Go to Polygence Scholars page
Katherine Brown's cover illustration
Polygence Scholar2022
Katherine Brown's profile

Katherine Brown

Oakdale High SchoolClass of 2024Oakdale, CA


Hi, I'm Katherine, class of 2024 at Oakdale High School. Through my own personal experience and interest in technology, I became quickly captivated by computational neuroscience. I started my first project in an effort to gain familiarity with the subject and to propose future applications of the discipline. I am most interested in the link between cognition and bioelectricity in the brain and hope to study this in the form of neural engineering post-high school. Other than academics, I run the Body Brain Initiative which is an organization created to spread practical solutions to nervous system regulation, chronic illness recovery, and advocacy for public health policy reform. Additionally, I operate my livestock operation through FFA and participate in leadership opportunities across the country.


  • "How do modern bio-circuit models (ring attractor network) and machine learning (convolutional neural networks) capture human cognition, and how do they compare?" with mentor Anne (Dec. 14, 2022)

Project Portfolio

How do modern bio-circuit models (ring attractor network) and machine learning (convolutional neural networks) capture human cognition, and how do they compare?

Started June 30, 2022

Abstract or project description

Though we have made notable advances in neuroscientific research and technological development, the exact relationship between the biological hardware of the brain and conscious cognition has yet to be discovered. We know what the hardware is at a biophysical level and how neurons communicate, but the individual complexities of how those pieces interact to give rise to a cognitive experience are harder to understand.
So far, researchers have developed modern bio-circuit models that capture and record different aspects of the brain, such as the spiking in individual neurons, and more complicated models that capture navigational information like ring attractor networks. These capture lab measurements well, but are often not as useful for cognitive tasks the way current machine learning approaches are. However, machine learning approaches are often missing a human element or understanding that can make them act in unpredictable or inhuman ways despite their high achievements. To understand cognition, it is important to understand these two types of models (bio and machine learning) and their capabilities. This project as such aims to review the benefits and limitations of bio-circuit models and machine learning models. To do so, we must understand and familiarize ourselves with the types of models, their primary functions, and their limitations.

Modern bio-circuit models can capture many features of working memory (WM), how memory is processed, and memory tagging & consolidation. These models capture something about the interaction between neurons that is often missing from non-biological models. However, these models are difficult to use in a computational task and do not possess all tools needed to complete cognitive tasks. The tools required remain unknown; what we do know is that there is a property of neurons and neuronal networks that is acting as an unknown variable.

Comparing bio-circuit models against machine learning acts as a form of cross-examination or a system of equations in a math problem to get closer to a missing variable. Machine learning is notorious for being compared to human-level intelligence, meaning its analytical and problem-solving abilities are also functioning as a model of processes similar to the human brain's capabilities.

Knowing each model's capabilities and strengths is essential to differentiate the two types of models. However, knowing their strengths and weaknesses does not necessarily define how accurate they are to even a single, well-understood functional aspect of the human brain. As the realm of cognitive science advances and connects to neuroscience, and as neuroscience further connects to computer science, we become closer to achieving an accurate model of all functions of the brain, even those we have yet to fully comprehend. This knowledge gap, as well as the resources we have today, allow us to study the quality and accuracy of these models. As such, we aim to answer this question: how do modern bio-circuit models (ring attractor networks) and machine learning (convolutional neural networks) capture human cognition, and how do they compare?