scholarly journals Humans monitor learning progress in curiosity-driven exploration

2020 ◽  
Author(s):  
Alexandr Ten ◽  
Pramod Kaushik ◽  
Pierre-Yves Oudeyer ◽  
Jacqueline Gottlieb
Keyword(s):  
Computational Modeling ◽  
Empirical Evidence ◽  
Time Scales ◽  
Human Cognition ◽  
Experimental Paradigm ◽  
Percent Correct ◽  
Intrinsic Goals ◽  
Computational Research ◽  
Learning Progress ◽  
Best Fit

Curiosity-driven learning is foundational to human cognition. Byenabling humans to autonomously decide when and what to learn,curiosity has been argued to be crucial for self-organizing temporally extended learning curricula. However, the mechanisms drivingpeople to set intrinsic goals, when they are free to explore multiplelearning activities, are still poorly understood. Computational theories propose different heuristics, including competence measures(e.g. percent correct, or PC) and learning progress (LP), that could beused as intrinsic utility functions to efficiently organize exploration.Such intrinsic utilities constitute computationally cheap but smartheuristics to prevent people from laboring in vain on random activities, while still motivating them to self-challenge on difficult learnable activities. Here, we provide empirical evidence for these ideasby means of a novel experimental paradigm and computational modeling. We show that while humans rely on competence information to avoid easy tasks, models that include an LP component provide the best fit to task selection data. These results provide a new bridge between research on artificial and biological curiosity, reveal strategies that are used by humans but have not been considered in computational research, and provide new tools for probing how humans become intrinsically motivated to learn and acquire interests and skills on extended time scales/

scholarly journals Humans monitor learning progress in curiosity-driven exploration

2020 ◽  
Author(s):  
Alexandr Ten ◽  
Pramod Kaushik ◽  
Pierre-Yves Oudeyer ◽  
Jacqueline Gottlieb
Keyword(s):  
Computational Modeling ◽  
Empirical Evidence ◽  
Learning Activities ◽  
Human Cognition ◽  
Experimental Paradigm ◽  
Percent Correct ◽  
Intrinsic Goals ◽  
Computational Research ◽  
Learning Progress ◽  
Best Fit

Curiosity-driven learning is foundational to human cognition. By enabling humans to autonomously decide when and what to learn, curiosity has been argued to be crucial for self-organizing temporally extended learning curricula. However, the mechanisms driving people to set intrinsic goals, when they are free to explore multiple learning activities, are still poorly understood. Computational theories propose different heuristics, including competence measures (e.g. percent correct, or PC) and learning progress (LP), that could be used as intrinsic utility functions to efficiently organize exploration. Such intrinsic utilities constitute computationally cheap but smart heuristics to prevent people from laboring in vain on random activities, while still motivating them to self-challenge on difficult learnable activities. Here, we provide empirical evidence for these ideas by means of a novel experimental paradigm and computational modeling. We show that while humans rely on competence information to avoid easy tasks, models that include an LP component provide the best fit to task selection data. These results provide a new bridge between research on artificial and biological curiosity, reveal strategies that are used by humans but have not been considered in computational research, and provide new tools for probing how humans become intrinsically motivated to learn and acquire interests and skills on extended time scales.

scholarly journals Humans monitor learning progress in curiosity-driven exploration

2021 ◽  
Vol 12 (1) ◽  
Author(s):  
Alexandr Ten ◽  
Pramod Kaushik ◽  
Pierre-Yves Oudeyer ◽  
Jacqueline Gottlieb
Keyword(s):  
Empirical Evidence ◽  
Free Choice ◽  
Learning Activities ◽  
Human Cognition ◽  
Experimental Paradigm ◽  
Percent Correct ◽  
Intrinsic Goals ◽  
Computational Research ◽  
Learning Progress ◽  
Best Fit

AbstractCuriosity-driven learning is foundational to human cognition. By enabling humans to autonomously decide when and what to learn, curiosity has been argued to be crucial for self-organizing temporally extended learning curricula. However, the mechanisms driving people to set intrinsic goals, when they are free to explore multiple learning activities, are still poorly understood. Computational theories propose different heuristics, including competence measures (e.g., percent correct) and learning progress, that could be used as intrinsic utility functions to efficiently organize exploration. Such intrinsic utilities constitute computationally cheap but smart heuristics to prevent people from laboring in vain on unlearnable activities, while still motivating them to self-challenge on difficult learnable activities. Here, we provide empirical evidence for these ideas by means of a free-choice experimental paradigm and computational modeling. We show that while humans rely on competence information to avoid easy tasks, models that include a learning-progress component provide the best fit to task selection data. These results bridge the research in artificial and biological curiosity, reveal strategies that are used by humans but have not been considered in computational research, and introduce tools for probing how humans become intrinsically motivated to learn and acquire interests and skills on extended time scales.

Neural Correspondence to Human Cognition from Analysis to Intuition – Implications of Display Design for Cognition

2017 ◽  
Vol 61 (1) ◽  
pp. 51-55 ◽  
Author(s):  
Joseph K. Nuamah ◽  
Younho Seong
Keyword(s):  
Objective Measure ◽  
Mental Effort ◽  
Human Cognition ◽  
Behavioral Measures ◽  
Display Design ◽  
Percent Correct ◽  
Task Load ◽  
Frontal Midline Theta ◽  
General Difference ◽  
Nasa Task Load Index

Psychophysiological measures can be used to determine whether a particular display produces a general difference in brain function. Such information might be valuable in efforts to improve usability in display design. In this preliminary study, we aimed to use the electroencephalography (EEG) task load index (TLI), given by the ratio of mean frontal midline theta energy to mean parietal alpha energy, to provide insight into the mental effort required by participants performing intuition-inducing and analysis-inducing tasks. We employed behavioral measures (reaction time and percent correct), and a subjective measure (NASA-Task Load Index) to validate the objective measure (TLI). The results we obtained were consistent with our hypothesis that mental effort required for analysis-inducing tasks would be different from that required for intuition-inducing tasks. Although our sample size was small, we were able to obtain a significant positive correlation between NASA-Task Load Index and TLI.

Weak evidence for a strong case against modularity in developmental disorders

2002 ◽  
Vol 25 (6) ◽  
pp. 764-765
Author(s):  
Ralph-Axel Müller
Keyword(s):  
Computational Modeling ◽  
Empirical Evidence ◽  
Cognitive Neuroscience ◽  
Developmental Disorders ◽  
Complete Understanding ◽  
Strong Case ◽  
Weak Evidence

Thomas & Karmiloff-Smith (T&K-S) provide evidence from computational modeling against modular assumptions of “Residual Normality” (RN) in developmental disorders. Even though I agree with their criticism, I find their choice of empirical evidence disappointing. Cognitive neuroscience cannot as yet provide a complete understanding of most developmental disorders, but what is known is more than enough to debunk the idea of RN.

scholarly journals Stability analysis of three exoplanet systems

2020 ◽  
Vol 494 (2) ◽  
pp. 2280-2288
Author(s):  
J P Marshall ◽  
J Horner ◽  
R A Wittenmyer ◽  
J T Clark ◽  
M W Mengel
Keyword(s):  
Time Scales ◽  
Planetary Systems ◽  
Orbital Parameters ◽  
Exoplanetary Systems ◽  
Best Fitting ◽  
The Stability ◽  
Short Time ◽  
Exoplanet Systems ◽  
Best Fit

ABSTRACT The orbital solutions of published multiplanet systems are not necessarily dynamically stable on time-scales comparable to the lifetime of the system as a whole. For this reason, dynamical tests of the architectures of proposed exoplanetary systems are a critical tool to probe the stability and feasibility of the candidate planetary systems, with the potential to point the way towards refined orbital parameters of those planets. Such studies can even help in the identification of additional companions in such systems. Here, we examine the dynamical stability of three planetary systems, orbiting HD 67087, HD 110014, and HD 133131A. We use the published radial velocity measurements of the target stars to determine the best-fitting orbital solutions for these planetary systems using the systemic console. We then employ the N-body integrator mercury to test the stability of a range of orbital solutions lying within 3σ of the nominal best fit for a duration of 100 Myr. From the results of the N-body integrations, we infer the best-fitting orbital parameters using the Bayesian package astroemperor. We find that both HD 110014 and HD 133131A have long-term stable architectures that lie within the 1σ uncertainties of the nominal best fit to their previously determined orbital solutions. However, the HD 67087 system exhibits a strong tendency towards instability on short time-scales. We compare these results to the predictions made from consideration of the angular momentum deficit criterion, and find that its predictions are consistent with our findings.

Multimodal constructs – multimodal constructions? The role of constructions in the working memory

2017 ◽  
Vol 3 (s1) ◽  
Author(s):  
Thomas Hoffmann
Keyword(s):  
Working Memory ◽  
Empirical Evidence ◽  
Human Cognition ◽  
Long Term Memory ◽  
Symbolic System ◽  
Term Memory ◽  
The Status ◽  
Precise Role

AbstractLanguage is a symbolic system, whose basic units are arbitrary and conventionalized pairings of form and meaning. In fact, in light of substantive empirical evidence, Construction Grammar approaches advocate the view that not only words but all levels of grammatical description – from morphemes, words, and idioms to abstract phrasal patterns as well as larger discourse patterns – comprise form-meaning pairings, which are collectively referred to as constructions. In this paper, I will discuss the status of multimodal usage-events (multimodal constructs) for the potential entrenchment of multimodal constructions and their implications for human cognition in general. As I will argue, constructionist approaches need to pay more attention to the role of the working memory in assembling and interpreting constructions. Drawing on verbal as well as gesture constructions, I will show that it is essential to distinguish entrenched constructions that are stored in the long-term memory from form-meaning pairings that are assembled in the working memory (online constructions). Once this distinction is made, the precise role of multimodal constructs and the nature of multimodal constructions can finally be disentangled.

scholarly journals How Far we Can Go Without Looking Under the Skin: The Bounds of Cognitive Science

2015 ◽  
Vol 40 (1) ◽  
pp. 91-109
Author(s):  
Łukasz Afeltowicz ◽  
Witold Wachowski
Keyword(s):  
Cognitive Science ◽  
Empirical Evidence ◽  
Distributed Cognition ◽  
Extended Mind ◽  
Cognitive System ◽  
Human Cognition ◽  
External Representations

Abstract The aim of this paper is to discuss the concept of distributed cognition (DCog) in the context of classic questions posed by mainstream cognitive science. We support our remarks by appealing to empirical evidence from the fields of cognitive science and ethnography. Particular attention is paid to the structure and functioning of a cognitive system, as well as its external representations. We analyze the problem of how far we can push the study of human cognition without taking into account what is underneath an individual’s skin. In light of our discussion, a distinction between DCog and the extended mind becomes important.

Investigating human prediction: sequence termination provides insights into predictive statistics

2021 ◽  
Author(s):  
Isaac Treves
Keyword(s):  
Computational Models ◽  
Frequency Estimation ◽  
Reaction Time Task ◽  
Prior Exposure ◽  
Human Cognition ◽  
Passive Exposure ◽  
Future Work ◽  
Best Fit ◽  
Prediction Sequence ◽  
Insight Into

Prediction is a fundamental process in human cognition. Prediction means extracting one or more statistics from the distribution of past inputs and using that information to make a decision. What are the statistics underlying human predictions, and how do they change with training? To investigate these questions, we designed a sequence termination task, where participants watch temporally unfolding sequences and terminate them when they can predict the next item. We then test how well the participants’ termination points are predicted by computational models. We contrast frequency estimation models (How often did this symbol appear in the sequence?), transition models (How often did symbol A follow symbol B?), and a chunking model (What are the patterns of symbols?). In an online experiment with 65 adults, we find that participants are best fit by a transition-counting model. To assess the effect of training, we manipulated passive exposure to the sequences prior to the sequence termination task. Contrary to our expectations, prior exposure to sequences had no effect on termination performance– whether tested statistically or computationally, and despite good power. Lastly, training specifically on the termination task may shift responses towards chunking. These results provide insight into the representations, or information in mind, behind prediction. However, the lack of an effect of prior exposure makes it clear that sequence termination measures explicit, or conscious, prediction. Future work could examine whether representations in explicit prediction tasks like sequence termination are different from implicit, or unconscious, tasks like the serial reaction time task.

scholarly journals Optical variability of quasars: a damped random walk

2013 ◽  
Vol 9 (S304) ◽  
pp. 395-398 ◽  
Author(s):  
Željko Ivezić ◽  
Chelsea MacLeod
Keyword(s):  
Random Walk ◽  
Time Scales ◽  
Model Parameters ◽  
Optical Wavelength ◽  
Long Time ◽  
Asymptotic Variability ◽  
Short Time ◽  
Quasar Variability ◽  
Best Fit

AbstractA damped random walk is a stochastic process, defined by an exponential covariance matrix that behaves as a random walk for short time scales and asymptotically achieves a finite variability amplitude at long time scales. Over the last few years, it has been demonstrated, mostly but not exclusively using SDSS data, that a damped random walk model provides a satisfactory statistical description of observed quasar variability in the optical wavelength range, for rest-frame timescales from 5 days to 2000 days. The best-fit characteristic timescale and asymptotic variability amplitude scale with the luminosity, black hole mass, and rest wavelength, and appear independent of redshift. In addition to providing insights into the physics of quasar variability, the best-fit model parameters can be used to efficiently separate quasars from stars in imaging surveys with adequate long-term multi-epoch data, such as expected from LSST.

scholarly journals PREDICTION AND ERROR-BASED LEARNING IN L2 PROCESSING AND ACQUISITION

2021 ◽  
pp. 1-26
Author(s):  
Giulia Bovolenta ◽  
Emma Marsden
Keyword(s):  
Language Acquisition ◽  
Computational Modeling ◽  
Language Learning ◽  
Empirical Evidence ◽  
Language Processing ◽  
Prediction Error ◽  
Predictive Processing ◽  
L2 Processing ◽  
L1 And L2

Abstract There is currently much interest in the role of prediction in language processing, both in L1 and L2. For language acquisition researchers, this has prompted debate on the role that predictive processing may play in both L1 and L2 language learning, if any. In this conceptual review, we explore the role of prediction and prediction error as a potential learning aid. We examine different proposed prediction mechanisms and the empirical evidence for them, alongside the factors constraining prediction for both L1 and L2 speakers. We then review the evidence on the role of prediction in learning languages. We report computational modeling that underpins a number of proposals on the role of prediction in L1 and L2 learning, then lay out the empirical evidence supporting the predictions made by modeling, from research into priming and adaptation. Finally, we point out the limitations of these mechanisms in both L1 and L2 speakers.

Sign in / Sign up

Export Citation Format

Share Document