Data-Centric Approach to Constrained Machine Learning - A Case Study on Conway’s Game of Life

Aug 3, 2026

Upload by Ezra

This paper presents a data-centric approach to constrained machine learning using Conway’s Game of Life. The study trains a minimal-parameter neural architecture to learn transition rules for multi-step prediction under strict limits on trainable parameters. Extensive quantitative evaluation shows that carefully designing the training dataset yields consistent gains in convergence and accuracy, largely independent of initialization and optimization choices. The results emphasize the practical value of domain-expert insights in building effective learning systems for real-world constrained settings.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages16

Reading Time40 min

AccessFree to download

Abstract

Introduction

Conway’s Game of Life

Problem Setup and Method

Training Data Design

Experimental Results

Quantitative Analysis

Discussion and Contributions

Data-Centric Approach to Constrained Machine Learning: A Case Study on Conway’s Game of Life Anton Bibin∗ Skoltech Skoltech Agro Moscow, Russia [a.bibin@skoltech.ru](a.bibin@skoltech.ru) Anton Dereventsov∗ Lirio LLC Behavioral Reinforcement Learning Lab Knoxville, TN, USA [adereventsov@lirio.com](adereventsov@lirio.com) arXiv :2408 . 12778v1 [ cs .LG] 23 Aug 2024 ABSTRACT This paper focuses on a data-centric approach to machine learning applications in the context of Conway’s Game of Life. Specifically, we consider the task of training a minimal architecture network to learn the transition rules of Game of Life for a given number of steps ahead, which is known to be challenging due to restrictions on the allowed number of trainable parameters. An extensive quantitative analysis showcases the benefits of utilizing a strategically designed training dataset, with its advantages persisting regardless of other parameters of the learning configuration, such as network initialization weights or optimization algorithm. Importantly, our findings highlight the integral role of domain expert insights in creating effective machine learning applications for constrained real-world scenarios. CCS CONCEPTS • Computing methodologies → Supervised learning by regression; Continuous space search; • Theory of computation → Pattern matching; Nonconvex optimization; • Information systems → Extraction, transformation and loading. KEYWORDS Data-Centric ML, Supervised Learning, Data Design, Game of Life ACM Reference Format: Anton Bibin and Anton Dereventsov. 2024. Data-Centric Approach to Constrained Machine Learning: A Case Study on Conway’s Game of Life. In Proceedings of The Second International Workshop on Resource-Efficient Learning for Knowledge Discovery (RelKD 2024) . ACM, New York, NY, USA, 16 pages. [https://doi.org/XXXXXXX.XXXXXXX](https://doi.org/XXXXXXX.XXXXXXX) 1 INTRODUCTION In this work we consider the problem of learning the rules of Conway’s Game of Life, posed as an image-to-image translation task. Such a setting is inspired by [57], where the authors investigate the ability of neural networks to learn the rules of Conway’s Game ∗ Equal contribution. Authors listed alphabetically. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for components of this work owned by others than the author(s) must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission [and/or a fee. Request permissions from permissions@acm.org](and/or a fee. Request permissions from permissions@acm.org). RelKD 2024, August 25–29, 2024, Barcelona, Spain © 2024 Copyright held by the owner/author(s) . Publication rights licensed to ACM. ACM ISBN 978-1-4503-XXXX-X/18/06. . . $15.00 [https://doi.org/XXXXXXX.XXXXXXX](https://doi.org/XXXXXXX.XXXXXXX) (a) State at time 0 (b) State at time 1 (c) State at time 2 (d) State at time 3 (e) State at time 4 (f) State at time 5 Figure 1: An example of a state trajectory in the Game of Life. Alive cells are white and dead cells are black. of Life from the given state transition images. The authors have observed that such a task is often not achievable by conventional machine learning approaches and is only attainable under a sufficient network overparameterization (about 5 − 10 times for a successful 1-or 2-step prediction). We restrict the setting by only working with minimally sufficient architectures and do not allow network overparameterization, which effectively puts us in a domain of constrained machine learning. To offset the lack of usual machine learning leniency, we allow for meticulous control of the training data. Specifically, we de

FAQs

What is the main learning task studied for Conway’s Game of Life?

The work formulates rule learning as an image-to-image translation task, training a minimal architecture to predict transitions ahead by a given number of steps.

Why is this problem considered difficult in conventional machine learning settings?

The paper notes that learning such transition rules usually requires sufficient network overparameterization; restricting to minimal architectures places the work in a constrained learning regime.

How does training data design affect performance under constrained settings?

A strategically constructed training dataset—rather than randomly generated boards—significantly improves convergence rate and speed, including for multi-step prediction tasks, regardless of initialization and optimization choices.

Aug 3, 2026

Upload by Ezra