AI Terminology Course
AI Terminology
/
Advanced

Weights & Biases

Definition

The internal mathematical parameters that a neural network adjusts during training. Weights determine the strength of the connection between two neurons. Biases shift the activation threshold.

Explain Like I'm New

Weights are 'How much does this matter?'. If you are predicting house prices, the 'Square Footage' input will have a massive Weight (very important), and the 'Color of the front door' input will have a tiny Weight (not important). Biases are the baseline starting point.

Real World Example

When OpenAI says GPT-4 has 'Trillions of Parameters', they literally mean there are trillions of individual Weights and Biases inside the mathematical file. Training the AI is just the process of adjusting those trillions of numbers up and down slightly until the AI gets smart.

Common Use Cases

  • •Model parameters
  • •Mathematical foundations of ML

Interview Questions

basic

  • During the training process, does the AI adjust the data, or does it adjust the Weights and Biases?

intermediate

  • In the classic linear equation `y = mx + b`, which variable represents the Weight, and which represents the Bias?

Flash Cards

Question

What is adjusted?

Click to reveal answer
Answer

It adjusts the Weights and Biases. The data remains strictly unchanged.

Question

Weight vs Bias in equation?

Click to reveal answer
Answer

The slope `m` is the Weight (how heavily 'x' impacts the outcome). The y-intercept `b` is the Bias (shifting the baseline up or down).