MATILDA is an online tool which makes available our new methodology known as Instance Space Analysis. It provides:

  • visualisations of the instance space for a problem
    • showing the location of benchmark test instances across the instance space
    • showing the strengths (“footprints”) and weaknesses of algorithms across the instance space
    • summarising the properties of the instances that an algorithm finds easy or hard
  • objective (unbiased) metrics of algorithmic power via footprint analysis
  • synthetic generation of new test instances at specific locations in the instance space (e.g. real-world-like instances, or instances with controllable properties)
  • automated algorithm selection tools to assist deployment of the best algorithm for a given instance.

Our efforts thus far have created instance space analyses for the following problem classes:

  1. Optimisation
    • Combinatorial Optimisation Problems
      • Graph Colouring
      • Travelling Salesman Problem
      • Knapsack Problem
      • Timetabling
    • Mathematical Programming
      • Linear Programming
      • Mixed Integer Programming
    • Continuous Optimisation
      • Black-box Single Objective
      • Black-box Multi-Objective
      • Job Shop Scheduling
  2. Learning and Model Fitting
    • Machine Learning
      • Classification
      • Regression
      • Clustering
    • Time Series Analysis
      • Time Series Forecasting
    • Image Analysis
      • Facial Age Estimation

If you have additional problems you would like to make available in MATILDA, please contact us (matilda-team@unimelb.edu.au)

VIDEO TUTORIAL: Introduction to Instance Space Analysis

USING MATILDA

The engine behind MATILDA is powered by MATLAB code, also available to download and run offline. MATILDA's online platform enables researchers to:

VIDEO TUTORIAL: How to Perform an Instance Space Analysis

Research Publications

The methodology used by MATILDA for visualizing and understanding the strengths and weaknesses of different algorithms has been originally described in a series of three papers focusing on graph colouring as a case study:

The early ideas for the instance space analysis methodology are summarised in three earlier papers:

Additional publications from the MATILDA team have applied this methodology to a wide variety of other problem domains, including: