United States

Computational models are greatly useful in cognitive science in revealing the mechanisms of learning and decision making. However, it is hard to know whether all meaningful variance in behavior has been account for by the best-fit model selected through model comparison. In this work, we propose to use recurrent neural networks (RNNs) to assess the limits of predictability afforded by a model of behavior, and reveal what (if anything) is missing in the cognitive models. We apply this approach in a complex reward-learning task with a large choice space and rich individual variability. The RNN models outperform the best known cognitive model through the entire learning phase. By analyzing and comparing model predictions, we show that the RNN models are more accurate at capturing the temporal dependency between subsequent choices, and better at identifying the subspace in the space of choices where participants&#39; behavior is more likely to reside. The RNNs can also capture individual differences across participants by utilizing an embedding. The usefulness of this approach suggests promising applications of using RNNs to predict human behavior in complex cognitive tasks, in order to reveal cognitive mechanisms and their variability.

CogSci 2021

Using Recurrent Neural Networks to Understand Human Reward Learning

### Looking for Cogsci 2023? Please [click here](https://underline.io/events/405/reception) to switch to 2023 edition.


**Dear Colleagues,**

Welcome to the 43rd Annual Conference of the Cognitive Science Society! 

This year’s conference brings together the latest research from all Cognitive Science and also highlights this year’s theme **Comparative Cognition: Animal Minds.** 

The program features four plenary speakers: Friederike Range (University of Veterinary Medicine, Vienna), Onur Güntürkün (Ruhr University Bochum), Julia Fischer (University of Göttingen), and Erich Jarvis (The Rockefeller University). The keynotes and invited symposia showcase the broad spectrum of ideas that encompass the conference theme of comparative cognition. 

Further, the program includes the Rumelhart Prize keynote presentation by Susan Goldin-Meadow (University of Chicago), and symposia to celebrate the winner of the Elman Prize: Wei Ji Ma (New York University), the Heineken Award winner Robert Zatorre (McGill University), and the Robert J. Glushko Dissertation Prizes session.

The program committee for CogSci 2021 received 1,025 submissions, including 710 full papers, 275 member abstracts, 8 publication-based talks, as well as 15 proposals for symposia, and 17 for workshops and tutorials. After a rigorous review process, the committee selected 152 papers for oral presentation and inclusion in the conference proceedings (14.8%), 332 papers for poster presentation and inclusion in the proceedings (32.4%), and 175 papers for poster presentation with inclusion of abstracts in the proceedings (17.1%) We also selected 275 submitted member abstracts. In addition, we accepted 6 publication-based talks, 12 symposia, and 9 workshops and tutorials to make for an exciting and inclusive program.

We hope that you enjoy the conference this year!<br><br>

**Your Program Chairs,**
![](https://assets.underline.io/uploads/markdown_image/1/image/99984908caba5e6453e5fc5ba34f3507.jpg)

**Learn more about the Vienna Cognitive Science Hub at the University of Vienna, hosts of CogSci 2021**

<iframe width="560" height="315" src="https://www.youtube.com/embed/464h1m3iU68" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe>

To review the proceedings for the conference or review the searchable program, please visit the [website](https://cognitivesciencesociety.org/program/).


The Cognitive Science Society brings together researchers from around the world who hold a common goal: understanding the nature of the human mind. The mission of the Society is to promote Cognitive Science as a discipline, and to foster scientific interchange among researchers in various areas of study, including Artificial Intelligence, Linguistics, Anthropology, Psychology, Neuroscience, Philosophy, and Education. The Society is a non-profit professional organization and its activities include sponsoring an annual conference and publishing the journals Cognitive Science and TopiCS.

Computational models are greatly useful in cognitive science in revealing the mechanisms of learning and decision making. However, it is hard to know whether all meaningful variance in behavior has been account for by the best-fit model selected through model comparison. In this work, we propose to use recurrent neural networks (RNNs) to assess the limits of predictability afforded by a model of behavior, and reveal what (if anything) is missing in the cognitive models. We apply this approach in a complex reward-learning task with a large choice space and rich individual variability. The RNN models outperform the best known cognitive model through the entire learning phase. By analyzing and comparing model predictions, we show that the RNN models are more accurate at capturing the temporal dependency between subsequent choices, and better at identifying the subspace in the space of choices where participants' behavior is more likely to reside. The RNNs can also capture individual differences across participants by utilizing an embedding. The usefulness of this approach suggests promising applications of using RNNs to predict human behavior in complex cognitive tasks, in order to reveal cognitive mechanisms and their variability.

poster

Recent theories suggest that emotion words may facilitate the development of emotion concepts. However, most research investigating this relation in early childhood has been correlational. To assess whether emotion words causally influence emotion concept development, we conducted a pre-test post-test study examining which facial configurations 3-year-olds associate with complex emotional scenarios (annoyed, disgusted, and nervous). Between pre- and post-test, children were randomly assigned to one of three conditions. Children observed a facial configuration paired with a scenario while presented with either an explicit emotion label, a vague emotion label, or irrelevant information. Data from 54 children (36 female, mean age= 3.53) revealed that children’s average change in number of correct responses from pre-test to post-test by condition were as follows: Explicit=1.00 (SD=1.68); Vague=0.11 (SD=1.45); Irrelevant=-0.28, (SD=1.71). These results hold implications for how specific emotion words may casually influence children’s ability to learn new emotion concepts.

Emotion Words May Connect Complex Emotional Events and Facial Expressions in Early Childhood

Improving the magnitude processing during early childhood is essential for further developing numerical cognition and mathematical skills. In this regard, there is an intense debate about whether numbers are processed using a number-specific system or a general magnitude processing system. Additionally, the evidence available focuses on the magnitude’s interference but not on the translation and integration process. This study aims to analyze the ability to map between non-symbolic and symbolic numerical information and integrate numbers and space. For this purpose, we designed an observational between-subjects study using two-cross format comparison tasks involving the integration and interconversion between magnitudes and formats.
Improving the magnitude processing during early childhood is essential for further developing numerical cognition and mathematical skills. In this regard, there is an intense debate about whether numbers are processed using a number-specific system or a general magnitude processing system. Additionally, the evidence available focuses on the magnitude’s interference but not on the translation and integration process. This study aims to analyze the ability to map between non-symbolic and symbolic numerical information and integrate numbers and space. For this purpose, we designed an
observational between-subjects study using two-cross format comparison tasks involving the integration and interconversion between magnitudes and formats. We will assess for each task the relation between ratio and performance, and the discrimination thresholds in 5-, 10- and 15-years old Colombian children to explore the developmental trajectory of these numerical cognition processes.

Mapping between numerical and non-numerical magnitude information: An observational study of the integration and interconversion between magnitudes and formats in Colombian children

Previous research suggests that humans use multimodal information during emotional communication. Only a few studies have investigated how emotional information is presented in non-verbal channels. We examined the role of hand gestures in communicating emotions. In a between-subject design, we analyzed the size, frequency, and type of gestures under encouraged and spontaneous gesturing conditions. Participants (N=36) were asked to describe narratives with emotional content. Preliminary analyses with 18 individuals show that, interestingly, gesture frequencies and gesture use with specific emotional phrases did not differ between groups. Also, there was not a significant difference between the two groups in the types of gestures they produced. However, the gestures in the encouraged condition were significantly bigger in size compared to gestures in the spontaneous condition. These diverse preliminary results suggest that people’s hand gestures during emotional communication should be investigated with all their dimensions instead of focusing on one aspect.

The Role of Hand Gestures in Emotion Communication

How do people ask questions to zero in on a correct answer? Although we can formally define an optimal query to maximize information gain, algorithms for finding this optimal guess may impose large resource costs in space (memory) and time (computation). 
To understand how people trade off the information gain and the computational difficulty of choosing the ideal query, we turned to a large dataset of 380,000 guesses made during a number-guessing game with Amazon Alexa. We analyzed whether the arithmetic difficulty of following the optimal strategy predicts how far a guess deviates from theoretically optimal query. We find that when memory load is higher, and when more arithmetic operations need to be performed, human guesses deviate more from the most informative query.
These results suggest human computational resource constraints limit how people seek out informative questions.

Cognitive cost and information gain trade off in a large-scale number guessing game

Children vary widely in their lexical development. These differences have been shown to carry through later ages and can even be found in adults. One parameter often found as a predictor of varying vocabulary size is speed of processing, measuring the reaction time in familiar word recognition tasks. However, the underlying nature of speed of processing is still unclear: Is it a purely linguistic phenomenon or is it tied to the general cognitive abilities of the child. Our study aims to shed light on the nature of speed of processing, by testing 17-month-olds in a word learning experiment and assessing their reaction time during word recognition for both newly acquired and familiar words as well as visual reaction times. Our results can thus disentangle how broad or narrow lexical speed of processing is and help us better understand the origin of its link with language skills across the lifespan.

Investigating the nature of infants' lexical speed of processing

According to theories of event-predictive cognition, neural processing focuses on the next relevant interaction targets. Evidence for this notion comes from the anticipatory crossmodal congruency effect (aCCE), which implies that spatial body representations are mapped onto future goal locations in advance of a goal-directed action. Here we present a free energy based normative process model that accounts for the aCCE quantitatively by applying crossmodal mappings between vision and touch as well as active inference. A comparison with a diffusion model shows that our model accounts for the response time distributions and the aCCE with a sparser set of parameters. However, the temporal dynamics of the model require further fine tuning to account for all aspects of the aCCE. The model shows how the free energy framework can be used to account for behavioral data in general and how to implement theories of event-predictive cognition in a normative cognitive process model.

Modeling the Anticipatory Remapping of Spatial Body Representations: A Free Energy Approach

Spacing is a highly effective encoding strategy that has been shown to benefit memory in a variety of domains. Recent work has emphasized the evaluation of spaced practice under conditions that more closely reflect daily life. This work found that spacing repetitions over days, weeks, or months is effective over retention intervals as long as one year. One aspect of spaced study that has received less attention, however, is the relationship between the number of acquisition sessions and final retention. That is, if a student preparing for an exam plans to allocate 10 hours to preparing for that exam, is there an optimal, or perhaps minimal, number of study sessions that they should engage in to best leverage the benefits of spacing? In the present experiment we had participants complete 16 practice tests of Japanese-English pairs (e.g., boushi – hat). These practice tests were either all completed in one session, or distributed across two, three, or four sessions. These sessions were spaced either 1, 7, 30, or 90 days apart. Participants completed four test trials following a retention interval of 90 days, 180 days, or some variable length. Our results suggest that the number of acquisition sessions monotonically enhanced first-trial test performance as well as relearning, though evidence for enhanced relearning between one and two sessions was ambiguous. Unexpectedly, these monotonic trends were stable across practice lags and retention intervals. These findings suggest that, in addition to the temporal lag between practice episodes, the number of sessions over which one elects to distribute those episodes also has ramifications for long-term retention, and that each additional session yields meaningful benefits.

Additional acquisition sessions monotonically benefit retention and relearning

The physical world is rich with social information that people readily detect and extract, such as inferring that someone was present when we encounter a stack of rocks in the woods. How do people recognize that a physical scene contains social information? Research in developmental psychology has argued that this capacity is supported by a sensitivity to violations of randomness. Here we present a computational model of this idea and test its explanatory power in a quantitative manner. Our model infers agency by estimating the likelihood that a scene would arise naturally, as determined by human intuitive physics instantiated as a physics engine. Our results suggest that people's ability to detect agency in a physical scene is sensitive not only to the superficial visual properties, but also to the underlying physical generative process. Our results highlight how people use intuitive physics to decide when to engage in nuanced social reasoning.

Detecting the involvement of agents through physical reasoning

Core knowledge concepts such as object behavior principles provide a rich inventory of primitives for thinking and learning in the natural world. However, it remains unexplored how these concepts are reused for problem-solving and communication in highly abstract domains.

We analyze a large-scale natural language study drawing on the Abstraction and Reasoning Corpus (ARC), a set of highly abstract visual tasks where solvers construct outputs from input grids according to an inferred pattern. ARC explicitly incorporates core knowledge principles without any real-world objects. In the study, subjects solved and communicated the inferred patterns of ARC tasks via written explanations for other subjects attempting to solve tasks using only the explanations.

We examine how subjects solve, communicate, and interpret these explanations, and we show that subjects use fundamentally abstract core knowledge properties–object cohesion and contact causality–to reason about, understand, and communicate the inference tasks with language.

Core knowledge objects in reasoning and language use for highly abstract inductive tasks

Young infants reason about the goals, preferences, and actions of others. State of the art computational models, however, still fail in such reasoning. The Baby Intuitions Benchmark (BIB) was designed to test agency reasoning in AI using an infant behavioral paradigm. While BIB’s presentation of simple animations makes it particularly suitable for testing AI, such vignettes have yet to be validated with infants.

In this pilot, 11-month-old infants watched two sets of animations from BIB, one on agents’ consistent preferences and the other on agents’ efficient actions. Infants looked longer towards violations in agents’ behavior in both the preference (N = 24, β = 3.24 p = .040) and efficiency task (N = 24, β = 4.50 p = .016).

These preliminary results suggest that infants’ agency reasoning is abstract enough to be elicited by simple animations and validate BIB as a test of agency reasoning for humans and AIs.

Using Recurrent Neural Networks to Understand Human Reward Learning

Downloads

Next from CogSci 2021

Emotion Words May Connect Complex Emotional Events and Facial Expressions in Early Childhood

Similar lecture

Speechless Reader Model: A neurocognitive model for human reading reveals cognitive underpinnings of baboon lexical decision behavior.