Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Trackable Agent-based Evolution Models at Wafer Scale

Matthew Andres Moreno, Connor Yang, Emily Dolson, Luis Zaman

TL;DR

This work tackles scaling agent-based evolution to wafer-scale hardware while preserving phylogenetic observability. It introduces an asynchronous island-model GA tailored for the Cerebras Wafer-Scale Engine and a surface-based hereditary stratigraphy framework for post hoc phylogeny tracking, implemented in Cerebras Software Language. Benchmarking shows significant runtime improvements and detectable phylometric signals under adaptive dynamics, with emulated and on-hardware experiments validating end-state genome reconstructions. The results suggest the potential to reach quadrillions of replications per day at wafer scale, enabling new inquiries in evolutionary biology and artificial life on emerging HPC platforms.

Abstract

Continuing improvements in computing hardware are poised to transform capabilities for in silico modeling of cross-scale phenomena underlying major open questions in evolutionary biology and artificial life, such as transitions in individuality, eco-evolutionary dynamics, and rare evolutionary events. Emerging ML/AI-oriented hardware accelerators, like the 850,000 processor Cerebras Wafer Scale Engine (WSE), hold particular promise. However, practical challenges remain in conducting informative evolution experiments that efficiently utilize these platforms' large processor counts. Here, we focus on the problem of extracting phylogenetic information from agent-based evolution on the WSE platform. This goal drove significant refinements to decentralized in silico phylogenetic tracking, reported here. These improvements yield order-of-magnitude performance improvements. We also present an asynchronous island-based genetic algorithm (GA) framework for WSE hardware. Emulated and on-hardware GA benchmarks with a simple tracking-enabled agent model clock upwards of 1 million generations a minute for population sizes reaching 16 million agents. We validate phylogenetic reconstructions from these trials and demonstrate their suitability for inference of underlying evolutionary conditions. In particular, we demonstrate extraction, from wafer-scale simulation, of clear phylometric signals that differentiate runs with adaptive dynamics enabled versus disabled. Together, these benchmark and validation trials reflect strong potential for highly scalable agent-based evolution simulation that is both efficient and observable. Developed capabilities will bring entirely new classes of previously intractable research questions within reach, benefiting further explorations within the evolutionary biology and artificial life communities across a variety of emerging high-performance computing platforms.

Trackable Agent-based Evolution Models at Wafer Scale

TL;DR

This work tackles scaling agent-based evolution to wafer-scale hardware while preserving phylogenetic observability. It introduces an asynchronous island-model GA tailored for the Cerebras Wafer-Scale Engine and a surface-based hereditary stratigraphy framework for post hoc phylogeny tracking, implemented in Cerebras Software Language. Benchmarking shows significant runtime improvements and detectable phylometric signals under adaptive dynamics, with emulated and on-hardware experiments validating end-state genome reconstructions. The results suggest the potential to reach quadrillions of replications per day at wafer scale, enabling new inquiries in evolutionary biology and artificial life on emerging HPC platforms.

Abstract

Continuing improvements in computing hardware are poised to transform capabilities for in silico modeling of cross-scale phenomena underlying major open questions in evolutionary biology and artificial life, such as transitions in individuality, eco-evolutionary dynamics, and rare evolutionary events. Emerging ML/AI-oriented hardware accelerators, like the 850,000 processor Cerebras Wafer Scale Engine (WSE), hold particular promise. However, practical challenges remain in conducting informative evolution experiments that efficiently utilize these platforms' large processor counts. Here, we focus on the problem of extracting phylogenetic information from agent-based evolution on the WSE platform. This goal drove significant refinements to decentralized in silico phylogenetic tracking, reported here. These improvements yield order-of-magnitude performance improvements. We also present an asynchronous island-based genetic algorithm (GA) framework for WSE hardware. Emulated and on-hardware GA benchmarks with a simple tracking-enabled agent model clock upwards of 1 million generations a minute for population sizes reaching 16 million agents. We validate phylogenetic reconstructions from these trials and demonstrate their suitability for inference of underlying evolutionary conditions. In particular, we demonstrate extraction, from wafer-scale simulation, of clear phylometric signals that differentiate runs with adaptive dynamics enabled versus disabled. Together, these benchmark and validation trials reflect strong potential for highly scalable agent-based evolution simulation that is both efficient and observable. Developed capabilities will bring entirely new classes of previously intractable research questions within reach, benefiting further explorations within the evolutionary biology and artificial life communities across a variety of emerging high-performance computing platforms.
Paper Structure (20 sections, 8 figures)

This paper contains 20 sections, 8 figures.

Table of Contents

  1. Introduction
  2. Progress Toward Scale-up in Artificial Life
  3. Untapped Emerging Hardware
  4. Maintaining Observability
  5. Decentralized Phylogenetic Tracking
  6. Contributions
  7. Methods
  8. Distributed Phylogenetic Tracking
  9. Surface-based Hereditary Stratigraphy Algorithms
  10. WSE Architecture and Programming Model
  11. Asynchronous Island-model Genetic Algorithm
  12. Genome Model
  13. Software and Data Availability
  14. Results and Discussion
  15. Surface Algorithm Benchmark on Conventional CPU
  16. ...and 5 more sections

Figures (8)

  • Figure 1: Strategy for trackable distributed evolution simulation.Hstrat markers (🧬) attached to replicating agents (🦠) in a many-CPU runtime (left panel) enable post hoc estimation of relatedness between lineages, enabling approximate phylogenetic reconstruction (right panel).
  • Figure 2: Surface-based hereditary stratigraphy implementations.Visualizations of steady (left) and tilted (right) surface site selection policies. Top-row heatmaps show evolution of time-since-last-deposition for each site on a 256-bit field over the course of 4,096 time steps. The bottom row shows retention spans for 3,000 ingested time points. Vertical lines span durations between ingestion and elimination for differentia appended at successive time points. Time points previously eliminated are marked in red. Time elapses from bottom to top in both visualizations.
  • Figure 3: Column vs. surface-based hereditary stratigraphy.Contrast of existing sorted-order "column"-based stratum retention framework with proposed explicitly addressed "surface"-based approach.
  • Figure 4: Island model GA implementation for WSE.Neighboring PEs exchange agents (🦠) via asynchronous send/receive operations from dedicated buffers ("migration"), with on-completion callbacks setting "ready" flags to copy between main population and ready buffer.
  • Figure 5: Hereditary stratigraphy algorithm benchmarks.Comparison of per-generation operation time for column- and surface-based steady and tilted retention policies, lower is better. Top and bottom panels show Python and Zig implementations, respectively. Trivial is a simple hardcoded retention decision, provided as a baseline control. Background hatching indicates significant outcomes (Mann-Whitney U test; $n=20$).
  • ...and 3 more figures