Model learning with personalized interpretability estimation (ML-PIE)

Virgolin, Marco; Lorenzo, Andrea De; Randone, Francesca; Medvet, Eric; Wahde, Mattias

doi:10.1145/3449726.3463166

Cited by 16 publications

(10 citation statements)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…𝛼-dom. ), and a simple extension of NSGA-II as mentioned in [79], where non-dominated sorting assigns an artificial worst-possible rank to duplicate solutions. We refer to the latter as NSGA-II with penalization of duplicates, NSGA-II+PD in short.…”

Section: Methodsmentioning

confidence: 99%

“…For NSGA-II applied to discrete optimization, in [30] strategies are explored to remove duplicate solutions from the population. One such strategy is used for MOGP in [79], where NSGA-II is modified so that duplicate solutions are assigned the lowest priority to survive selection. Together with classic NSGA-II, SPEA2, and the 𝛼-dominance based algorithms, we also include this algorithm in our comparisons.…”

Section: Prior Work On Improving Nsga-ii For Gpmentioning

confidence: 99%

“…At the end of a single run of MOGP, decision makers can choose the model that strikes the right balance between accuracy and interpretability. Since interpretability is hard or impossible to define (in general terms) [45,79], the common way by which interpretability in pursued in MOGP for SR is by minimization of solution size (or derivations thereof, see e.g., the related work section in [78]), i.e., the number of components that constitutes the solution. Minimizing size is typically in conflict with maximizing accuracy in (MO)GP, because (MO)GP typically discovers better solutions by refining the function approximation they represent, i.e., by incorporating additional components [40].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Evolvability Degeneration in Multi-Objective Genetic Programming for Symbolic Regression

Liu¹,

Virgolin²,

Alderliesten³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Genetic programming (GP) is one of the best approaches today to discover symbolic regression models. To find models that trade off accuracy and complexity, the non-dominated sorting genetic algorithm II (NSGA-II) is widely used. Unfortunately, it has been shown that NSGA-II can be inefficient: in early generations, lowcomplexity models over-replicate and take over most of the population. Consequently, studies have proposed different approaches to promote diversity. Here, we study the root of this problem, in order to design a superior approach. We find that the over-replication of low complexity-models is due to a lack of evolvability, i.e., the inability to produce offspring with improved accuracy. We therefore extend NSGA-II to track, over time, the evolvability of models of different levels of complexity. With this information, we limit how many models of each complexity level are allowed to survive the generation. We compare this new version of NSGA-II, evoNSGA-II, with the use of seven existing multi-objective GP approaches on ten widely-used data sets, and find that evoNSGA-II is equal or superior to using these approaches in almost all comparisons. Furthermore, our results confirm that evoNSGA-II behaves as intended: models that are more evolvable form the majority of the population. Code: https://github.com/dzhliu/evoNSGA-II CCS CONCEPTS• Computing methodologies → Genetic programming; • Applied computing → Multi-criterion optimization and decisionmaking.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Prior Work On Improving Nsga-ii For Gpmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Evolvability Degeneration in Multi-Objective Genetic Programming for Symbolic Regression

Liu¹,

Virgolin²,

Alderliesten³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…Moreover, researchers should strive to include user studies, to assess whether the proposed objective (interpretability, trust, etc.) obtains the desired effect in practice [32].…”

Section: Open Problems and Possible Directions In Gp For Imlmentioning

confidence: 97%

“…In our experience in clinical applications [28,29,30], obtaining a good objective function of what the user truly needs requires several sessions of interaction. A promising direction here is to use ML itself to learn what specific users find to be more or less interpretable in an automatic fashion [31,32]. Moreover, researchers should strive to include user studies, to assess whether the proposed objective (interpretability, trust, etc.)…”

Section: Open Problems and Possible Directions In Gp For Imlmentioning

confidence: 99%

Less is More: A Call to Focus on Simpler Models in Genetic Programming for Interpretable Machine Learning

Virgolin¹,

Medvet²,

Alderliesten³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Interpretability can be critical for the safe and responsible use of machine learning models in high-stakes applications. So far, evolutionary computation (EC), in particular in the form of genetic programming (GP), represents a key enabler for the discovery of interpretable machine learning (IML) models. In this short paper, we argue that research in GP for IML needs to focus on searching in the space of low-complexity models, by investigating new kinds of search strategies and recombination methods. Moreover, based on our experience of bringing research into clinical practice, we believe that research should strive to design better ways of modelling and pursuing interpretability, for the obtained solutions to ultimately be most useful.

show abstract