Distributions of the neuronal weights calculated by different methods, including single neuron correlation analysis, model sensitivity analysis, support vector machine recursive feature elimination, mutual information, and local-learning-based method. Neurons on the left side of the vertical dash line show real recordings. Neurons on the right are simulated and generated independently from the task. The five approaches show the different abilities of eliminating noisy neurons. The weights of the simulated neurons, learning from our proposed method, are all close to 0, while many of them are assigned with relatively large values by other methods [70].
