United States

Modern machine learning models for computer vision exceed humans in accuracy on specific visual recognition tasks, notably on datasets like ImageNet. However, high accuracy can be achieved in many ways. The particular decision function found by a machine learning system is determined not only by the data to which the system is exposed, but also the inductive biases of the model, which are typically harder to characterize. In this work, we follow a recent trend of in-depth behavioral analyses of neural network models that go beyond accuracy as an evaluation metric by looking at patterns of errors. Our focus is on comparing a suite of standard Convolutional Neural Networks (CNNs) and a recently-proposed attention-based network, the Vision Transformer (ViT), which relaxes the translation-invariance constraint of CNNs and therefore represents a model with a weaker set of inductive biases. Attention-based networks have previously been shown to achieve higher accuracy than CNNs on vision tasks, and we demonstrate, using new metrics for examining error consistency with more granularity, that their errors are also more consistent with those of humans. These results have implications both for building more human-like vision models, as well as for understanding visual object recognition in humans.

CogSci 2021

Are Convolutional Neural Networks or Transformers more like human vision?

### 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.

poster

Social robotics has shown expansive growth in areas related to companionship/assistance for older adults. Critically, everyday interactions with artificial agents often involve spoken language in the context of a shared visual environment. Therefore, language interfaces for these applications must account for the distinctive nature of visually-situated communication revealed by psycholinguistic studies. In traditional frameworks, "rational" speakers were thought to avoid redundancy, yet human-human communication research shows that both younger and older speakers include redundant information (e.g., color adjectives) in descriptions to facilitate listeners' visual search. However, this "cooperative" use of redundant expressions hinges on beliefs about listeners' perception (e.g., "pop-out" nature of human color processing). We explored the incidence and nature of younger and older speakers' redundant descriptions for a robot partner in different visual environments. Whereas both age groups produced redundant descriptions, there were important age differences for when these descriptions occurred and for the properties encoded in them.

Younger and Older Speakers' Use of Linguistic Redundancy with a Social Robot

Colexification, or the expression of multiple concepts by the same word, is ubiquitous in language. Colexifications may appear rule-like, as when an artifact is used for an activity ('repair the shower'/'take a shower'), or similarity-based ('child' refers to both 'young person' and 'descendant'). We investigate whether these two modes of generalization (rules and similarity) reflect how people structure new meanings. We propose computational models based on rules, similarity, and a hybrid of the two, and correlate model predictions to human behavior—in a novel task, participants generalized labels across colexified meanings. We found that a model using similarity correlated much better with human behavior than rules. Further, the similarity model was significantly outperformed by a hybrid model of the two mechanisms. However, the difference in correlations was modest, suggesting that a framework which combines rules and similarity largely relies on similarity-based generalization to characterize human expectations about colexification.

Modeling rules and similarity in colexification

We present a computational cognitive model of referential choice that models and explains the choice between a wide variety of referring forms using a small set of features important to situated contexts. By combining explainable machine learning techniques, data collected in situated contexts, and recent computational models of cognitive status, we produce an accurate and explainable model of referential choice that provides an intuitive pragmatic account of this process in humans, and an intuitive method for computationally enabling this capability in robots and other autonomous agents.

Givenness Hierarchy Theoretic Referential Choice in Situated Contexts

We present a computational model of interactions between a Speaker and a Hearer in a signalling game. The partly cooperative/partly competitive interaction is intended to reproduce essential aspects of human communication, with the of approximating human language use and analysing it by means of simulation studies.
This was accomplished by implementing a language that accommodates for compositional signals, allowing agents to express infinitely many meanings with a finite set of signals. Personal attributes integral to human decision making such as sympathy and trust were implemented as adjustable parameters, providing the opportunity to create and study individuals with different 'personalities'. Over the course of several rounds, agents learn their optimal strategies using the Moran algorithm. The model was able to substantiate widely confirmed notions about human communication such as correlations between truthfulness and trust, demonstrating the possibility of correlation between aspects of linguistic cognition and social aspects of language use.

Modelling Human Communication as a Rejection Game

The integration of behavioral phenomena into mechanistic models of cognitive function is a fundamental staple of cognitive science. Yet, researchers are beginning to accumulate increasing amounts of data without having the temporal or monetary resources to integrate these data into scientific theories. We seek to overcome these limitations by incorporating existing machine learning techniques into an open-source pipeline for the automated construction of quantitative models. This pipeline leverages the use of neural architecture search to automate the discovery of interpretable model architectures, and automatic differentiation to automate the fitting of model parameters to data. We evaluate the utility of these methods based on their ability to recover quantitative models of human information processing from synthetic data. We find that these methods are capable of recovering basic quantitative motifs from models of psychophysics, learning and decision making. We also highlight weaknesses of this framework and discuss future directions for their mitigation.

Recovering Quantitative Models of Human Information Processing with Differentiable Architecture Search

In an effort towards predicting mental workload while driving, previous research found interactions between working memory load and visuospatial demands, which complicates the accurate prediction of momentary mental workload. To investigate this interaction, the cognitive concepts working memory load and visuospatial attention were integrated into a cognitive driving model using the cognitive architecture ACT-R. The model was developed to safely drive on a multi-lane highway with ongoing traffic while performing a secondary n-back task using speed signs. To manipulate visuospatial demands, the model must drive through a construction site with reduced lane-width in certain blocks of the experiment. Furthermore, it is able to handle complex driving situations such as overtaking traffic while adjusting the speed according to the n-back task. The behavioral results show a negative effect on driving performance with increasing task difficulty of the secondary task. Additionally, the model indicates an interaction at a common, task-unspecific level.

Utilizing ACT-R to investigate interactions between working memory and visuospatial attention while driving

The Abstraction and Reasoning Challenge (ARC) is a set of tasks where one must induce a program from a few given input-output examples, and apply it to a new input. Although humans can easily solve most tasks, ARC is challenging for state-of-the-art algorithms. We hypothesize that humans use intuitive program induction, and interpret ARC tasks by constructing "natural programs'' in language.

We experimentally study "natural programs'' by formulating a two-player game: A participant solves and then communicates the program to another participant using natural language; the second participant must solve the task using the description alone. We find at least 361 out of 400 tasks can be solved from a natural language description, demonstrating that natural language is sufficient in transmitting these natural programs. 

We compared the natural language programs to computer programs constructed using two separate state-of-the-art program synthesis approaches, and conducted a study on leveraging natural language annotations to improve performance of program-synthesis tools.

Language annotated Abstraction and Reasoning Corpus

Although Artificial Intelligence (AI) is expected to outperform humans in many domains of decision-making, the process by which AI arrives at its superior decisions is often hidden and too complex for humans to fully grasp. As a result, humans may find it difficult to learn from AI, and accordingly, our knowledge about whether and how humans learn from AI is also limited. In this paper, we aim to expand our understanding by examining human decision-making in the board game Go. Our analysis of 1.3 million move decisions made by professional Go players suggests that people learned to make decisions like AI after they observe reasoning processes of AI, rather than mere actions of AI. Follow-up analyses compared the decision quality of two groups of players: those who had access to AI programs and those who did not. In line with the initial results, decision quality significantly improved for the players with AI access after they gained access to reasoning processes of AI, but not for the players without AI access. Our results demonstrate that humans can learn from AI even in a complex domain where the computation process of AI is also complicated.

Human Learning from Artificial Intelligence: Evidence from Human Go Players’ Decisions after AlphaGo

Digitally-enabled means for judgment aggregation have renewed interest in "wisdom of the crowd'' effects and kick-started collective intelligence design as an emerging field in the cognitive and computational sciences. A keenly debated question here is whether social influence helps or hinders collective accuracy on estimation tasks, with recently introduced network theories offering a reconciliation of seemingly contradictory past results. Yet, despite a growing body of literature linking social network structure and the accuracy of collective beliefs, strategies for exploiting network structure to harness crowd wisdom are under-explored. In this paper, we introduce a potential new tool for collective intelligence design informed by such network theories: rewiring algorithms. We provide a proof of concept through agent-based modelling and simulation, showing that rewiring algorithms that dynamically manipulate the structure of communicating social networks can increase the accuracy of collective estimations in the absence of knowledge of the ground truth.

Rewiring the Wisdom of the Crowd

Theory-of-Mind inference is natural for humans but poses significant computational challenges. The core difficulty can be traced back to the exponential growth of paths to consider in planning given a mental state. In this paper we tackle this problem in a search-and-rescue task implemented in Minecraft. Our goal is to infer differences in knowledge from participants' continuous-time trajectory. By abstracting the spatio-temporal state space and the reward function together, we surface natural decision points, on which we compare the participants' behavior to myopic rational agents of varying knowledgeability. Collectively, the abstraction and rational agent analysis yield successful inference of participants' knowledge states and reveal distinct patterns of their exploratory behavior.

Are Convolutional Neural Networks or Transformers more like human vision?

Downloads

Next from CogSci 2021

Younger and Older Speakers' Use of Linguistic Redundancy with a Social Robot

Similar lecture

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