Excited to share a glimpse of my recent session at IISc in Prof. Anirban Chakraborty's course, where we looked into the basics of Deep Learning. We started with the feedforward aspect of a neural network, exploring how it is essentially a composition of functions, with each function representing a module or a layer. In this process, the input is transformed into activations, which ultimately become the output. At this stage, we also consider the label or the ground truth and compute the loss. This loss provides a scalar value indicating our satisfaction with the output. We then backpropagate the gradient of the loss with respect to the output throughout our network, using this gradient to update weights. Each module performs two key computations: the gradient of the output with respect to the input, and the gradient of the output with respect to the weights. These gradients are then multiplied with the gradient coming in from the next layer. This process exemplifies how the chain rule functions. It offers a straightforward yet profound insight into how neural networks learn and evolve. Eager to hear your thoughts and insights on this fundamental aspect of AI and machine learning!