PROBING PROFICIENCY-RELATED NEURAL REPRESENTATIONS WITH PCA AND ICA

Author

• Onila R. Narayana Mudalige Don, The University Of MontanaFollow

Year of Award

2026

Document Type

Thesis

Degree Type

Master of Science (MS)

Degree Name

Computer Science

Department or School/College

Computer Science

Committee Chair

Lucy L. W. Owen Ph.D.

Commitee Members

Zedong Peng Ph.D. Emily Stone Ph.D.

Keywords

Computational Neuroscience, Functional Magnetic Resonance Imaging, Principal Component Analysis, Independent Component Analysis, Second Language Acquisition, Functional Connectivity

Subject Categories

Bioelectrical and Neuroengineering | Biological Engineering | Computational Engineering | Other Computer Engineering

Abstract

How proficiency-related information is represented in task fMRI depends not only on the data themselves, but on the representational lens used to summarize them. This thesis examines second-language (L2) proficiency as a problem of representational organization rather than as a simple classification exercise. Using task fMRI from adult language learners performing semantic animacy judgments in their native language (L1) and second language (L2), I derive shared low-dimensional network representations with principal component analysis (PCA) and independent component analysis (ICA), then evaluate those representations under matched leakage-controlled decoding pipelines.

Across analyses, the central comparison is between representational frameworks rather than between feature sets alone. PCA produces broad, overlapping variance-maximizing components whose component-wise summaries are already strongly predictive of proficiency. ICA produces more spatially focal and network-selective components whose predictive value emerges more clearly when targeted functional connectivity is added among informative components. Dimensionality sweeps further show that proficiency-related signal is not tied to a single arbitrary number of components, but depends on representational scale, and that this scale differs for PCA and ICA. Cross-decomposition matching reveals only modest one-to-one spatial correspondence overall, yet several matched PCA–ICA pairs still exhibit partially shared predictive behavior, indicating that the two decompositions recover complementary but non-equivalent aspects of the same underlying signal.

These findings support three main conclusions. First, decoding can be used as a controlled probe of representational structure when the downstream predictive pipeline is held fixed. Second, proficiency-related neural information is multi-faceted: some of it is visible in marginal component summaries, while some emerges more clearly in relations among components. Third, common contrast-based constructions such as ∆(L2 − L1) do not necessarily preserve the most informative individual-difference signal. Together, the thesis advances a representational account of L2 proficiency and provides a reproducible comparative framework for studying how different low-dimensional decompositions expose different aspects of brain-behavior relationships.

Recommended Citation

Narayana Mudalige Don, Onila R., "PROBING PROFICIENCY-RELATED NEURAL REPRESENTATIONS WITH PCA AND ICA" (2026). Graduate Student Theses, Dissertations, & Professional Papers. 12634.
https://scholarworks.umt.edu/etd/12634