Active learning reveals underlying decision strategies

Parpart, Paula; Schulz, Eric; Speekenbrink, Maarten; Love, Bradley C.

doi:10.1101/239558

Cited by 8 publications

(11 citation statements)

References 40 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In category learning, Markant and Gureckis (2014) found that active learners sampled more along the line of the category boundaries and performed better than passive learners. Parpart, Schulz, Speekenbrink, and Love (2017) found that participants' queries in a feature-based active learning task with binary outcomes were more in line with a weight-based strategy than a rank-based strategy.…”

Section: Active Learningmentioning

confidence: 89%

Active Function Learning

Jones

Schulz²,

Meder

et al. 2018

Preprint

Self Cite

View full text Add to dashboard Cite

How do people actively explore to learn about functional rules, that is, how continuous inputs map onto continuous outputs? We introduce a novel paradigm to investigate information search in continuous, multi-feature function learning scenarios. Participants either actively selected or passively observed information to learn about an underlying linear function. We develop and compare different variants of rule-based (linear regression) and non-parametric (Gaussian process regression) active learning approaches to model participants' active learning behavior. Our results show that participants' performance is best described by a rule-based model that attempts to efficiently learn linear functions with a focus on high and uncertain outcomes. These results advance our understanding of how people actively search for information to learn about functional relations in the environment.

show abstract

Section: Active Learningmentioning

confidence: 89%

Active Function Learning

Jones

Schulz²,

Meder

et al. 2018

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…where τ is a free temperature parameter. We followed previous work (Wu et al, 2018;Parpart, Schulz, Speekenbrink, & Love, 2017) The results of this analysis revealed a mean pseudo-R 2 of 0.041 over all orders and degrees, which was low but significantly better than chance (t(33) = 20.52, p < 0.001 d = 1.86, BF > 100). Moreover, the estimated median temperature parameter was τ = 1.02, indicating a relatively wide spread of predictions.…”

Section: Computational Modelingmentioning

confidence: 84%

“…The general predictive performance of many models was relatively similar and rather low. This might be due to the overall complexity of choices, since there were 216 possible options on every trial, making it difficult to compare among candidate models (also see Parpart et al, 2017).…”

Section: Discussionmentioning

confidence: 99%

Exploring the space of human exploration

Schulz

Bertram

Hofer

et al. 2019

Preprint

Self Cite

View full text Add to dashboard Cite

What drives people's exploration in complex scenarios where they have to actively acquire information? How do people adapt their selection of queries to the environment? We explore these questions using Entropy Mastermind, a novel variant of the Mastermind code-breaking game, in which participants have to guess a secret code by making useful queries. Participants solved games more efficiently if the entropy of the game environment was low; moreover, people adapted their initial queries to the scenario they were in. We also investigated whether it would be possible to predict participants' queries within the generalized Sharma-Mittal information-theoretic framework. Although predicting individual queries was difficult, the modeling framework offered important insights on human behavior. Entropy Mastermind opens up rich possibilities for modeling and behavioral research.

show abstract

“…A recent study by Parpart, Schulz, Speekenbrink, and Love (2017) also focused on interactions between strategy selection and learning the structure of the environment. They formulated active learning versions of WADD and TTB, assuming that people using the former will learn cue weights and the latter cue order.…”

Section: Interactions Between Strategy Selection and Cue Weight Learningmentioning

confidence: 99%

Linear integration and lexicographic models of choice: A cue weight learning perspective.

Stojić

Olsson

Analytis

2020

Journal of Experimental Psychology: Learning, Memory, and Cogni

View full text Add to dashboard Cite

Choosing between options characterized by multiple cues can be a daunting task. People may integrate all information at hand or just use lexicographic strategies that ignore most of it. Notably, integrative strategies require knowing exact cue weights, whereas lexicographic heuristics can operate by merely knowing the importance order of cues. Here we study how using integrative or lexicographic strategies interacts with learning about cues. In our choicelearning-estimation paradigm people first make choices, learning about cues from qualities of chosen options, and then estimate qualities of new options. We developed delta-elimination (DE), a new lexicographic strategy that generalizes previous heuristics to any type of environment, and compared it to the integrative weighted-additive (WADD) strategy. Our results show that participants learned cue weights regardless of the choice strategy employed. The group of people best described by the DE strategy learned cue weights, exactly like the group best described by the WADD. Still, there was an interaction between the adopted strategy and the cue weight learning process: the DE users learned cue weights slower than the WADD users. This work advances the study of lexicographic choice strategies, both empirically and theoretically, and deepens our understanding of strategy selection, in particular the interaction between the strategy used and learning the structure of the environment.

show abstract

Active learning reveals underlying decision strategies

Cited by 8 publications

References 40 publications

Active Function Learning

Active Function Learning

Exploring the space of human exploration

Linear integration and lexicographic models of choice: A cue weight learning perspective.

Contact Info

Product

Resources

About