He Weight Initialization for Neural Networks
Initialization, Normalization & Regularization DS practice problem on Onlearn.
Difficulty: medium.
Topics: Understanding Neural Network Weight Initialization Strategies, ReLU Activation Stability, Fan-in calculation, Normal Distribution Sampling, Vanishing Gradient Problem, Weight Matrix Dimensionality, Deep Learning Foundations, Linear Algebra, Probability and Statistics, Numerical Optimization, Neural Network Architecture, Weight Initialization, Activation Functions, Gradient Flow, Variance Scaling, Tensor Operations.
Implement a function 'he initialize' that takes the number of input units (fan in) and output units (fan out) and returns a weight matrix of shape (fan in, fan out) initialized using the He Normal strategy. Use numpy for random generation with a fixed seed for reproducibility.