Research

My research lies at the intersection of Psychology, Computational Cognitive Science, and Cognitive Neuroscience, from low-level perceptual learning to complex decision-making, applying Deep Learning models to neuroimaging and behavioural data.

Currently, I am investigating semantic navigation strategies and their neural underpinnings in human adults. Additionally, I am exploring how children learn to navigate semantic spaces as they develop, and how these skills predict other cognitive cognitive processes like generalization, inference, exploration-exploitation strategies, and Pavlovian bias.

Current and past research projects include:

• Computational Models of CPS
• Exploring CPS Brain Dynamics
• Boosting CPS skills

Computational Models of CPS

How do humans efficiently solve the exploration/exploitation trade-off in CPS?

In this study, we provide compelling evidence that children balance novelty and appropriateness to generate creative associations by optimally regulating the level of exploration in the semantic search.

(A) Computational modeling pipeline using word2vec to extract word embeddings from target and response words. A similarity matrix was generated with target words as rows and responses as columns, which was used with a softmax action selection rule to build and fit the semantic explorer model to children's data via a negative log-likelihood (NLL) loss function. (B) Results include a scatterplot showing real versus fitted parameters. (C) Bar plots displaying loss values for three models (with error bars for SEM). (D) Line plots representing the loss landscape across participants. (E) raincloud plots visualizing beta parameter differences between conditions.

Publication:
Rastelli et al., PNAS Nexus (2022).

Exploring CPS Brain Dynamics

How the brain support semantic exploration and creativity in narrative generation?

In this study we investigate the neural mechanisms underlying semantic control during narrative creation, employing advanced fMRI techniques alongside cutting-edge large language models. Our findings uncover distinct patterns of brain activity that govern semantic exploration and facilitate the creative process of story generation, offering new insights into the cognitive and neural dynamics that fuel creativity.

The Connectome Harmonic Decomposition reveals spatial frequency patterns underlying the modulation of semantic control during story generation. a. Data analysis pipeline for implementing the Connectome Harmonic Decomposition. b. Brain plots showing examples of harmonics. c. Group-level searchlight classification identifies differences between conditions across the harmonic space by comparing classification accuracy to chance level, with statistical significance indicated by horizontal lines. d. Group-level searchlight regression analyzes how behavioral features relate to harmonics, measuring model performance by comparing R² scores to a baseline with permuted features, with statistical significance marked by horizontal lines.

Related publications:
Rastelli et al., biorxiv (2024).

Boosting CPS

How can we foster CPS in individuals?

The research explores whether simulated visual hallucinations in virtual reality can enhance cognitive flexibility (CF), an essential component for adapting to changing environments. Using VR videos altered by the DeepDream algorithm to simulate hallucinations, participants showed increased flexibility in their semantic networks and a reduction in automatic responses, suggesting that those experiences may facilitate the exploration of creative, less conventional strategies in problem-solving.

Experimental Design and Stimuli. (a) Visual stimuli were presented in virtual reality (VR), consisting of panoramic 360° videos of natural scenes (indicated by red frames) and their DeepDream-modified versions (blue frames). (b) Experimental design overview. Recurring arrows indicate the counterbalanced order of conditions across participants. (c) Schematic representation of the Alternative Uses Task (AUT). (d) Schematic representation of the Stroop task. (e) Radar plot displaying the results of the Altered States of Consciousness (ASC) questionnaire. Red and blue areas correspond to the OR and DD conditions, respectively. Statistical significance: *p < 0.05; **p < 0.01; ***p < 0.001. (Rastelli, 2022)

Related publications:
Rastelli et al. Sci Rep (2022);