明我们的方法对于某些类型的网络可以收敛到真实的梯度。 在前馈网络和卷积网络中，我们通过实验表明，我们的方法能够学习近似梯度，并且可以达到或超过基于精确梯度的学习性能。 学习反馈权重提供了一种生物上合理的机制，以实现良好的性能，而无需精确的、预先指定的学习规则。 显示英文摘要

摘要： Backpropagation is driving today's artificial neural networks (ANNs). However, despite extensive research, it remains unclear if the brain implements this algorithm. Among neuroscientists, reinforcement learning (RL) algorithms are often seen as a realistic alternative: neurons can randomly introduce change, and use unspecific feedback signals to observe their effect on the cost and thus approximate their gradient. However, the convergence rate of such learning scales poorly with the number of involved neurons. Here we propose a hybrid learning approach. Each neuron uses an RL-type strategy to learn how to approximate the gradients that backpropagation would provide. We provide proof that our approach converges to the true gradient for certain classes of networks. In both feedforward and convolutional networks, we empirically show that our approach learns to approximate the gradient, and can match or the performance of exact gradient-based learning. Learning feedback weights provides a biologically plausible mechanism of achieving good performance, without the need for precise, pre-specified learning rules.