Active Function Learning

Jones, Angela; Schulz, Eric; Meder, Björn; Ruggeri, Azzurra

doi:10.1101/262394

Cited by 8 publications

(9 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…While the proposed sampling strategy we introduced is relatively inflexible, this policy could reflect the use of rational metareasoning (Lieder & Griffiths, 2017), with participants deploying a heuristic with a favourable trade-off between its utility in giving relatively informative evidence for a variety of common functional relationships (including the most a priori plausible, positive linear), while requiring little cognitive effort to adapt to existing sampled points. This also coheres with previous findings in active function learning, where participants' choices most closely fit a simpler linear regression policy rather than a generalized GP policy when learning in linear domains (Jones et al, 2018).…”

Section: Discussionsupporting

confidence: 88%

“…Previous work in active learning has suggested that people can effectively learn linear functions by focusing their sampling on regions of high uncertainty (Jones, Schulz, Meder, & Ruggeri, 2018). An uncertainty-based sampling policy could also be useful for non-linear functions, as maximizing information gain about the extrema of a function eliminates the need for extrapolation, which can otherwise be inaccurate in non-linear functions (DeLosh et al, 1997;Kalish et al, 2004).…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Sampling heuristics for active function learning

Gelpí¹,

Saxena²,

Lifchits³

et al. 2021

Preprint

View full text Add to dashboard Cite

People are capable of learning diverse functional relationships from data; nevertheless, they are most accurate when learning linear relationships, and deviate further from estimating the true relationship when presented with non-linear functions. We investigate whether, when given the opportunity to learn actively, people choose samples in an efficient fashion, and whether better sampling policies improve their ability to learn linear and non-linear functions. We find that, across multiple different function families, people make informative sampling choices consistent with a simple, low-effort policy that minimizes uncertainty at extreme values without requiring adaptation to evidence. While participants were most accurate at learning linear functions, those who more closely adhered to the simple sampling strategy also made better predictions across all non-linear functions. We discuss how the use of this heuristic might reflect rational allocation of limited cognitive resources.

show abstract

Section: Discussionsupporting

confidence: 88%

Section: Introductionmentioning

confidence: 99%

Sampling heuristics for active function learning

Gelpí¹,

Saxena²,

Lifchits³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…However, if they encode the abstract functional relation between the cause and its effect, they will be able to make novel causal predictions. Two studies that have investigated compositional reasoning in preschoolers (Piantadosi & Aslin, 2016) and noncausal function learning in adults (Jones, Schulz, Meder, & Ruggeri, 2018) begin to suggest that this may well be the case.…”

Section: Discussionmentioning

confidence: 99%

Transformations and Transfer: Preschool Children Understand Abstract Relations and Reason Analogically in a Causal Task

2020

View full text Add to dashboard Cite

Previous research suggests that preschoolers struggle with understanding abstract relations and with reasoning by analogy. Four experiments find, in contrast, that 3-and 4-year-olds (N = 168) are surprisingly adept at relational and analogical reasoning within a causal context. In earlier studies preschoolers routinely favored images that share thematic or perceptual commonalities with a target image (object matches) over choices that match the target along abstract relations (relational matches). The present studies embed such choice tasks within a causeand-effect framework. Without causal framing, preschoolers strongly favor object matches, replicating the results of previous studies. But with causal framing, preschoolers succeed at analogical transfer (i.e., choose relational matches). These findings suggest that causal framing facilitates early analogical reasoning. We gratefully acknowledge the Bezos Family Foundation for funding this research. We thank Rosie Aboody,

show abstract

“…As few tasks have combined all these components (but see [11] for active learning with probabilistic multidimensional stimuli), it remains unclear how people learn actively in an environment with complex rules (with multiple and potentially an unknown number of relevant dimensions) and probabilistic feedback. To study this, we developed a novel decision task: participants were asked to configure three-dimensional stimuli by choosing what features to use in each dimension, earning rewards that were probabilistically determined by features in a subset or all of these dimensions.…”

Section: Introductionmentioning

confidence: 99%

Humans combine value learning and hypothesis testing strategically in multi-dimensional probabilistic reward learning

et al. 2022

View full text Add to dashboard Cite

Realistic and complex decision tasks often allow for many possible solutions. How do we find the correct one? Introspection suggests a process of trying out solutions one after the other until success. However, such methodical serial testing may be too slow, especially in environments with noisy feedback. Alternatively, the underlying learning process may involve implicit reinforcement learning that learns about many possibilities in parallel. Here we designed a multi-dimensional probabilistic active-learning task tailored to study how people learn to solve such complex problems. Participants configured three-dimensional stimuli by selecting features for each dimension and received probabilistic reward feedback. We manipulated task complexity by changing how many feature dimensions were relevant to maximizing reward, as well as whether this information was provided to the participants. To investigate how participants learn the task, we examined models of serial hypothesis testing, feature-based reinforcement learning, and combinations of the two strategies. Model comparison revealed evidence for hypothesis testing that relies on reinforcement-learning when selecting what hypothesis to test. The extent to which participants engaged in hypothesis testing depended on the instructed task complexity: people tended to serially test hypotheses when instructed that there were fewer relevant dimensions, and relied more on gradual and parallel learning of feature values when the task was more complex. This demonstrates a strategic use of task information to balance the costs and benefits of the two methods of learning.

show abstract

Active Function Learning

Cited by 8 publications

References 23 publications

Sampling heuristics for active function learning

Sampling heuristics for active function learning

Transformations and Transfer: Preschool Children Understand Abstract Relations and Reason Analogically in a Causal Task

Humans combine value learning and hypothesis testing strategically in multi-dimensional probabilistic reward learning

Contact Info

Product

Resources

About