liveBook · Manning

Global optimum

As explained in Build AI Drug Discovery Pipelines

A global minimum is a point at which a function attains its lowest value over its entire domain.

A global optimum is the point where a function reaches its lowest value across its entire domain. In logistic regression for cytochrome P450 inhibition, minimizing binary cross-entropy seeks the parameter values with the smallest aggregate prediction error. section 5.1.1

The binary cross-entropy cost is convex, so gradient descent is guaranteed to reach a global minimum when the learning rate avoids instability and enough parameter updates are made. There is no closed-form solution for this cost function. section 5.1.1

Prediction quality and loss

For a true class of 1, predicted probabilities of 0.3 and 0.8 produce losses of 1.20 and 0.22, respectively; the lower loss indicates the better prediction. section 5.1.1

How Build AI Drug Discovery Pipelines covers it

How deep it goes
Explains convexity, learning-rate stability, update count, and the absence of a closed-form solution.
Examples
Compares binary cross-entropy losses for predicted probabilities 0.3 and 0.8 when the true class is 1.
What you can do after
Apply gradient descent to minimize a convex binary cross-entropy cost under suitable learning-rate and update conditions.
How it connects
Connects global optima with logistic regression, binary cross-entropy, convex functions, sigmoid predictions, and gradient descent.

Also explained in

Compare how 6 books explain global optimum →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Build AI Drug Discovery Pipelines. The section links point to the parts of the book it draws on.