Review of resting-state fMRI features across rodents and humans, comparing acquisition, preprocessing, and analysis choices that influence functional connectivity. The paper surveys time-averaged and time-varying metrics (e.g., sliding windows, quasi-periodic and co-activation patterns), links them to structural connectivity, and discusses applications to neurological and psychiatric disorders. It closes with open problems to make rodent-to-human translation more reliable.
Using Human Connectome Project data, the authors test whether parcel-wise complexity metrics (correlation dimension, approximate entropy, Lyapunov exponent) mainly reflect neural dynamics or are driven by basic BOLD properties. They find these complexity measures are reliable across scans and strongly tied to each region’s frequency profile, indicating that frequency content can confound interpretations of “complexity” in rs-fMRI.
Presents a modular ecosystem of containerized tools and workflows for petascale neuroimaging, enabling reproducible end-to-end processing across modalities like electron microscopy and X-ray microtomography. The framework standardizes storage, execution, and optimization and is demonstrated on synapse-level connectome estimation and large-scale cell density mapping.
Introduces an automated pipeline for estimating cellular densities and identifying laminar transitions in neuroanatomical images. The method combines patch extraction, cell detection, and sparse approximation for count data using total-variation regularization to robustly recover layer boundaries and density profiles across cortical and retinal samples.
Proposes ArCaDe, an automated approach to estimate spatially varying cell densities and detect laminar structure in cortical and retinal tissue. By modeling counts with a sparse, TV-regularized estimator and coupling it with patch-wise cell detection, the method recovers cytoarchitectonic transitions without manual annotation.
Demonstrates an NLP pipeline that identifies software and methods referenced in arXiv articles and uses the extracted names to explore usage patterns across research areas. The study argues that automated extraction at scale complements traditional literature surveys for tracking technologies and methods in scientific work.
Demonstrates robust large-scale intrinsic connectivity networks in white matter at rest and during task, providing templates for future network and clinical studies.
Dissertation on quasi-periodic patterns and dynamic network biomarkers across the Alzheimer's disease spectrum.
Links infraslow cortical dynamics and quasi-periodic patterns to shifts between externally and internally oriented attention states.
Introduces QPPLab, an open-source toolbox for robust detection and visualization of quasi-periodic patterns across diverse fMRI datasets.
Proposes a voxelwise projection method to better encode resting-state networks for multimodal deep learning pipelines.
Examines how default mode network dynamics and QPPs differ in improvisational musicians during creative tasks.
Uses voxel-wise fusion of functional networks and gray matter volume within a deep learning framework to classify Alzheimer’s disease.
Decomposes spatial and spectral contributions to the BOLD global signal in rats, clarifying its neural and non-neural components.
Explores Graph of Thoughts as a framework for non-linear, multi-step LLM reasoning, connecting it to planning, search, and modern structured prompting workflows.
Breaks down the math and architecture behind KANs, evaluates empirical results, and frames where they challenge or complement conventional deep nets.
Deep dive into quasi-periodic patterns and complex PCA in fMRI, highlighting how these tools map large-scale brain dynamics and inform connectivity analyses.
A debate-style session weaving together GPT-4 "Sparks of AGI", theoretical CS views of consciousness, and plant cognition to stress-test claims about intelligence.
Explains how AlphaFold’s structure predictions reshaped structural biology, why it mattered enough to influence a Nobel, and what it signals for AI in the sciences.
Traces the evolution of LLMs, outlines ChatGPT’s architecture at a high level, and hosts a structured debate on bias, deployment risk, and responsible use.
Synthesizes four major theories of consciousness, their predictions, and empirical challenges, with an eye toward how they intersect with AI and brain modeling.
Compares DALL·E 2 and Imagen on architecture, data, and evaluation; uses them as a case study in generative models, alignment, and visual creativity.
Walks through the Nature paper revealing alternating somato-cognitive networks in motor cortex and the implications for control, mapping, and neuromodulation.
Analyzes GATO as an early generalist agent: multi-task training, limitations, and what it taught us about scaling and specialization.
Presents Upside-Down RL as a goal-conditioned, supervised framing of RL and connects it to later return-conditioned and sequence-model approaches.
Builds from fundamentals to Rainbow DQN, unpacking each component and why the combined method stabilized and advanced deep value-based RL.
Introduces the RL loop, MDPs, value functions, and core intuition that grounds later deep RL methods.
Seminar on ArCaDe: automated estimation of cellular densities and laminar structure from neuroanatomical images, bridging sparse recovery and interpretable morphometrics.
EMBC presentation introducing ArCaDe as a robust, data-driven framework for mapping cytoarchitecture at scale.
Research seminar on early ArCaDe results and applications to cortical and retinal tissue.
Undergraduate talk on using NLP to trace software and method usage across scientific literature.