Local Optima in Diversity Optimization: Non-trivial Offspring Population is Essential

Denis Antipov,Aneta Neumann,Frank Neumann

2024-07-12

Abstract:The main goal of diversity optimization is to find a diverse set of solutions which satisfy some lower bound on their fitness. Evolutionary algorithms (EAs) are often used for such tasks, since they are naturally designed to optimize populations of solutions. This approach to diversity optimization, called EDO, has been previously studied from theoretical perspective, but most studies considered only EAs with a trivial offspring population such as the $(\mu + 1)$ EA. In this paper we give an example instance of a $k$-vertex cover problem, which highlights a critical difference of the diversity optimization from the regular single-objective optimization, namely that there might be a locally optimal population from which we can escape only by replacing at least two individuals at once, which the $(\mu + 1)$ algorithms cannot do. We also show that the $(\mu + \lambda)$ EA with $\lambda \ge \mu$ can effectively find a diverse population on $k$-vertex cover, if using a mutation operator inspired by Branson and Sutton (TCS 2023). To avoid the problem of subset selection which arises in the $(\mu + \lambda)$ EA when it optimizes diversity, we also propose the $(1_\mu + 1_\mu)$ EA$_D$, which is an analogue of the $(1 + 1)$ EA for populations, and which is also efficient at optimizing diversity on the $k$-vertex cover problem.

Neural and Evolutionary Computing,Artificial Intelligence

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: in diversity optimization, when using a simple offspring population (such as (µ + 1) EA), the algorithm may get trapped in a local optimal solution and be unable to escape. Specifically, the paper shows that in a k - vertex cover problem instance, there exists a locally optimal population. To escape from this population and improve diversity, at least two individuals must be replaced simultaneously, but (µ + 1) EA can only replace one individual at a time, so it cannot be achieved.

To solve this problem, the paper proposes and studies the following points:

1. **The importance of complex offspring populations**:
- The paper proves through specific examples that simple offspring populations (such as (µ + 1) EA) cannot effectively find the global optimal solution in some cases.
- It is proposed to use more complex offspring populations (such as (µ + λ) EA), where \( \lambda \geq \mu \), which can effectively find diverse populations.

2. **New algorithm variants**:
- To overcome the difficulties of (µ + λ) EA in selecting subsets, the paper proposes a new algorithm variant (1µ + 1µ) EAD, which is similar to (1 + 1) EA but applicable to the population space.
- This new algorithm also shows efficient performance on the k - vertex cover problem.

3. **Theoretical analysis**:
- The paper conducts a theoretical analysis of the proposed algorithms and proves that they can find the population with optimal diversity within the expected time under specific conditions.
- Specifically, if \( k\mu = o(\sqrt{n}) \), then (1µ + 1µ) EAD and (µ + λ) EAD (where \( \lambda \geq \mu \)) have a probability of at least \( 2^{-k\mu(1 - o(1))} \) of finding the population with optimal diversity in each iteration, and their running time is dominated by the geometric distribution \( \text{Geom}(2^{-k\mu(1 - o(1))}) \).

### Key formulas

- **Total Hamming distance**:
Given the bit strings of a population \( P \), the total Hamming distance is defined as:
\[
D(P)=\sum_{x,y \in P} H(x, y)
\]
where \( H(x, y) \) is the Hamming distance, and the sum is taken over all unique unordered pairs of individuals.

- **Position contribution**:
The total Hamming distance can be calculated by the number of 1s in each position:
\[
D(P)=\sum_{i = 1}^{n} n_i (|P| - n_i)
\]
where \( n_i \) is the number of individuals in population \( P \) with 1 in the \( i \)-th position.

### Conclusion

The main contribution of the paper is to reveal the limitations of simple offspring populations in diversity optimization and propose an effective solution, that is, using non - trivial offspring populations (such as (µ + λ) EA or (1µ + 1µ) EAD). These algorithms can efficiently find the population with optimal diversity under specific conditions, thus solving the local optimal problem in diversity optimization.