United States

Our decisions are accompanied by a sense of confidence, a metacognitive assessment of how likely those decisions are to be correct, but the mechanisms that underlie this capacity remain poorly understood. A number of recent behavioral and neural data have suggested that decisions are made in accord with an optimal `balance-of-evidence&#39; rule, whereas confidence is estimated using a heuristic `response-congruent-evidence&#39; rule. We developed a deep neural network model optimized to classify images and predict its own likelihood of being correct, and found that this model naturally accounts for some of the key behavioral dissociations between decisions and confidence ratings. Further investigation revealed that neither the `balance-of-evidence&#39; rule nor the `response-congruent-evidence&#39; rule fully characterized the strategy that the model learned. We argue instead that the model learns to flexibly approximate the distribution of its training data, and, analogously, that apparently suboptimal features of human confidence ratings may arise from optimization for the statistics of naturalistic settings.

CogSci 2021

A Task-Optimized Neural Network Model of Decision Confidence

### Welcome!
The 62nd Annual Meeting of the Association for Computational Linguistics (ACL 2024) will take place in Bangkok, Thailand from August 11th to 16th, 2024. <br>Our Virtual Poster Sessions will take place online Thursday, August 22, 2024.<br><br><br>

You are required to register for this event. <br>**Please register [here](https://2024.aclweb.org/registration). **

If you have already registered, please check your inbox for an email from Underline granting you access to ACL 2024 content.


Please register!

ACL 2024

The 62nd Annual Meeting of the Association for Computational Linguistics (ACL 2024) will take place in Bangkok, Thailand from August 11th to 16th, 2024. More information will be announced soon.

Our decisions are accompanied by a sense of confidence, a metacognitive assessment of how likely those decisions are to be correct, but the mechanisms that underlie this capacity remain poorly understood. A number of recent behavioral and neural data have suggested that decisions are made in accord with an optimal `balance-of-evidence' rule, whereas confidence is estimated using a heuristic `response-congruent-evidence' rule. We developed a deep neural network model optimized to classify images and predict its own likelihood of being correct, and found that this model naturally accounts for some of the key behavioral dissociations between decisions and confidence ratings. Further investigation revealed that neither the `balance-of-evidence' rule nor the `response-congruent-evidence' rule fully characterized the strategy that the model learned. We argue instead that the model learns to flexibly approximate the distribution of its training data, and, analogously, that apparently suboptimal features of human confidence ratings may arise from optimization for the statistics of naturalistic settings.

poster

### 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.

The neural mechanisms supporting flexible relational inferences, especially in novel situations, are a major focus of current research. In the complementary learning systems framework, pattern separation in the hippocampus allows rapid learning in novel environments, while slower learning in neocortex accumulates small weight changes to extract systematic structure from well-learned environments. In this work, we adapt this framework to a task from a recent fMRI experiment where novel transitive inferences must be made according to implicit relational structure. We show that computational models capturing the basic cognitive properties of these two systems can explain relational transitive inferences in both familiar and novel environments, and reproduce key phenomena observed in the fMRI experiment.

Complementary Structure-Learning Neural Networks for Relational Reasoning

Neurophysiological studies in non-human primates suggest that perceptual decision-making consists of two stages of information processing: sensory evidence accumulation and response selection. Recent work with humans shows that the sensory evidence accumulation process can be tracked with the CPP component derived from EEG. As most studies in the field use simple motion stimuli, it remains unclear whether these processes generalize to more complex and socially important stimuli such as biological motion. In the present study, we used point-light displays with 4 levels of coherence and recorded EEG as human subjects (N=14) performed a perceptual decision-making task. Our results show that biological motion elicited a CPP component whose peak rate tracks the coherence level of the stimuli, albeit with a later onset than observed previously. These results suggest that similar decision-making mechanisms may play a role in biological motion perception.

Biological motion perception in perceptual decision-making framework: ERP evidence in humans

Localizing function in the brain has been an elusive long-term goal in the study of cognition. Prior studies have utilized four reference abilities (RAs) to capture cognition (Salthouse, 2009). Full-brain cortical networks have been tied to these abilities using common multi-voxel patterns across subjects in distinct age groups. Using voxel searchlights the current study explores purely local cortical representations of cognition, less commonly explored. This work analyzes 240 subjects’ responses to cognitive tasks from the four RAs. The current study further employs representational similarity analysis (RSA, Kriegeskorte, Mur, & Bandettini, 2008) to the similarity of brain activities from tasks within the same RA; RSA can capture representational consistencies within each subject even when exact voxel pattern may vary across subjects. We found distinct topographical localizations for each RA that were mostly consistent across age and suggested refinements of broader functional divisions of the brain from prior literature.

Identifying local cognitive representations in the brain across age spans through voxel searchlights and representational similarity analysis

Embodied cognition theory predicts that semantic processing shares processing resources with sensorimotor systems. The present study aimed to reveal mechanisms for impact of motor simulation on semantic processing of objects, which are manipulated by hands. We measured activation of inferior parietal lobule (IPL) by functional near-infrared spectroscopy (fNIRS) to examine the effect of constraint on hand movement. Participants were faced with two words representing name of objects that can be manipulated by hand (e.g. cup) or objects that cannot be manipulated by hand (e.g. windmill), and answered which object was larger. We analyzed effect of two factors on IPL activity: hand constraint and hand manipulability of the objects represented by the words. We found that (1) IPL activity for hand-manipulable objects was significantly higher than non-manipulable objects under control condition, and that (2) the difference in IPL activity between hand manipulable and non-manipulable objects was significantly reduced under hand-constraint condition.

Hand constraint affects semantic processing of hand-manipulable objects: An fNIRS study

The surprising efficiency of triple conjunction search has created a puzzle for modelers who link visual feature binding to selective attention, igniting an ongoing debate on whether features are bound with or without attention. Nordfang and Wolfe (2014) identified feature sharing and grouping as important factors in solving the puzzle and thereby established new constraints for models of visual search. Here we extend our neural dynamic model of scene perception and visual search (Grieben et al., 2020) to account for these constraints without the need for preattentive binding. By demonstrating that visual search is not only guided top-down, but that its efficiency is affected by bottom-up salience, we address a major theoretical weakness of models of conjunctive visual search (Proulx, 2007). We show how these complex interactions emerge naturally from the underlying neural dynamics.

A neural dynamic process model of combined bottom-up and top-down guidance in triple conjunction visual search

Speech and music are fundamental human communication modes. To what extent they rely on specific brain networks or exploit general auditory mechanisms based on their spectrotemporal acoustic structure is debated. We aimed at defining connectivity patterns modulated by attention to auditory content and spectrotemporal information, using fMRI. Participants tried to recognise sung speech stimuli that were gradually deprived of spectral or temporal information. Although auditory cortices appeared to specialise in temporal (left) and spectral (right) encoding, modularity of the bilateral connectivity network was largely unaffected by spectrotemporal degradations or attention. However, while participants' recognition decreased when necessary acoustic information - spectral for attention to melodies, temporal for attention to sentences - was degraded, efficiency of information flow in the network increased, and different subnetworks emerged. This suggests that the loss of crucial spectral (melody) or temporal (speech) information is compensated within the network by recruiting more and differential neural ressources.

Spectrotemporal cues and attention modulate neural networks for speech and music

A lot of studies on the N400 event-related brain potential (ERP) component have tried to understand its functional significance. Recently, the N400 was modeled in a neural network model as the update of a probabilistic representation of sentence meaning. The change in activation induced by a word was proposed to reflect a semantic prediction error that drives adaptation of the model’s connections (Rabovsky et al., 2018), which implies that more negative N400 amplitudes should lead to greater adaptation. By experimentally manipulating expectancy in a sentence reading task (n=33), we could show that this manipulation did not only influence N400 amplitudes, but also implicit memory: reaction times in a perceptual identification task were significantly faster for previously unexpected words. Additionally, difference in N400 amplitude correlated with the reaction time benefit for unexpected compared to expected items. These finding support the interpretation of the N400 as a prediction error and learning signal.

The N400 event-related potential component reflects a learning signal

This study investigated the effects of semantic transparency and familiarity on the lexical processing of German compounds. We measured event-related potentials (ERPs) while participants saw compound triplets that held the same head like /garten/ (‘garden’): (a) semantically transparent compounds /Gemüsegarten/ (‘vegetable garden’), (b) semantically opaque compounds /Kindergarten/ (‘kinder garden’), and (c) possible but nonexistent novel compounds /Gesichtsgarten/ (‘face garden’). Participants made nonword decisions to compounds with a scrambled constituent /Semüregarten/. 

ERP potentials for semantically transparent and opaque compounds were alike, irrespective of whether the manipulation was on the modifier or the head. By contrast, novel compounds showed strong N400 effects relative to both transparent and opaque compounds. These findings indicate that – during compound processing – the brain differentiates between familiar and novel, but not between transparent and opaque. Compound processing independent of semantic compositionality differs from that in other Indo-European languages and stresses the importance of cross-language comparisons.

Kindergarten versus ‘Gartenkinder’: EEG-evidence on the effects of familiarity and semantic transparency on German compounds

Children born preterm are at increased risk for white matter injury, impaired cognitive development and lower academic achievement. Here, we examined the association between fractional anisotropy and volume of select white matter tracts at age 5 with mathematics skills at age 7 in children born preterm (<33 weeks gestational age, n=52) without severe neurological complications and children born full-term (38-41 weeks gestational age, n=34). The preterm group had significantly lower mathematics scores and lower volume in several white matter tracts. Using multiple linear regression models, we examined white matter tracts that have previously been associated with mathematical cognition. We found a significant interaction with term status: fractional anisotropy of the corticospinal tract, and volume of corticospinal tract and parietal superior longitudinal fasciculus were significantly associated with mathematics skills in children born full-term, but not in children born preterm. These findings indicate white matter plasticity following preterm birth.

White Matter Tract Properties and Mathematics Skills: A Longitudinal Study of Children Born Preterm and Full-term

This study investigated whether positive or negative media exposure to non-local accents influences children’s language attitudes of those accents. Following exposure to heroic or villainous cartoon characters with either a regional (British; Experiment 1, N=89) or non-native (Experiment 2, N=84) accents, children were tested on a friend preference task, during which they selected—given a choice between a locally-accented child and a British-accented (Experiment 1) or Korean-accented (Experiment 2) child—which one they wanted to be friends with. Consistent with previous literature, children selected native-accented speakers at above chance rates, both when paired with British- (M=55%, β=0.25, p=0.03) and Korean-accented children (M=86%, β=2.63, p<.001). There was no evidence that exposure to heroic vs. villainous characters influenced children’s preferences for native-accented children in either experiment. Follow-up work is investigating whether protracted exposure to evil/heroic characters with non-local accents influences children’s evaluations to non-local varieties across a wider range of tasks.

Downloads

Next from CogSci 2021

Complementary Structure-Learning Neural Networks for Relational Reasoning

Similar lecture

Confidence Under the Hood: An Investigation into the Confidence-Probability Alignment in Large Language Models

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES