Inicialización Kaiming
Inicialización de Kaiming, también conocida como Inicialización He, is a technique used to initialize the weights of redes neuronales, particularly those employing funciones de activación like ReLU (Rectified Linear Unit). Proper inicialización de pesos is crucial as it can significantly affect the convergence speed and overall performance of the network during training.
The method was introduced by Kaiming He and his collaborators in a 2015 paper. It aims to address the issues of vanishing and la explosión de gradientes that can occur during the training of deep networks. These issues arise when weights are initialized too small or too large, leading to ineffective learning.
En la Inicialización de Kaiming, los pesos se extraen de una distribución gaussiana with a mean of 0 and a variance that is inversely proportional to the number of input units (fan-in) for each neuron. Specifically, the weights are initialized using the formula:
w ~ N(0, sqrt(2 / fan_in))Where w represents the weight, N denotes a distribución normal, and fan_in is the number of input connections to the neuron. This approach helps maintain a balanced distribution of activations throughout the layers of the network, promoting effective gradient flow during backpropagation.
Kaiming Initialization is particularly effective in deep networks with ReLU and its variants but can be adapted for other activation functions with slight modifications. As a result, it has become a standard practice in training aprendizaje profundo modelos, contribuyendo a tasas de convergencia más rápidas y estables.