Marcelo Gomes Mattar

This study shows humans allocate working memory during planning based on decision relevance, not salience, using computational models and behavioral experiments.

This paper shows neural networks can develop diverse memory strategies for free recall, including an item index code similar to memory palace techniques, which outperforms temporal context models.

The paper uses interpretable neural network models to predict human information gathering behavior and identifies brain regions involved in value of information processing.

This study shows that flexible decisions are made by sampling memories in a resource-rational way, using gaze patterns and a neural network model.

This paper develops a recurrent neural network model that learns to select computations by combining rational meta-reasoning with meta-learning, demonstrating its ability to approximate optimal symbolic models and replicate neural dynamics in macaque and human planning tasks.

The paper shows that episodic memory helps people make better decisions when future needs are uncertain by recalling detailed past events.

This study shows that the brain uses shared neural subspaces to flexibly perform and combine multiple tasks by reusing representations across different task-related information.

This paper presents a computational theory of infant gaze based on rational information gathering, showing that gaze preference is nonmonotonic with exposure due to the interplay of stimulus informativeness and future utility.

The paper develops a corticostriatal neural network model to show how cortical input compression enables distinct control and dynamic functions in the striatum across time-sensitive tasks.

This paper introduces DynamicRL, a method to estimate how humans adapt their reinforcement learning strategies over time, showing parameter updates resemble policy gradient ascent.

This study compares parameter estimation methods in reinforcement learning models, finding that deep learning outperforms traditional optimization in reliability and generalizability.

The paper develops a neural network model of planning that explains human behavior and hippocampal replay patterns in rodents.

The paper investigates how monkeys switch between learned rules and continuously update them during a perceptual categorization task, revealing distinct computational processes for rule switching and rule learning.

This paper presents a rational analysis of cognitive fatigue and boredom as opportunity cost measures in controlled behavior, offering a unified framework for understanding these phenomena.

The paper shows that reverse sequential replay supports non-local reinforcement learning in humans and is prioritized based on future utility.

This study investigates how brain activity and connectivity reorganize during learning, finding distinct patterns in early and later stages of value learning.

The paper investigates whether movement-related activity in dopamine neurons can be explained by reward prediction error (RPE) and finds that choice encoding is distinct from RPE.

The paper examines how individual differences in response precision correlate with adaptation bias in perceptual tasks using a Bayesian framework.

The paper proposes a theory that prioritizes memory access during decision-making, using hippocampal replay to explain planning and unify findings on replay mechanisms.

The paper tests if visual adaptation decorrelates shape representations in humans using behavioral and fMRI experiments, finding improved discrimination and decorrelation in object-selective cortex.

The paper predicts future motor learning from baseline sensorimotor network connectivity in humans.

This study uses machine learning to uncover dynamic subgraphs in brain networks, revealing how they constrain modular organization and explain inter-individual differences in network reconfiguration.