Home Page
About me
Hi! My name is Heitor Baldo. I hold a BS in Mathematics and an MS in Applied and Computational Mathematics, both from the University of Campinas, and a PhD in Bioinformatics (Mathematical Neuroscience) from the University of São Paulo. I was a visiting postdoctoral researcher at Leipzig University in Germany, and I am currently a postdoctoral fellow at the University of São Paulo. I’m also an affiliate researcher at the IGDORE Institute. My background spans mathematics, applied mathematics, and computer science, with an emphasis on computational / mathematical neuroscience and artificial intelligence.
Academic Curriculum Vitae Résumé (two-pages)
Research Interests
1. Brain-Inspired Artificial Intelligence
Development and application of brain-inspired AI algorithms, encompassing spiking neural networks (SNNs), stochastic SNNs, ultra-LIF SNNs, and oscillatory neural networks (ONNs), together with biologically grounded learning rules such as dopaminergic reinforcement learning and neuromodulated synaptic plasticity.
2. Emergence in Cognitive Multi-Agent Systems.
Investigating how complex collective behaviors arise from interactions among cognitive agents, encompassing MARL-trained ethology-based animal agents, LLM-based multi-agent systems of human behavior, and LLM-based opinion-dynamics systems. A central focus is the emergence and evolution of interaction network topology, using graph theory and applied algebraic topology to identify structural signatures of phenomena such as cooperation, hierarchy, social organization, consensus, polarization, and fragmentation, and to understand the feedback between individual cognition, network topology, and collective behavior.
3. Topological and Geometric Deep Learning for Neuroscience.
Development and application of geometric, topological, and manifold-aware learning architectures (graph, hypergraph, simplicial, and sheaf neural networks, and manifold-valued models, adapted to the structure of neural and connectomic data). The goal is to build models whose inductive biases respect the higher-order, geometric, and topological nature of brain data, improving both predictive performance and interpretability relative to generic deep learning approaches.
4. Brain Connectivity Inference, Graph Theoretic and Topological Data Analysis of Brain Networks.
Brain connectivity inference methods, focusing on directed connectivity inference in the frequency domain. Multivariate autoregressive methods, such as partial directed coherence (PDC) and variants, directed transfer function (DTF), and related estimators, recover directed connectivity networks from neural signals (EEG, fMRI, MEG). From these inferred networks, we go beyond pairwise graphs toward multiway, multilayer, temporal, and dynamic representations that capture directed, higher-order neural interactions through hypergraphs, simplicial complexes, and digraph-based complexes. We characterize these structures through discrete geometry (Ollivier–Ricci and Forman–Ricci curvatures, finite geometries, and combinatorial invariants) and through graph theoretic and topological data analysis (persistent homology, filtration-based descriptors, Q-analysis, and network summary statistics), quantifying structural and functional organization across scales.
5. Neural Manifolds and Cognitive Representations.
Investigating the geometric and topological organization of neural and artificial latent representations, and determining how manifold structure encodes cognitive maps, learned world models, and behavioral states.
6. Brain Coding and Decoding
Reconstruction of stimuli, intentions, and cognitive states from measured neural activity, encompassing both the encoding problem (how sensory and cognitive variables are represented) and the decoding problem (recovering those variables from recorded signals). This integrates the representational insights of the algebraic, manifold, and topological directions above with statistical and machine learning decoders, with relevance to brain-computer interfaces (BCIs) and neurotechnology.