580 California St., Suite 400
San Francisco, CA, 94104
Key research themes
1. How do connectionist neural mass and field models elucidate effective connectivity and cortical dynamics in brain function?
This research area focuses on computational models that capture neuronal population dynamics and their interactions in the brain, especially as applied in Dynamic Causal Modeling (DCM). It addresses modeling at a mesoscopic scale, balancing biological realism and tractability to infer effective connectivity across brain regions from electrophysiological and neuroimaging data. Emphasis is placed on different classes of models (neural mass, neural field, conductance-based), their mathematical formulations (ODEs, PDEs), and their ability to explain spectral and laminar-specific responses, accounting for synaptic and ion-channel dynamics. This theme matters because it provides the mechanistic basis for interpreting brain signals, bridging microscopic physiology and macroscopic measurements, and enables testing hypotheses regarding synaptic function and cortical organization.
Key finding: This paper reviews different neuronal population models utilized in Dynamic Causal Modeling, distinguishing neural mass models that treat populations as point processes from neural field models incorporating spatial... Read more
Key finding: By introducing a Bayesian framework using Markov Chain Monte Carlo methods, this work enhances connective field (CF) modeling, which assesses the spatial dependencies between activity in distinct cortical visual field areas.... Read more
Key finding: This comprehensive review situates connectionist models in the context of neural computation principles, emphasizing knowledge representation as distributed connection weights and processing as emergent patterns of activation... Read more
2. How can connectionist neural networks model linguistic processing integrating symbolic rules and associative memory?
This area investigates connectionist architectures designed to model language phenomena where rule-governed processes and memory-based associations coexist. The goal is to simulate linguistic generalization and irregularity in morphology and syntax, capturing empirical linguistic data such as verb inflections in complex languages (e.g., German). These models aim to reconcile the utility of symbolic rules with emergent connectionist representations in layered neural models, reflecting cognitive theories about language processing that blend rule application and lexical storage. This serves to deepen understanding of language acquisition and processing from a neurocomputational perspective and informs computational linguistics and cognitive science.
Key finding: This work demonstrates a modular neural network architecture combining a connectionist short-term memory with an associative memory lexicon to learn the German verb paradigm. The network simulates dual-route processing:... Read more
Key finding: This article elaborates on connectionist models applied to syntactic processing, highlighting how such networks learn to represent constituent structure, recursion, and grammar directly from exposure to raw input. It... Read more
Key finding: This paper presents an approach where shared weights in multi-task neural networks capture high-level features common across structurally analogous tasks, facilitating transfer learning in language-related tasks. It shows... Read more
3. How do connectionist frameworks enable modeling of neural coding, population representations, and cognitive decision processes?
This theme concerns computational models using connectionist neural networks to represent population-level neural activity and model how such populations encode sensory stimuli and support cognition, including decision making. It explores normative and generative models of neural encoding and decoding, statistical properties of responses to natural stimuli, and neural implementations of reinforcement learning and cognitive biases. This line of work is crucial for linking neuronal population dynamics with behavioral and cognitive outcomes, advancing interpretability of neural codes, and developing biologically plausible AI architectures.
Key finding: This study establishes that model neurons with oriented receptive fields, appropriately modeled normalization, and encoding noise produce Gaussian-distributed responses to natural images with scale-invariant stimulus-driven... Read more
Key finding: Proposing a normative framework combining efficient coding and Bayesian decoding, this paper models neural populations as jointly optimizing stimulus encoding and probabilistic decoding via a variational autoencoder... Read more
Key finding: Extending classical neural network models with reaction-diffusion dynamics inspired by quantum hydrodynamics, this work introduces a deterministic framework called Neurohydrodynamics. By incorporating a neuropotential... Read more
Key finding: This experimental and modeling study reveals that humans exhibit systematic choice and inference biases during probabilistic learning tasks involving multiple cues with unequal reward contingencies. A biophysically plausible... Read more