Exploring mechanisms of typical and abnormal cognitive development: neurodevelopmental computational models of theory of mind and general intelligence
Bibliographic record
Abstract
A useful approach to better understand the mechanisms underlying cognitive development has been that of constructive artificial neural networks (CANNs). This thesis presents several CANN models that contribute to our understanding of two typical and abnormal developmental phenomena.The first two manuscripts explore the mechanisms underlying false-belief (FB) task transitions. Typically-developing preschoolers go through two transitions on verbal FB tasks, in which they have to say where an agent will search to find (approach) or avoid (avoidance) an object that was moved from location A to location B in the agent's absence. Transition 1 occurs as children go from failure to success on the approach task, and Transition 2 occurs as children go from succeeding only at approach to succeeding also at avoidance tasks. Are these transitions due to learning about beliefs or to other factors? The first manuscript presents a model of a non-verbal FB task (which uses looking time rather than a verbal measure). The model captured the transitions observed with verbal tasks, predicting that transitions would be observed on non-verbal tasks. Results suggest that initial failure could be due to observing more true-belief (TB) than FB searches, and that Transition 1 may not be due to learning about beliefs but to overcoming default TB attributions by learning to distinguish FB from TB situations. Results also suggest Transition 2 may be due to avoidance goals being represented by more varied behaviour than approach goals.Autistic children usually fail at verbal approach FB tasks, even when they are older than the typical age of success. The second manuscript explores the impact of simulating specific autistic deficits on Transition 1. First, it is thought that social deficits in autism may be related to abnormal connectivity between the brain regions used in FB tasks. I explored this hypothesis by impairing in one group of networks the connectivity of the input unit providing the information about the agent, while in a second group of networks I impaired a start or end location input unit. Results suggest that the information from the agent node is computationally crucial to Transition 1, as only the first group had impaired performance. I next simulated the decreased autistic attention to social stimuli by replacing a random half of all network training patterns by random patterns, simulating observations of random situations. Because there is currently some doubt as to whether specific, early behavioural treatment of autism improves later deficits, I simulated different times of treatment by manipulating the duration of the attention impairment in networks. As the duration of the impairment was reduced, performance progressively improved, showing that computationally, early treatment can be beneficial for autism. In the third manuscript, I explored whether white-matter integrity (WMI) could be manipulated to simulate a range of performances on Raven's Standard Progressive Matrices (SPM), a popular test of intelligence requiring subjects to analyze a matrix to find which figure, out of a few alternatives, best fits the missing figure in the matrix. Different levels of WMI have been associated with typical, age-related cognitive improvements and decline, as well as with preterm birth. To explore the effects of different levels of WMI, I incorporated different noise proportions in the activation values of my SPM model. Best performance was obtained with no impairment, but as WMI was reduced, the model's success rate was lowered to first capture the success rate of typically-developing 9-year-olds on the SPM, and with more noise it then captured the performance of 9-year-olds born preterm. These results thus computationally support a link between WMI and typical and impaired cognitive development.In sum, these results show that CANNs are unique tools to advance our understanding of typical and abnormal mechanisms of development.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.003 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.001 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.003 |
| Scholarly communication | 0.002 | 0.003 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".