@machinelearningtogo: ➡️Part 7 of learning ML code from scratch: One of the most important concepts that allows our model to bend and understand complex patterns. No linearity anymore. Without this one curve, every neural network would collapse into a linear thing. Every neuron does the same three things. First the sum: each input times its weight, all added up, plus the bias. That gives one number, z. Second the curve: z is plugged into the activation function, and the height of the curve at z is the neuron's output, a. Third the hand over: a is simply the input for the next neuron, times a weight, into the next sum, into the next curve. In the video the numbers are real, taken from the trained XOR network: input 1 and 0, weights minus 5.6 and 5.7, bias 2.9, so z is minus 2.7, the sigmoid turns it into 0.06, and the answer neuron turns that into 0.95, a clear yes. There are many curves, each with a job. Sigmoid squashes everything to between 0 and 1, a confidence. Tanh squashes to between minus 1 and 1. ReLU cuts off everything below zero and keeps the rest. Softmax picks one of many. ChatGPT uses GELU, a smooth cousin of ReLU. So a neuron is a sum, a curve, and a hand over. The curve is what makes a network more than a line. #machinelearning #neuralnetworks #activationfunction #ai #coding

QuantMachineLearn

Open In TikTok:

Region: DE

Sunday 20 September 2026 09:13:54 GMT