Supervised learning trains a parameterized model from examples pairing each input with a known desired output, also called a target or ground truth. The model adjusts its parameters so its outputs move closer to those targets, then applies the learned function to previously unseen inputs during inferencing. section 1.1 section 7.1
A typical supervised-learning workflow is:
- Design a parameterized model architecture, such as a weighted-sum function. section 1.1
- Prepare labeled pairs of sampled inputs and desired outputs; manual labeling may be the most arduous step. section 1.1 section 7.1
- Iteratively adjust weights so predictions on training inputs more closely match their corresponding targets. section 1.1 section 7.1
- Deploy the trained model for inferencing on inputs it has never seen before, while treating correctness as probabilistic rather than guaranteed. section 7.1
Threat estimation from hardness and sharpness
A linear model computes threat as y(hardness, sharpness) = w0 × hardness + w1 × sharpness + b. For an input with hardness 0.01 and sharpness 0.02, a prediction of 50.3 can be moved toward the target −0.9 by reducing the bias; repeated adjustments across examples bring the parameters toward useful values. Threat scores then support decisions such as approach, ignore, or run away. section 1.1
| y(hardness, sharpness) = w0 × hardness + w1 × sharpness + b |
Use the equation to connect labeled examples with a concrete parameterized prediction function. section 1.1
