Adjusting Default Outputs with Bias

Bias in machine learning is an additional value that allows a model to adjust based on input data influences.

Alongside weights, it is one of the crucial elements that a model learns, enabling consistent output even in the absence of specific input values.

In simple terms, bias in a machine learning model adjusts the default output when there is no input value.

The formula can be represented as follows:

$y = w_1 x_1 + w_2 x_2 + ... + w_n x_n + b$

• w1, w2, ..., wn: Weights of each feature

• b: Bias

• x1, x2, ..., xn: Input values

Here, b helps the model adjust the predicted value in a consistent direction.

Without bias, the model's output would be zero when all input values are zero, making realistic predictions challenging in a machine learning model.

Weights and bias are both critical elements that a machine learning model learns, but their roles differ as outlined below.

Let's consider a simple model to predict a student's exam score.

• x: Hours studied

• w: Influence of study time on the score (weight)

• b: Expected base score even if the student hasn’t studied (bias)

$exam\_score = w \times hours\_studied + b$

If b = 0, a student who does not study at all would score 0.

However, if we assume that even without studying a student can get a base score of around 30, then b = 30.

In this way, bias helps machine learning models reflect data more realistically.

In artificial neural networks, each neuron has its own bias.

Bias enables a neuron to generate meaningful output even in the absence of input values, playing a crucial role in helping machine learning models learn accurate patterns.

In the next lesson, we'll tackle a quick quiz to review the concepts we've covered so far.