liveBook · Manning

Pairwise and population search

As explained in Self-Improving Agents

Search methods that compare candidates pairwise or select among a population of candidate variants.

Pairwise and population search keeps multiple candidates in play: it generates variants and uses a signal to select among them, exploring more of the search space than a single-threaded hill-climber. section 1.4.2

Examples include:

  • SPO compares candidates pairwise; GEPA breeds reflectively from a population using genetic-Pareto search. section 1.4.2
  • MIPROv2 proposes instruction candidates using a Bayesian model of what tends to work; AFlow searches candidate workflows in a tree. section 1.4.2
Figure 1.11 The search family arranged by cost against improvement reach. Hill-climbing is the cheapest and most limited, pairwise and population methods cost more and reach further, reinforcement-learning searches fit policy-shaped artifacts, and evolutionary archive search is the most expensive and reaches the furthest. Meta-search sits apart, since it changes the search itself rather than spending more effort on the same curve.

Compare this family’s cost and improvement reach with other search approaches. section 1.4.2

How Self-Improving Agents covers it

How deep it goes
Defines the family and names representative methods, without detailing their implementation procedures.
Examples
Names pairwise SPO, reflective genetic-Pareto GEPA, Bayesian-model-guided MIPROv2, and tree-search AFlow.
What you can do after
Evaluate its trade-off with hill-climbing: it explores more of the search space at higher cost.
How it connects
Connects to hill-climbing, SPO, GEPA, MIPROv2, and AFlow as neighboring or representative search approaches.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Self-Improving Agents. The section links point to the parts of the book it draws on.