A Proposed Infrastructure for Adding Online Interaction to Any Evolutionary Domain

Paul A. Szerlip, Kenneth O. StanleyarXiv 2014 · arXiv

Abstract

In the fields of Artificial Life (ALife) and Evolutionary Computation (EC), research often occurs in isolation, with experimental code frequently retired after publication. This lack of persistence and the high technical barrier to creating collaborative online platforms limit the potential for large-scale, human-in-the-loop discovery. This paper introduces the Worldwide Infrastructure for Neuroevolution (WIN), a software library designed to transform standalone evolutionary experiments into collaborative online platforms. By providing a standardized toolkit for storage, retrieval, and branching, WIN enables researchers to harness human intuition and "crowd-sourced" search. We demonstrate the library's versatility through two distinct domains: image evolution and creature morphology.

Introduction

Traditional evolutionary experiments are typically finite; a researcher defines a domain, runs an algorithm, and analyzes the results. However, collaborative systems like Picbreeder have demonstrated that human-computer interaction can navigate complex search spaces more effectively than automated algorithms alone. Despite their success, such systems are notoriously difficult to build, often requiring over a year of development for the underlying web infrastructure. This high barrier to entry prevents the broader research community from leveraging the power of online communities.

To address this, we propose the WIN library, which simplifies the process of connecting evolutionary domains to a persistent, web-based backend. The goal is to shift the paradigm from isolated, one-off experiments to "never-ending" searches where every discovery serves as a potential starting point for future exploration.

Methodology

The WIN library is built on a modern web stack designed for scalability and ease of integration.

Technology Stack

The library utilizes Node.js for its lightweight, event-driven architecture, making it ideal for handling concurrent web services. Data management is handled by MongoDB, a NoSQL database that facilitates the storage and cataloging of complex, non-relational research artifacts.

Branching and Phylogeny

The core innovation of WIN is its built-in logic for maintaining relationships between data points. Every object stored in the system represents a potential "branching point." When a user selects an artifact to evolve further, the system creates a new branch, automatically tracking the lineage. This creates a traceable history, or phylogeny, of the search process. This "stepping stone" approach ensures that progress is accumulated over time rather than reset for every new experimental run.

Demonstration Domains

To prove the library's versatility, we implemented two distinct evolutionary domains using the WIN infrastructure:

  1. win-Picbreeder: A reimplementation of the collaborative image evolution domain, where users evolve complex patterns through interactive selection.
  2. win-IESoR: A domain focused on evolving the morphology and control of two-dimensional "ambulating creatures" represented as graphs or networks.
Figure 1: Phylogenies generated by the WIN library for two different domains: evolved images (left) and evolved creature morphologies/networks (right).

Key Insights and Results

The primary achievement of this work is the successful creation of a prototype infrastructure that significantly reduces the labor required to build online evolutionary systems.

Democratizing Contribution

By lowering the technical barrier, WIN allows non-experts to contribute to neuroevolutionary research. The branching mechanic enables novice users to explore search spaces simply by selecting interesting results and observing where further evolution leads. This human-in-the-loop approach is particularly effective in domains where the objective function is subjective or difficult to define mathematically.

Phylogeny Generation

As shown in Figure 1, the system effectively tracks and visualizes the evolutionary history of user discoveries. The branching trees (phylogenies) illustrate how complex artifacts emerge from simpler ancestors through successive rounds of human-guided selection. This demonstrates that the same underlying library can manage vastly different data types—from pixel-based images to structural network graphs—while maintaining a consistent evolutionary logic.

Conclusion and Future Work

The WIN library represents a significant step toward making collaborative evolutionary search a standard tool for researchers. By providing a centralized repository (WIN Online), we aim to aggregate diverse evolutionary experiments into a single, accessible platform.

Future work will focus on improving the scalability of the infrastructure and exploring gamification strategies to maintain long-term user engagement. While the current prototype successfully lowers the barrier for web-based integration, further engineering is required to make the connection of high-performance computing clusters and complex physics simulators truly "trivial" for the average researcher.

Summary updated

Abstract

To address the difficulty of creating online collaborative evolutionary systems, this paper presents a new prototype library called Worldwide Infrastructure for Neuroevolution (WIN) and its accompanying site WIN Online (http://winark.org/). The WIN library is a collection of software packages built on top of Node.js that reduce the complexity of creating fully persistent, online, and interactive (or automated) evolutionary platforms around any domain. WIN Online is the public interface for WIN, providing an online collection of domains built with the WIN library that lets novice and expert users browse and meaningfully contribute to ongoing experiments. The long term goal of WIN is to make it trivial to connect any platform to the world, providing both a stream of online users, and archives of data and discoveries for later extension by humans or computers.
BibTeX
@misc{szerlip2014proposed,
  author = {Paul A. Szerlip and Kenneth O. Stanley},
  title = {{A Proposed Infrastructure for Adding Online Interaction to Any Evolutionary Domain}},
  year = {2014},
  eprint = {1407.3000},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1407.3000},
}
Figure