Using a Machine Learning Framework to Understand Intelligent Behaviour in Humans and Non-human Primates - Thesis Abstract

Aug 3, 2026

Upload by Eliana

This thesis investigates how cognitive neuroscience and machine learning jointly explain the dynamic neural processes underlying intelligent behaviour. It centers on the idea that the human brain evolved across a generalist-specialist ecological landscape, then translates these roles into computational motifs such as reservoir computing and task-optimised networks. Empirical results in non-human primates show a temporal evolution in prefrontal cortex strategies during learning, from uniform coding of task features to task-optimised representations. The work also demonstrates that prefrontal cortex can encode abstract environmental structures for transfer to novel but related tasks, and reconceptualises working memory as a dynamic computational resource. Human EEG data supports this view by showing filtering and maintenance of task-relevant information, with efficient low-dimensional neural codes emerging at decision time.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages188

Reading Time7 h 54 min

AccessFree to download

Thesis Abstract

Aims and Computational Framing

Temporal Evolution in Prefrontal Cortex

Abstract Structure Learning and Task Transfer

Working Memory as Dynamic Computational Resource

Acknowledgements

Supervisors and Academic Contributions

Friends and Family Support

Contributions

Independent Work by Thesis Sections

Data Collection and Analysis Collaboration

Using a machine learning framework to understand intelligent behaviour in humans and non-human primates Michał Józef Wójcik New College University of Oxford Trinity Term 2023 A thesis submitted in partial fulfilment of the requirements for the degree of Doctor of Philosophy in Experimental Psychology Using a machine learning framework to understand intelligent behaviour in humans and non-human primates Michał Józef Wójcik, New College, Oxford Trinity Term, 2023 Thesis abstract This thesis explores the intersection of cognitive neuroscience and machine learning to shed light on the dynamic nature of neural processes supporting intelligent behaviour. Drawing upon research proposing that the human brain evolved in a generalist-specialist ecological landscape, I focus on neural processes that support adaptability at different time scales. Using a machine learning perspective, these generalist and specialist processes were recast as computational motifs, akin to those found in reservoir computing and taskoptimised networks, respectively. The thesis posits the existence of a temporal evolution in neural strategies, from high-dimensional, generalist representations to low-dimensional, task-optimised ones, in the prefrontal cortex (PFC) over learning. This hypothesis is corroborated empirically in non-human primates, revealing that the PFC transitioned from initially coding task features uniformly to creating a taskoptimised representation. Furthermore, the thesis confirms the capacity of the PFC to capture abstract environmental structures, thus facilitating the learning of novel but related tasks. Additionally, applying a machine-learning lens, it reconceptualises working memory as a dynamic computational resource. Human EEG data supports this view, showing working memory's role in filtering and maintaining task-relevant information, with efficient, low-dimensional neural codes emerging at decision time. The studies presented here, while rooted in a machine learning framework, underscore the significance of understanding the inherent nuances and limitations of this paradigm in comprehending the complexity and richness of neural processes. This thesis was generously funded by the Clarendon Fund and the Saven European Fund. Acknowledgements First and foremost, my deep gratitude goes to my supervisors, the late Prof Mark Stokes, Prof Laurence Hunt, and Prof Nick Myers. This thesis stands as a testament to their unwavering, and perhaps more crucially, enthusiastic support. Mark, in particular, was a beacon of inspiration. The countless hours of intense discussions with him have shaped the conceptual heart of this work. He will be sorely missed not only as a mentor but also as a friend. Laurence graciously welcomed me as his student during one of the most challenging phases of my DPhil. His uncanny instinct for truth and precision has profoundly inﬂuenced the quality of this research. And to Nick, I am indebted for the vast ocean of knowledge he possesses, an expanse which he generously shared. In my academic journey, I was fortunate to cross paths with Dr Dante Wasmuht and Dr Jake Stroud. Dante, with his infectious enthusiasm, unveiled the complex world of dynamical systems and the state-space perspective to me. Jake reshaped my understanding of brain function through the lens of machine learning, thus changing my trajectory as a scientist. Their contributions to my growth cannot be overstated, and it has been an honour to work alongside them. I also wanted to thank my friend Dr Emilia Piwek who is truly one of the brightest people I know and probably will ever meet. My heartfelt thanks for the countless hours of listening to my scientiﬁc rambling, often at the most unconventional times. I owe an immense debt of gratitude to my housemates and friends, Kevin, Alessandro, Mrinalini, and Vinita. The late-night discourses and shared culinary adventures have been a source of continuous emotional support and will

FAQs

What core research question does the thesis address?

The thesis examines how cognitive neuroscience and machine learning can jointly explain the neural mechanisms that support intelligent behaviour in humans and non-human primates.

How does the thesis link brain function to machine learning concepts?

It reframes generalist and specialist neural processes as computational motifs, including reservoir computing and task-optimised networks, to study how strategies evolve over learning.

What does the thesis find about prefrontal cortex activity during learning?

In non-human primates, the prefrontal cortex transitions from initially coding task features uniformly to producing task-optimised representations, and it can capture abstract environmental structures to enable learning of related novel tasks.

Aug 3, 2026

Upload by Eliana