United States

Most of the recent success in machine learning has been achieved in supervised learning and predicated on the availability of large amounts of labelled training data. On the other hand, effectively using readily available unlabelled data has proven a much more difficult endeavour. In contrast to algorithms, infants spontaneously learn from the available sensory information without explicit instructions, supervision or feedback. Thus, infant learning can be viewed as a highly successful approach to unsupervised learning. In this work, we explore the parallels between infant learning and recent successes of unsupervised machine learning in the area of contrastive learning. We examine how the principles of infant learning and developmental cognitive neuroscience can inform and inspire the development of novel contrastive learning algorithms. We focus on the phenomenon of category learning and explore how these principles can be applied to better understand and improve contrastive methods.

CogSci 2021

Can algorithms learn from babies? Exploring how infant learning can inform and inspire unsupervised learning algorithms

### 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

Children with Developmental Language Disorder (DLD) have deficits in verbal and nonverbal processing relative to typically developing (TD) peers. We examined working memory in DLD relative to age-matched TD peers (9-13 years) under the serial-order-in-a-box – complex span model. This model posits a time-based mechanism, Free Time, that governs how interference affects processing performance. Results showed that Free Time was positively associated with accuracy when recall and interference stimuli had verbal features (b = 0.00; stat = 3.11; p < .01), and combined verbal and nonverbal features (b = 0.00; stat = 3.05; p < .01). Group differences in this relationship were evident when recall stimuli had verbal features regardless of interference stimuli features (b = -0.00; stat = -3.66; p < .001; b = 0.00; stat = 2.97; p < .01). Findings suggest a greater role of Free Time for verbal than nonverbal content, which varies depending on participant characteristics.

Testing an interference-based model of working memory in children with developmental language disorder and their typically developing peers

A traditional view of understanding other’s mental states is that early in ontogeny infants start from a me-first position, and through themselves learn to understand others. Here we propose an opposite developmental trajectory, where infants might start out highly altercentric (Southgate, 2020) and through development increasingly rely on their own point of view. 

In a pre-registered cross-sectional study we present 1-6-year-old children with a task where altercentric modulation has been found with 14-month-old infants (Kampis & Kovács, 2020). There, infants tended to search longer in a box when another person believed an object to be present, than when she believed it was empty. We predicted this tendency to decrease with age. Preliminary results with n=191 children included based on their search in baseline trials, show a decrease in altercentric modulation with age (r=-.174, p= .016), which will be discussed together with the relationship between altercentrism and the development of self-concept.

Humans start out altercentric: the ontogenetic development of other-centered cognition

Reasoning by exclusion allows us to infer properties of unobserved objects from currently observed objects, formalized by P or Q, not P, therefore Q. Previous work suggested that, by age 3, children can use this kind of reasoning to infer the location of a hidden object after learning that another location is empty (e.g. Mody & Carey, 2016). In the current study, we asked whether children could use reasoning by exclusion to infer the identities of previously unobserved occluded objects in a task that required them to track the locations of multiple occluded objects. Forty-nine 4-7-year-olds viewed animated arrays of virtual “cards” depicting images which were then hidden by occluders. The occluders then swapped locations during the maintenance period. Children were asked to select which card was hidden in a probed location. During the encoding period, we manipulated whether children saw all the card faces (Face-up block) or all but one of the card faces (Exclusion block), for which children had to reason by exclusion to infer the target in half of the trials. We found that all children succeeded in the Face-up block, but only 6-year-olds succeed in the Exclusion block when they had to deploy logical reasoning to identify a previously-unseen hidden target. Our results suggest that children’s ability to reason by exclusion to infer the identity of a hidden target while tracking multiple objects and locations may undergo protracted development.

Children’s use of Reasoning by Exclusion to Track Identities of Occluded Objects

This study investigated processing/strategy complexity to activate magnitude representation for common fractions (1/4, 1/3, 1/2, 2/3, and 3/4), and the precision of their magnitude representations on the mental number line. We compared the magnitude comparison performance for pairs involving any one of common fractions and pairs of two uncommon fractions. We hypothesized the mean reaction time to be shorter for pairs involving a common fraction with a simple processing/strategy to activate the magnitude representation, and the magnitude distance effect to be smaller for pairs involving common fractions with precise magnitude representation. Bayesian mixed-effect regression analysis found a shorter mean reaction time for pairs involving common fractions except 1/4. Comparisons involving common fractions did not show a smaller magnitude distance effect. Thus, processing to activate magnitude representation is simpler for common (except 1/4) than for uncommon fractions, although common fractions do not have more precise magnitude representation than uncommon fractions.

Complexity of processing to activate magnitude representation for common fractions and precision of their magnitude representations in fraction magnitude comparison

From about six months of age onward, infants observing an action, such as a grasp, start to shift their gaze from the moving agent to the goal before the action is complete. A variety of factors that influence such goal-predictive gaze have been identified. However, the underlying cognitive processes are heavily debated. We propose that our minds structure sensorimotor dynamics into probabilistic, generative event-predictive models, and, choose actions with the objective to minimize predicted uncertainty. We implement this proposition by means of event-predictive learning and active inference. Trained on manual object-manipulations, the generation of goal-predictive gaze emerges: The model starts fixating the anticipated goal at the start of an observed event when a familiar agent (i.e., a hand) is involved. Meanwhile, the model keeps tracking unfamiliar agents (e.g., a claw) performing the same movement. We conclude that event-predictive learning combined with active inference may be critical for eliciting action-goal predictions.

I see where this is going: Modeling the development of infants' goal-predictive gaze

Visual comparison is used in education to convey important commonalties and differences. This process is more effective when the figures are spatially aligned so that the corresponding parts and relations are maximally clear (direct placement) (Matlen, Gentner, & Franconeri, 2020). Yet science textbooks often fail to follow this principle arranging figures meant to be compared (Jee et al., in prep)—perhaps in service of visual appeal. To explore whether this choice in fact maximizes visual appeal, we gave middle-school students illustrations characteristic of textbook figures, along with modified versions that followed direct placement principles. Students were significantly more likely to choose the direct placement version when given the goal of helping other students see differences among the figures (M=94%), than when given the goal to make the figure “look nice” (M=61%). These findings suggest that direct placements improve the educational value of a figure without sacrificing its aesthetic appeal.

Students Prefer to Learn from Figures that Include Spatial Supports for Comparison

Biologists, lay adults, and children alike value causal explanations of how biological entities work. Despite this, elementary school science education has historically lacked mechanistic content. In line with recent science education standards, we investigated the effects of mechanistic learning goals on understanding of an in-depth lesson about how the heart works. Children ages 6 to 11 who were given mechanistic learning goals performed better on knowledge assessments of the heart lesson and enjoyed their learning goal more than children who were given a relatively superficial learning goal—to focus on labels. Thus, learning goals orienting children towards mechanistic content during lessons enhance science learning and enjoyment.

Mechanistic Learning Goals Enhance Elementary Student Understanding and Enjoyment of Heart Lessons.

Humans are often faced with an exploration-versus-exploitation trade-off. A commonly used paradigm, multi-armed bandit, has shown humans to exhibit an "uncertainty bonus", which combines with estimated reward to drive exploration. However, previous studies often modeled belief updating using either a Bayesian model that assumed the reward contingency to remain stationary, or a reinforcement learning model. Separately, we previously showed that human learning in the bandit task is best captured by a dynamic-belief Bayesian model. We hypothesize that the estimated uncertainty bonus may depend on which learning model is employed. Here, we re-analyze a bandit dataset using all three learning models. We find that the dynamic-belief model captures human choice behavior best, while also uncovering a much larger uncertainty bonus than the other models. More broadly, our results also emphasize the importance of an appropriate learning model, as it is crucial for correctly characterizing the processes underlying human decision making.

Revisiting the Role of Uncertainty-Driven Exploration in a (Perceived) Non-Stationary World

Research shows that there is a strong cultural bias against women: Many people believe that women are less intellectually talented than men (Kirkcaldy et al., 2007; Upson & Friedman, 2012), and these beliefs emerge in childhood. For example, girls as young as six were less motivated than boys to pursue gender-neutral activities that were described as being for children who are “really smart” and more inclined toward activities intended for “hardworking” children (Bian, Cimpian, & Leslie, 2015). 
This study investigates whether framing activities as stereotypically masculine or feminine alters the impact of messages about intellectual ability on 6- and 7-year-olds’ selection of preferred activities. Preliminary findings (N=75) indicate that making activities stereotypically feminine reduces girls’ preference for "hardworking" over "smart" games. Such results suggest that presenting young girls with toys that are targeted specifically to them might counteract the effects of negative gender stereotypes about intelligence.

The Effects of Messages About Intellectual Ability on Children’s Activity Preferences

When an otherwise neutral stimulus signals an aversive event, we learn to use that information to avoid negative outcomes. Extended too broadly, however, fear learning can become maladaptive. In concept learning, people tend to restrict generalization to the narrowest category consistent with the learning set (Xu & Tenenbaum, 2007). We tested whether this pattern extends to fear conditioning. Shocks were associated with bananas (basic level group) or with fruit (superordinate group), but never with vegetables (both). In generalization, both groups were shown novel fruits and vegetables (no shocks). Surprisingly, while the basic level group’s Skin Conductance Response (SCR) reliably diminished from acquisition to generalization, the superordinate group’s SCR did not. Further, SCR to vegetables was greater in generalization for the superordinate group, suggesting that superordinate training encourages overgeneralizing threat beyond what concept learning models would predict. These novel findings have implications for both learning models and anxiety disorders (e.g., PTSD).

Can algorithms learn from babies? Exploring how infant learning can inform and inspire unsupervised learning algorithms

Downloads

Next from CogSci 2021

Testing an interference-based model of working memory in children with developmental language disorder and their typically developing peers

Similar lecture

The Anatomy of Discourse: Linguistic Predictors of Narrative and Argument Quality

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES