TUM.ai Deep Dive into Neural Networks: Understanding Depth & Width

Ever wondered what really happens inside a neural network and why certain architectures work better than others? Join us for a hands-on deep dive into the mathematics behind neural networks, hosted by TUM.ai members! We’ll explore the meaning of depth and width in neural networks, combining mathematical intuition with practical experiments in NumPy and PyTorch. 🔍 What we’ll cover: Understanding Depth: Forward and backward propagation, Jacobians, residual connections, and the Neural ODE perspective. Understanding Width: Infinite-width networks, Neural Tangent Kernels (NTK), and how network width influences learning dynamics. From Theory to Practice: Deriving gradients from scratch, implementing neural network components, and understanding how to debug and train deep networks. Expect a mix of mathematical foundations, theoretical insights, hands-on coding, and plenty of room for questions and discussions. 👥 Who can join? Whether you’re already part of TUM.ai or interested in joining our community, you’re welcome! This session is especially suited for those with some background in mathematics, machine learning, or neural networks who want to deepen their understanding. Come learn, exchange ideas, and connect with fellow AI enthusiasts! 🚀