liveBook · Manning

Neighborhood function

As explained in Build AI Drug Discovery Pipelines

A neighborhood function assigns the magnitude of a node update according to the node's distance from the best matching unit.

A neighborhood function controls how strongly a self-organizing map updates each node according to its distance from the best matching unit (BMU). The Gaussian form gives nearby nodes larger updates and distant nodes smaller ones. section 7.1.2

h(i,BMU,t)=exp(-d(i,            BMU)22σ(t)2)

The Gaussian equation makes the distance effect explicit: increasing distance reduces a node’s update, while the width σ controls how broadly the BMU influences the grid. section 7.1.2

Self-organizing molecular maps

During repeated SOM training, the BMU and neighboring nodes move toward each selected vector. The learning rate and neighborhood width typically decrease over time, shifting from broad organization to fine-tuning. The same approach can organize high-dimensional Morgan-fingerprint compounds on a two-dimensional rectangular grid. section 7.1.2

How Build AI Drug Discovery Pipelines covers it

How deep it goes
Defines the Gaussian calculation and its distance effect, then connects width and learning-rate schedules to training without deriving further internals.
Examples
Uses an RGB bead-and-grid analogy and Morgan-fingerprint compounds trained with MiniSom.
What it assumes
Assumes vectors, node distance, and BMUs, while defining the neighborhood function and update roles directly.
What you can do after
Apply it through MiniSom by selecting a Gaussian neighborhood and tuning its width alongside the learning rate.
How it connects
Connects self-organizing maps with best matching units, learning rate, Gaussian functions, topology, and Morgan fingerprints.

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.