登録して招待リンクを共有すると、動画再生報酬と紹介報酬を獲得できます。

Mathematica
@mathemetica
Math isn't escape. It's the map through the madness.
参加 October 2024
204 フォロー中    77.4K ファン
One neuron, by the numbers. Inputs 0.2, 0.5, 0.43 times weights 0.4, 0.26, 0.3 give 0.08 + 0.13 + 0.129. Add bias 0.1: z = 0.439. Activation: σ(z) = 1/(1 + e⁻ᶻ), so σ(0.439) ≈ 0.608. That number is passed to every neuron in the next layer. Training picks the weights w that minimise the average loss (1/n) Σ ℓ(h(xᵢ; w), yᵢ), by stepping down the loss surface along −∇L. Learning is weighted sums, a squashing function, and gradient descent.
もっと見る