# Generating Random Parameters in Feedforward Neural Networks with Random Hidden Nodes: Drawbacks of the Standard Method and How to Improve It

## TL;DR

This paper critiques the standard uniform distribution method for initializing neural network parameters, highlighting its drawbacks, and proposes an improved approach that preserves nonlinear regions and enables uniform slope distribution.

## Contribution

It introduces a novel method for generating random neural network parameters that enhances nonlinearity and slope uniformity compared to traditional approaches.

## Key findings

- The new method maintains nonlinear regions within the input hypercube.
- It allows for activation functions with uniformly distributed slope angles.
- The approach improves the expressiveness of randomly initialized neural networks.

## Abstract

The standard method of generating random weights and biases in feedforward neural networks with random hidden nodes, selects them both from the uniform distribution over the same fixed interval. In this work, we show the drawbacks of this approach and propose a new method of generating random parameters. This method ensures the most nonlinear fragments of sigmoids, which are most useful in modeling target function nonlinearity, are kept in the input hypercube. In addition, we show how to generate activation functions with uniformly distributed slope angles.