Like our brains, Artificial Neural Networks are a collection of Artificial Neurons.
To make predictions, Neuron requires three parameters.
1. Inputs → Numerical inputs
2. Weights → Each inputs are associated with a weight.
3. Bias → Neurons cautious number.
To better understand, a movie say "The Penguin" received 5 user ratings and let's take an average rating for all of the below inputs. With these inputs, let's find out whether this movie will be recommended by the Neuron or not.
User Rating → x1 = 4.8/5
Genre → x2 = 0.8 (80% match for a specific genre like sci-fi)
Watch time → x3 = 0.9 or 90% completion rate
Now with these inputs, weights are assigned to each input. These weights are assigned first randomly, then over the period of multiple backpropagation, the neural network becomes smart enough to improve the weights. For now, let's assign random weights for each of the inputs.
w1 → 0.7
w2 → 0.3
w3 → 0.4
bias → 0.5
Calculation :
→(4.8x0.7) + (0.8 x 0.3) + (0.9 x 0.4) - 0.5 = 3.36+ 0.24 + 0.36 - 0.5 = 3.46 ~ 3.5
Using the sigmoid formula:
f(x) = 1 / (1 + e^(-x)), let's assign the value 3.5 obtained in the above step to x.
f(x) = 1 / (1 + e^(-x)) with x = 3.5
e^(-3.5) ≈ 0.0302866
1 + 0.0302866 = 1.0302866
1 / 1.0302866 ≈ 0.9706
As you can see the output is 0.97 which is 97%.
Activation Mechanism:
The Neuron outputs a value of 0.9706 (or 97.06%), which can also be interpreted as a strong activation in favour of recommending the movie. In simpler terms, the neuron is highly confident (about 97% sure) that it should recommend this movie.
This is how Netflix, YouTube and other platforms make their initial predictions before adding complex layers.
Hope this clarifies how individual Neuron helps Neural networks in making predictions. Remember, neural networks consists of many such Neurons working together in interconnected layers.
#ai #llm #learnthebasics
