/courses/machine-learning/deep-learning-foundations/Recent content in Deep Learning Foundations on DATATWEETSHugoenCopyright (c) 2025 DatatweetsFri, 14 Nov 2025 09:00:00 +0200/courses/machine-learning/deep-learning-foundations/lesson-1-introduction-to-neural-networks/Fri, 14 Nov 2025 09:00:00 +0200/courses/machine-learning/deep-learning-foundations/lesson-1-introduction-to-neural-networks/Discover the ideas behind neural networks: the biological inspiration, the artificial neuron, how layers stack into a network, what weights and biases do, why activation functions add nonlinearity, and how a network learns. You will meet the real Pima diabetes dataset as the running example for the whole module./courses/machine-learning/deep-learning-foundations/lesson-2-math-and-numpy-foundations-for-neural-networks/Fri, 14 Nov 2025 09:00:00 +0200/courses/machine-learning/deep-learning-foundations/lesson-2-math-and-numpy-foundations-for-neural-networks/Master the math behind neural networks using NumPy: vectors, matrices, the dot product, matrix multiplication, and broadcasting. You will standardize the real diabetes dataset, implement a single neuron, and build a full dense-layer forward pass with the X @ W + b pattern./courses/machine-learning/deep-learning-foundations/lesson-3-gradient-descent-for-neural-networks/Fri, 14 Nov 2025 09:00:00 +0200/courses/machine-learning/deep-learning-foundations/lesson-3-gradient-descent-for-neural-networks/Learn the loss function (binary cross-entropy), the gradient, and gradient descent, the algorithm that lets a network improve by stepping downhill. You will train a small numpy network on the real Pima Diabetes dataset and see how the learning rate controls whether learning crawls, converges, or blows up./courses/machine-learning/deep-learning-foundations/lesson-4-backpropagation/Fri, 14 Nov 2025 09:00:00 +0200/courses/machine-learning/deep-learning-foundations/lesson-4-backpropagation/Learn how backpropagation turns a single loss value into a gradient for every weight in your network. You will trace a forward pass that caches activations, then walk a backward pass through a 2-layer network on the real Pima Diabetes dataset, deriving dz2, dW2, da1, dz1, and dW1 by hand and in NumPy./courses/machine-learning/deep-learning-foundations/lesson-5-optimizers-sgd-rmsprop-and-adam/Fri, 14 Nov 2025 09:00:00 +0200/courses/machine-learning/deep-learning-foundations/lesson-5-optimizers-sgd-rmsprop-and-adam/Plain gradient descent works, but it is slow and noisy. In this lesson you will build momentum, RMSprop, and Adam from scratch in NumPy, understand their update equations, and compare all four optimizers training the same network on the real Diabetes dataset./courses/machine-learning/deep-learning-foundations/lesson-6-regularizing-neural-networks/Fri, 14 Nov 2025 09:00:00 +0200/courses/machine-learning/deep-learning-foundations/lesson-6-regularizing-neural-networks/Learn why deep networks overfit and how to fight back. You will watch a 64-unit network memorize the training data perfectly, then add L2 weight decay to your from-scratch NumPy network to close the train/test gap, and meet dropout and early stopping along the way.