https://sankethbk.github.io/blog/tags/weights/Recent content in Weights on Sanketh's BlogHugo -- 0.169.0en-usMon, 19 Oct 2026 00:00:00 +0530https://sankethbk.github.io/blog/posts/ml/labs/2026-10-19-lab-1-mnist-from-pixels-to-neural-network/Mon, 19 Oct 2026 00:00:00 +0530https://sankethbk.github.io/blog/posts/ml/labs/2026-10-19-lab-1-mnist-from-pixels-to-neural-network/<h1 id="lab-11-what-do-the-weights-of-a-neural-network-actually-represent">Lab 1.1: What Do the Weights of a Neural Network Actually Represent?</h1> <p><strong>MNIST internals series</strong> — the practical companion to the <a href="https://sankethbk.github.io/blog/posts/ml/">ML notes</a> (softmax &amp; cross-entropy, backprop, training a 2-layer network). This time we stop deriving and start <em>looking</em>.</p> <p>Anyone who has trained a model eventually asks:</p> <blockquote> <p>The model is just a pile of numbers. In this lab&rsquo;s case, exactly 7,860 of them (7,850 weights + 10 biases). <strong>What do those numbers actually <em>mean</em>?</strong></p> </blockquote>