Accueil > Pages Perso > Fabien Mathy

Fabien Mathy

PR -  UCA

Dernières publicationsHAL

pour l'idHal "fabien-mathy" :

titre
Hold-out strategy for selecting learning models: application to categorization subjected to presentation orders
auteur
Giulia Mezzadri, Thomas Laloë, Fabien Mathy, Patricia Reynaud-Bouret
article
Journal of Mathematical Psychology, Elsevier, 2022, 109 (102691), ⟨10.1016/j.jmp.2022.102691⟩
annee_publi
2022
resume
In this article, we develop a new general inference method for selecting learning models. The method relies upon a specific hold-out cross-validation, which takes into account the dependency within the data. This allows us to retrieve the model that best fits the learning strategy of a single individual. The novelty of our approach lies on the choice of the testing set, both in the experimental design and in the data analysis. This individual approach is then applied to two category learning models (ALCOVE and Component-cue) on data-sets manipulating presentation order, after verification of the reliability of our method. We found that both models performed equally well during transfer, but Component-cue best fits the majority of participants during learning. To further analyze these models, we also investigated a potential relation between the underlying mechanisms of the models and the actual types of presentation order assigned to participants.
typdoc
Article dans une revue
Accès au texte intégral et bibtex
https://hal.archives-ouvertes.fr/hal-03284595/file/Hold-out%20strategy%20for%20selecting%20learning%20models%3A%20application%20to%20categorization%20subjected%20to%20presentation%20orders%20-%202022.pdf BibTex
titre
Differential use of Transitional Probabilities and Frequency in Statistical Learning of Pseudowords
auteur
Laura Lazartigues, Fabien Mathy, Frédéric Lavigne
article
International Conference on Interdisciplinary Advances in Statistical Learning, Jun 2022, San Sebastian, Spain
annee_publi
2022
resume
The ability to learn transitional probabilities (TPs) and frequency is central to language processing. Current evidence indicates that both frequency and transitional probability are involved in the memorization of sequences, but the questions of which prevails and why it would prevail in statistical learning remain unclear. The present study investigated the respective roles of transitional probability and frequency in statistical learning of pseudowords in two different tasks that focused on either prediction or recognition. The learning phase consisted of the repeated presentation of sixteen three-syllable pseudowords for which participants were asked to perform a target detection task on vowels (fully predictable based on TPs). The evolution of the rate of correct answer and response times during the learning phase was recorded. After the detection task, a two-alternative forced-choice task (2AFC) required participants to choose between a pseudoword and a lure. Results indicated a prevalence of TPs during the detection task, but a prevalence of frequency during the 2AFC task. Our findings suggests that TPs and frequency can be used flexibly depending on which process (learning or recognition) is more adapted to the task.
typdoc
Poster de conférence
Accès au texte intégral et bibtex
https://hal.archives-ouvertes.fr/hal-03688991/file/Lazartigues2022.pdf BibTex
titre
Investigating interactions between types of order in categorization
auteur
Giulia Mezzadri, Patricia Reynaud-Bouret, Thomas Laloë, Fabien Mathy
article
2022
annee_publi
2022
resume
This study investigates how different types of presentation orders influence category learning and generalization. We used a full factorial design (within-category, between-category, and across-blocks manipulations), each factor having two levels (rule-based vs. similarity-based, blocked vs. interleaved, and constant vs. variable orders, respectively). This research offers a unique and novel approach through both an individual and concurrent analysis of the studied factors. Moreover, the investigation of across-blocks manipulations is unprecedented. Using survival analysis techniques as well as non-parametric tests, we found that learning was impacted by both within-category and across-blocks manipulations (with across-blocks manipulation being the main predictor). More specifically, the rule-based and constant orders were found more beneficial than the similarity-based and variable orders, respectively. Also, within-category and between-category orders were found to affect generalization patterns. More precisely, participants in the rule-based order more often showed generalization patterns consistent with a rule-based strategy than participants in the similarity-based order. Likewise, participants in the interleaved order more often showed generalization patterns consistent with a rule-based strategy than participants in the blocked order. We conclude that combining a similarity-based order with a variable across-blocks manipulation delays learning, whereas combining a rule-based order with a constant across-blocks manipulation benefits learning. Furthermore, combining a rule-based order with an interleaved order favors the use of a rule-based strategy during category generalization.
typdoc
Pré-publication, Document de travail
Accès au texte intégral et bibtex
https://hal.archives-ouvertes.fr/hal-03349792/file/Investigating%20interactions%20between%20types%20of%20order%20in%20categorization%20-%202022.pdf BibTex
titre
Learning Higher‐Order Transitional Probabilities in Nonhuman Primates
auteur
Arnaud Rey, Joël Fagot, Fabien Mathy, Laura Lazartigues, Laure Tosatto, Guillem Bonafos, Jean‐marc Freyermuth, Frédéric Lavigne
article
Cognitive Science, Wiley, 2022, 46 (4), ⟨10.1111/cogs.13121⟩
annee_publi
2022
resume
The extraction of cooccurrences between two events, A and B, is a central learning mechanism shared by all species capable of associative learning. Formally, the cooccurrence of events A and B appearing in a sequence is measured by the transitional probability (TP) between these events, and it corresponds to the probability of the second stimulus given the first (i.e., p(B|A)). In the present study, nonhuman primates (Guinea baboons, Papio papio) were exposed to a serial version of the XOR (i.e., exclusive-OR), in which they had to process sequences of three stimuli: A, B, and C. In this manipulation, first-order TPs (i.e., AB and BC) were uninformative due to their transitional probabilities being equal to .5 (i.e., p(B|A) = p(C|B) = .5), while secondorder TPs were fully predictive of the upcoming stimulus (i.e., p(C|AB) = 1). In Experiment 1, we found that baboons were able to learn second-order TPs, while no learning occurred on first-order TPs. In Experiment 2, this pattern of results was replicated, and a final test ruled out an alternative interpretation in terms of proximity to the reward. These results indicate that a non-human primate species can learn a nonlinearly separable problem such as the XOR. They also provide fine-grained empirical data to test models of statistical learning on the interaction between the learning of different orders of TPs. Recent bioinspired models of associative learning are also introduced as promising alternatives to the modeling of statistical learning mechanisms.
typdoc
Article dans une revue
Accès au texte intégral et bibtex
https://hal.archives-ouvertes.fr/hal-03631173/file/RFMLTBFL2022CS.pdf BibTex
titre
An order-dependent transfer model in categorization
auteur
Giulia Mezzadri, Patricia Reynaud-Bouret, Thomas Laloë, Fabien Mathy
article
Journal of Mathematical Psychology, Elsevier, 2022, 107 (102634), ⟨10.1016/j.jmp.2021.102634⟩
annee_publi
2022
resume
Most categorization models are insensitive to the order in which stimuli are presented. However, a vast array of studies have shown that the sequence received during learning can influence how categories are formed. In this paper, the objective was to better account for effects of serial order. We developed a model called Ordinal General Context Model (OGCM) based on the Generalized Context Model (GCM), which we modified to incorporate ordinal information. OGCM incorporates serial order as a feature along ordinary physical features, allowing it to account for the effect of sequential order as a form of distortion of the feature space. The comparison between the models showed that integrating serial order during learning in the OGCM provided the best account of classification of the stimuli in our data-sets.
typdoc
Article dans une revue
Accès au texte intégral et bibtex
https://hal.archives-ouvertes.fr/hal-03225670/file/An%20order-dependent%20transfer%20model%20in%20categorization%20-%202021.pdf BibTex
titre
Statistical Learning of Second-Order Transitional Probabilities in Humans
auteur
Laura Lazartigues, Fabien Mathy, Frédéric Lavigne
article
62nd Virtual annual meeting of the Psychonomic Society, Nov 2021, Virtuel, United States
annee_publi
2021
resume
The order of stimuli within sequences and the transitional probabilities (TPs) these orders generate are central information in language acquisition, but less is known about how this type of information is extracted by general learning mechanisms. The present study focused on the statistical learning of second-order TPs (i.e., only the combination of two stimuli allowing to predict the third) of visual sequences. Eight three-item sequences exclusively governed by second-order TPs were presented. The response times were measured with oculometry. The task included a learning phase and a switch phase which reset the second-order TPs (e.g., the sequences ABC and BAF became respectively ABF and BAC). Results indicated a sole decrease of RTs between the second and the third stimulus and an increase of RTs during the switch phase that suggested the learnability of second-order TPs. We discuss this result in light of language acquisition.
typdoc
Poster de conférence
Accès au texte intégral et bibtex
https://hal.archives-ouvertes.fr/hal-03509457/file/Statistical%20Learning%20of%20Second-Order%20Transitional%20Probabilities%20in%20Humans.pdf BibTex
titre
Switching costs in verbal working memory complex span task performance
auteur
Miriam Debraise, Fabien Mathy
article
62th Virtual annual meeting of the Psychonomic Society, Nov 2021, Nouvelle Orléans, LA (virtual), United States
annee_publi
2021
typdoc
Poster de conférence
Accès au texte intégral et bibtex
https://hal.archives-ouvertes.fr/hal-03523304/file/DEBRAISEMiriam_Psynom21.pdf BibTex
titre
Benefits and pitfalls of data compression in visual working memory
auteur
Laura Lazartigues, Frédéric Lavigne, Carlos Aguilar, Nelson Cowan, Fabien Mathy
article
Attention, Perception, and Psychophysics, Springer Verlag, 2021, ⟨10.3758/s13414-021-02333-x⟩
annee_publi
2021
resume
Data compression in memory is a cognitive process allowing participants to cope with complexity to reduce information load. However, previous studies have not yet considered the hypothesis that this process could also lead to over-simplifying information due to haphazard amplification of the compression process itself. For instance, we could expect that the over-regularized features of a visual scene could produce false recognition of patterns, not because of storage capacity limits but because of an errant compression process. To prompt memory compression in our participants, we used multielement visual displays for which the underlying information varied in compressibility. The compressibility of our material could vary depending on the number of common features between the multi-dimensional objects in the displays. We measured both accuracy and response times by probing memory representations with probes that we hypothesized could modify the participants’ representations. We confirm that more compressible information facilitates performance, but a more novel finding is that compression can produce both typical memory errors and lengthened response times. Our findings provide clearer evidence of the forms of compression that participants carry out.
typdoc
Article dans une revue
Accès au bibtex
BibTex
titre
Classification de patterns de catégorisation chez l'humain par deux modèles d'apprentissage
auteur
Giulia Mezzadri, Patricia Reynaud-Bouret, Thomas Laloë, Fabien Mathy
article
JDS 2020 : 52-ièmes journées de Statistique de la Société Française de Statistique (SFdS), Jun 2021, Nice (en ligne), France
annee_publi
2021
typdoc
Communication dans un congrès
Accès au texte intégral et bibtex
https://hal.archives-ouvertes.fr/hal-03508398/file/SFdS.pdf BibTex
titre
Probabilities, Dependencies and Frequency Are Not All Equally Involved in Artificial Word Learning
auteur
Laura Lazartigues, Fabien Mathy, Frédéric Lavigne
article
2021 Virtual APS Convention, May 2021, Virtuel, France
annee_publi
2021
resume
The ability to learn transitional probabilities (TPs), adjacent and non-adjacent dependencies and frequency is central to language processing. Current evidence indicates that both frequency and transitional probability impact the memorization of sequences, but it is not sure yet which type of information prevails. Also, adjacent and non-adjacent pairs do not seem to be equally learnable based on statistical learning. These three factors however are not yet integrated into a unique model predicting learning, whereas they are all supposed to play a critical role in statistical learning. The present study thus investigated the concurrent roles of transitional probability, frequency and adjacency within a single task aiming at having participants learn artificial words. Participants were exposed to four conditions, with each condition consisting in four words (sequences of three syllables without meaning) governed by a specific rule. The first condition tested first order TP between adjacent elements, where the second syllable only allowed to predict the third. The second condition tested first order TP between non-adjacent elements, where the first syllable only allowed to predict the third. The next condition tested second order TP, where only the combination of the two first syllables allowed to predict the third. The last condition tested the effect of the frequency for a given second order TP, by presenting two words five times per block, a third word two times per block and the fourth word eight times per block. All words were presented randomly in each block of the experiment. Participants had to perform a target detection task on the vowels that corresponded to the last syllable of each word, and which were fully predictable from the rules that governed the conditions. Correct answer rates and response times (RTs) were recorded. Each participant completed five sessions of ten blocks (eighty trials per block), one session per day for five days. A transfer block at the end of the last session was added to test whether participants could adapt to a switch inverting the vowels (but maintaining the rest of the rules). The results showed that words were learned in all conditions with an increase of correct answer rates, a decrease of RTs during the learning phase, and a drop of performance during the switch phase. Results further indicated easier learning of first order TPs with higher correct responses and shorter RTs compared second order TPs. We also observed an advantage for adjacent dependencies over non-adjacent ones. Moreover, a low frequency caused lower correct answer rates and slower RTs and seemed to interfere with the learning of new words . Analysis of frequency and first order TPs in the condition where the effect of the frequency for a given second order TP was tested indicated a prevalence of TPs over frequency at the end of the learning process.
typdoc
Poster de conférence
Accès au texte intégral et bibtex
https://hal.archives-ouvertes.fr/hal-03509434/file/Transitional%20Probabilities%2C%20Dependencies%20and%20Frequency%20are%20Not%20All%20Equally%20involved%20in%20Artificial%20Word%20Learning.pdf BibTex
  • + de résultats dans la Collection HAL du laboratoire BCL
  • Voir l'ensemble des résultats sur la plateforme HAL