F-Race and Iterated F-Race: An Overview

By Mauro Birattari, Zhi Yuan, Prasanna Balaprakash, Thomas Stützle. Experimental Methods for the Analysis of Optimization Algorithms, 2010

455 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computer Networks and Communications, Industrial and Manufacturing Engineering

View Paper: https://www.semanticscholar.org/paper/2fae91efed5573379669d40d6e9ae9625a4a3861

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Networks and Communications

In Industrial and Manufacturing Engineering

Citations in the corpus

Hyperparameter Search in Machine Learning (2015, methodology): “A wide variety of optimization methods have been used for hyperparameter search, including particle swarm optimization [26, 27], genetic algorithms [32], coupled simulated annealing [33] and racing algorithms [6].”
The irace package: Iterated racing for automatic algorithm configuration (2016, methodology): “I/F-Race was proposed [4, 8] as an extension of F-Race in order to search the configuration space more effectively by focusing on the most promising configurations.”
Hyperparameters and tuning strategies for random forest (2018, methodology): “other more sophisticated methods such as F-Race (Birattari et al., 2010), general simulated annealing (Bohachevsky et al.”
Automatically improving the anytime behaviour of optimisation algorithms (2014, methodology): “Iterated F-race (I/F-Race) [8, 13] is a method for automatic configuration that consists of three steps: (1) sampling new configurations according to a probability distribution, (2) selecting the best configurations from the newly sampled ones by means of F-Race, and (3) updating the probability dis”
Feature-based tuning of simulated annealing applied to the curriculum-based course timetabling problem (2014, methodology): “First, we ran an F-Race tuning (Birattari et al., 2010) of our algorithm over the training instances with a 95% confidence level, in order to establish a baseline for further comparisons.”
Parameter tuning for configuring and analyzing evolutionary algorithms (2011, background): “For instance, [4] uses the names qualitative, and quantitative parameters, [5] distinguishes symbolic and numeric parameters, [6] calls them categorical and numeri-”
Identifying Key Algorithm Parameters and Instance Features Using Forward Selection (2013, methodology): “Specifically, it has been convincingly demonstrated that methods for automated algorithm G. Nicosia and P. Pardalos (Eds.)”
Evolutionary Bilevel Optimization for Complex Control Tasks (2015, background): “These include racing algorithms that try to aggressively prune out bad parameters [20].”
Feature-based tuning of single-stage simulated annealing for examination timetabling (2015, methodology): “A subsequent F-Race limited to the failing instances suggested that, while for most parameters the winning value found in the preliminary race is also good for the failing instances, t 0 and w H must be tuned differently.”
Tunability: Importance of Hyperparameters of Machine Learning Algorithms (2018, background): “These search strategies range from simple grid or random search (Bergstra and Bengio, 2012) to more complex, iterative procedures such as Bayesian optimization (Hutter et al., 2011; Snoek et al., 2012; Bischl et al., 2017) or iterated F-racing (Birattari et al., 2010; Lang et al., 2017).”
An Efficient Approach for Assessing Hyperparameter Importance (2014, methodology): “…(Adenso-Diaz and Laguna, 2006; Nannen and Eiben, 2007; Hutter et al., 2009; Ansotegui et al., 2009) and racing algorithms (Maron and Moore, 1994; Birattari et al., 2010), recently, Bayesian optimization methods based on random forest models have been shown to compare favourably (Hutter et al.,…”
Hyperparameter optimization: Foundations, algorithms, best practices, and open challenges (2021, methodology): “The iterated racing (IR, Birattari et al., 2010) procedure is a general algorithm configuration method that optimizes for a configuration of a general (not necessarily ML) algorithm that performs well over a given distribution of (arbitrary) problems.”
A Survey of Automatic Parameter Tuning Methods for Metaheuristics (2020, background): “Details about how to update the probabilistic model in iterated F -Race could be found in [51] and [63].”