Search

Compositionality, or the ability to build complex representations from discrete elements, is an essential ingredient of human intelligence. Compositionality enables people to think productively, learn fast from limited experience, and generalize knowledge to new contexts without re-learning from scratch. It is also essential in information processing systems to efficiently represent structured data and has seen application in compression and symbolic Artificial Intelligence (AI). Historically, the notion of compositionality played a central role in linguistic theory and philosophy of mind. More recently, it is attracting a surge of interest throughout the domains of cognitive science. Compositional processes are leveraged for elucidating the nature of mental representations in cognition (Dehaene et al., 2022), understanding the functional organisation of the brain (Agrawal et al., 2019), or building Artificial Intelligence systems that are robust to changes in the environment (Hupkes et al., 2020).

Networks are a useful mathematical tool for capturing the complexity of the world. Using behavioral measures, we showed that human adults were sensitive to the high-level network structure underlying auditory sequences (such as communities) even when presented with incomplete information. Their performance was best explained by a mathematical model following associative learning principles and based on the integration of the transition probabilities between adjacent and non-adjacent elements with memory decay. In a follow up MEG study, we explored the neural correlates of this hypothesis. First, the comparison of the brain responses to tone transitions adhering or not to the community structure revealed an early difference, suggesting an automatic encoding of sequence structure. Second, time-resolved decoding allowed determining the duration and overlap of the representation of each tone. The decoding performance exhibited exponential decay, resulting in a significant overlap between the representations of successive tones, enabling associative learning through Hebbian rule.

Understanding the unique evolutionary trajectory taken by humanity is impossible without an appreciation of our singular capacities for symbolic cognition and behaviour, which have evolved into the diverse practices of communication, art, reasoning, and ritual we encounter today (Deacon, 1998; Donald, 1991). But how did these capacities evolve during the Late Pleistocene?

Although there are several representativeness-based models of the Lawyer-Engineer task, it remains unclear just why people rely on representativeness-based heuristics rather than on posterior probabilities. This is especially striking because subjects have access to the rational answer: irrational answers decrease dramatically in frequency formats (Gigerenzer&Hoffrage,1995). We argue that the availability of representativeness is explained by the fact that subjects (1) engage in question-answering behavior, as predicted by theories of linguistic semantics, and (2) recursively reason on each other’s mental states, as predicted by the Rational Speech Act Theory (Frank&Goodman,2012). To test this, in a norming study, we asked participants for frequency judgments on the components of Bayes' law, using pairs of real-world professions and related descriptions. In the main experiment, an independent group gave probability judgments on lawyers-engineers problems. We compared different models built from the normed values, and found that those incorporating (1) and (2) best predicted main-experiment responses.

Among primates, humans are special in their ability to create and manipulate highly elaborate structures of language, mathematics or music. We show that this sensitivity is present in a much simpler domain: the visual perception of geometric shapes. We asked human subjects to detect an intruder shape among six quadrilaterals. Although the intruder was defined by an identical amount of displacement of a single vertex, the results revealed a geometric regularity effect: detection was considerably easier with most regular shapes. This effect was replicated in several tasks and in both uneducated adults and preschoolers. Baboons, however, showed no such geometric regularity effect even after extensive training. Baboon behavior was captured by convolutional neural networks (CNNs) but a symbolic model was needed to fit human behavior. Our results indicate that the human propensity for symbolic abstraction permeates even elementary shape perception and they suggest a new putative signature of human singularity.

How do people learn complex rules? We introduce a novel paradigm called ”Track-A-Mole”, in which participants have tolearn about and predict the moves of a cartoon mole, whose movements are generated by graphical programs. Our resultsshow that participants can learn to predict richly structured programs, and often require only few observations to do so,showing rapid learning and early insights about the underlying patterns. Moreover, we found that how learnable a programis can be predicted by features related to its complexity and compressibility. Finally, participants also show interestingpatterns of generalizations, assuming more parsimonious rules first and then gradually adjusting their predictions to morecomplex regularities, as well as matching their predictions to the general direction of movements and producing sensi-ble errors. These results extend our understanding of complex rule learning and open up future opportunities to modelsequential pattern predictions as graphical program induction.

Studies of human expertise suggest that experts and novices “see“ problems differently. Experts not only acquire a bodyof domain-specific strategies and knowledge, but also learn to quickly identify when those concepts apply to problemswithin the domain. We propose modeling these elements as an iterative process of domain-specific language (DSL)learning, while jointly training a neural network to recognize when learned concepts apply to new problems. We showthat the algorithm solves problems more accurately and quickly than either a neural network alone, or a model that simplyacquires new concepts without learning when to use them. We also examine the implicit problem representations learnedby the neural network recognition model, and find that they increasingly come to reflect abstract relationships betweenproblems, rather than surface features, as the model acquires domain expertise. A full paper and additional details areavailable at: https://sites.google.com/view/neurally-guided-expertise-mit

On mental models theories, reasoners create mental representations of information which they manipulate in order toderive new conclusions. These theories have been uniquely successful at explaining a class of attractive fallacies involvingdisjunctions. The original theories have appealed to low-level matching mechanisms (Walsh & Johnson-Laird, 2004;Koralus & Mascarenhas, 2013) to compare the models of the premises and the models of the conclusion and predict ananswer. In three experiments, we show that the check for overlap in content involved in these accounts must take place ata high level of cognition in order to incorporate complex world knowledge. We introduce variants of illusory inferencesfrom disjunction whose acceptance is accurately predicted by independant measures of confidence in causal connections.We conclude that the Revised Mental Model Theory of Khemlani et al. (2018) holds promise, but cannot account for ourdata out of the box.

The purpose of this study was to determine if dual‐energy CT (DECT) vital iodine tumor burden (ViTB), a direct assessment of tumor vascularity, allows reliable response assessment in patients with GIST compared to established CT criteria such as RECIST1.1 and modified Choi (mChoi). From 03/2014 to 12/2019, 138 patients (64 years [32–94 years]) with biopsy proven GIST were entered in this prospective, multi‐center trial. All patients were treated with tyrosine kinase inhibitors (TKI) and underwent pre‐treatment and follow‐up DECT examinations for a minimum of 24 months. Response assessment was performed according to RECIST1.1, mChoi, vascular tumor burden (VTB) and DECT ViTB. A change in therapy management could be because of imaging (RECIST1.1 or mChoi) and/or clinical progression. The DECT ViTB criteria had the highest discrimination ability for progression‐free survival (PFS) of all criteria in both first line and second line and thereafter treatment, and was significantly superior to RECIST1.1 and mChoi (p <.034). Both, the mChoi and DECT ViTB criteria demonstrated a significantly early median time‐to‐progression (both delta 2.5 months; both p <.036). Multivariable analysis revealed 6 variables associated with shorter overall survival: secondary mutation (HR = 4.62), polymetastatic disease (HR = 3.02), metastatic second line and thereafter treatment (HR = 2.33), shorter PFS determined by the DECT ViTB criteria (HR = 1.72), multiple organ metastases (HR = 1.51) and lower age (HR = 1.04). DECT ViTB is a reliable response criteria and provides additional value for assessing TKI treatment in GIST patients. A significant superior response discrimination ability for median PFS was observed, including non‐responders at first follow‐up and patients developing resistance while on therapy. [ABSTRACT FROM AUTHOR]

Networks are a useful mathematical tool for capturing the complexity of the world. In a previous behavioral study, we showed that human adults were sensitive to the high-level network structure underlying auditory sequences, even when presented with incomplete information. Their performance was best explained by a mathematical model compatible with associative learning principles, based on the integration of the transition probabilities between adjacent and nonadjacent elements with a memory decay. In the present study, we explored the neural correlates of this hypothesis via magnetoencephalography (MEG). Participants (N = 23, 16 females) passively listened to sequences of tones organized in a sparse community network structure comprising two communities. An early difference (~150 ms) was observed in the brain responses to tone transitions with similar transition probability but occurring either within or between communities. This result implies a rapid and automatic encoding of the sequence structure. Using time-resolved decoding, we estimated the duration and overlap of the representation of each tone. The decoding performance exhibited exponential decay, resulting in a significant overlap between the representations of successive tones. Based on this extended decay profile, we estimated a long-horizon associative learning novelty index for each transition and found a correlation of this measure with the MEG signal. Overall, our study sheds light on the neural mechanisms underlying human sensitivity to network structures and highlights the potential role of Hebbian-like mechanisms in supporting learning at various temporal scales. [ABSTRACT FROM AUTHOR]