arXiv subjects

Julius Fridriksson

Publications and source records attributed to Julius Fridriksson.

7 recordsLinked to original sources

arXiv2026

Perturbation-based Regional Interpretability through Subtraction Mapping (PRISM): naming-error dissociations in language models and post-stroke aphasia

Mechanistic interpretability of large language models lacks spatially resolved, falsifiable tools for testing whether internal components are specialized for distinct cognitive operations. We adapt subtraction analysis, the standard framework of human neuroimaging, from biological brains to perturbed transformers, and apply the same logic to both substrates in parallel. Building on the Brain-LLM Unified Model (BLUM), which showed that layer-perturbed LLaVA-1.6-Vicuna-13B error profiles match the lesion patterns of aphasic patients, we develop PRISM (Perturbation-based Regional Interpretability through Subtraction Mapping). PRISM maps the seven clinical Philadelphia Naming Test categories, subtracts error classes pairwise, and treats each perturbation seed as a subject in a group analysis with threshold-free cluster enhancement along the layer axis. We run a structurally matched analysis on 213 chronic post-stroke aphasia patients using correlation-difference lesion-symptom mapping, and replicate both sides on held-out splits. The designs match in subject dimension (seeds, patients), spatial dimension (layers, atlas-parcellated cortex) and thresholding, but the contrast operator differs: a within-subject error-proportion difference for the LLM, a between-subject correlation difference for the cortex. Both substrates recover a robust phonemic-favoring dissociation, a deep layer cluster and a frontal-perisylvian cortical cluster, both replicating; the semantic-favoring direction is a consistently signed but non-significant trend on both. PRISM thus gives a falsifiable, spatially resolved test of functional-specialization claims in transformer language models. A confirmatory ROI-level intervention (PRISM Stage 3) licensing the strongest causal-mechanism claim is left to subsequent work.

cs.LG↗

arXiv2026

Recovering Lesion Parameters from Aphasic Picture Naming Error Profiles in Large Language Models

Interpretability methods for large language models (LLMs) describe internal state but do not directly test whether that state is causally sufficient to produce the observed behavior. In earlier work, we lesioned LLMs to produce error profiles in picture naming, a central task for assessing aphasia, and found that specific lesions produced errors resembling those of individual stroke survivors. Here we ask the inverse question: given an error profile, can the lesion parameters that produced it be recovered, and what does this inverse problem reveal about transformer computation? Lesions in LLaVA-Vicuna 13B were parameterized by layer index, modification percentage, and noise sigma across 4,840 configurations, and error profiles were characterized by a seven-category clinical taxonomy (correct, semantic, unrelated, formal, mixed, neologism, no-response). We trained a multi-task neural network to map error profiles back to perturbation parameters. The problem admitted a partial solution: across 10 independently trained inverse models, modification percentage and noise sigma were recoverable, whereas layer index was recoverable only within a neighborhood. In counterfactual validation, a fresh model instance perturbed with the recovered parameters reproduced the target behavior in 81.4% of cases. This dissociation between low layer recovery and high counterfactual fidelity is consistent with functional redundancy across transformer layers, a property not captured by standard interpretability methods. As an out-of-distribution test, we applied the trained model to picture-naming error profiles from 278 stroke survivors; recovered parameters were syndrome-discriminative, most strongly for perturbation intensity, indicating generalization beyond the training distribution. Counterfactual validation provides a general framework for LLM interpretability claims beyond inverse mapping.

cs.CL↗

arXiv2026

Lesioned Multimodal Language Models Reproduce Aphasic Picture-Naming Patterns

Aphasia following stroke commonly produces systematic naming errors with characteristic profiles, but whether general-purpose language models not designed for clinical simulation can reproduce these patterns remains untested. We investigated (1) whether lesions or controlled perturbations to a multimodal language model can reproduce different types of errors in picture naming, and (2) whether the framework can reproduce the complete error profile of individual persons with aphasia (PWAs). Using LLaVA 1.6, we evaluated perturbation configurations that varied the layer, proportion, and amount of noise applied to model units. We examined 278 PWAs on the Philadelphia Naming Test, classifying responses into seven categories using a validated neural classifier. Six of seven response categories (correct, semantic, mixed, unrelated, neologism, no response errors) emerged at clinically-comparable proportions across distinct parameter space regions, with formal paraphasia being the exception. Searching the perturbation space revealed configurations that reproduced the individual error profile in at least six of seven categories for 97.8% of PWAs and in all seven categories for 79.5% of PWAs. Monte Carlo baselines confirmed that this matching reflects joint inter-category structure rather than marginal overlap. These results establish a quantitative framework for reproducing individual aphasic error patterns in picture naming. They suggest the potential for language models to serve as digital twins of individuals with post-stroke aphasia.

cs.AI↗

arXiv2026

Topological inference on brain networks with application to lesion symptom mapping

Persistent homology (PH) characterizes the shape of brain networks through persistence features. Group comparison of persistence features from brain networks can be challenging as they are inherently heterogeneous. A recent scale-space representation of persistence diagrams (PDs) through heat diffusion reparameterizes them using a finite number of Fourier coefficients with respect to the Laplace--Beltrami (LB) eigenfunction expansion of the domain, providing a powerful vectorized algebraic representation for group comparisons. In this study, we develop a transposition-based permutation test for comparing multiple groups of PDs using heat-diffusion estimates. We evaluate the empirical performance of the spectral transposition test in capturing within- and between-group similarity and dissimilarity under varying levels of topological noise and cycle location variability. In application, we propose a topological lesion symptom mapping (TLSM) method based on the proposed framework. The method is applied to resting-state functional brain networks of individuals with post-stroke aphasia to identify characteristic cycles associated with varying levels of speech-language impairment.

stat.ME↗

arXiv2026

Stroke Lesions as a Rosetta Stone for Language Model Interpretability

Large language models (LLMs) have achieved remarkable capabilities, yet methods to verify which model components are truly necessary for language function remain limited. Current interpretability approaches rely on internal metrics and lack external validation. Here we present the Brain-LLM Unified Model (BLUM), a framework that leverages lesion-symptom mapping, the gold standard for establishing causal brain-behavior relationships for over a century, as an external reference structure for evaluating LLM perturbation effects. Using data from individuals with chronic post-stroke aphasia (N = 410), we trained symptom-to-lesion models that predict brain damage location from behavioral error profiles, applied systematic perturbations to transformer layers, administered identical clinical assessments to perturbed LLMs and human patients, and projected LLM error profiles into human lesion space. LLM error profiles were sufficiently similar to human error profiles that predicted lesions corresponded to actual lesions in error-matched humans above chance in 67% of picture naming conditions (p < 10^{-23}) and 68.3% of sentence completion conditions (p < 10^{-61}), with semantic-dominant errors mapping onto ventral-stream lesion patterns and phonemic-dominant errors onto dorsal-stream patterns. These findings open a new methodological avenue for LLM interpretability in which clinical neuroscience provides external validation, establishing human lesion-symptom mapping as a reference framework for evaluating artificial language systems and motivating direct investigation of whether behavioral alignment reflects shared computational principles.

cs.LG↗

arXiv2023

False Discovery Rate Control for Lesion-Symptom Mapping with Heterogeneous data via Weighted P-values

Lesion-symptom mapping studies provide insight into what areas of the brain are involved in different aspects of cognition. This is commonly done via behavioral testing in patients with a naturally occurring brain injury or lesions (e.g., strokes or brain tumors). This results in high-dimensional observational data where lesion status (present/absent) is non-uniformly distributed with some voxels having lesions in very few (or no) subjects. In this situation, mass univariate hypothesis tests have severe power heterogeneity where many tests are known a priori to have little to no power. Recent advancements in multiple testing methodologies allow researchers to weigh hypotheses according to side-information (e.g., information on power heterogeneity). In this paper, we propose the use of p-value weighting for voxel-based lesion-symptom mapping (VLSM) studies. The weights are created using the distribution of lesion status and spatial information to estimate different non-null prior probabilities for each hypothesis test through some common approaches. We provide a monotone minimum weight criterion which requires minimum a priori power information. Our methods are demonstrated on dependent simulated data and an aphasia study investigating which regions of the brain are associated with the severity of language impairment among stroke survivors. The results demonstrate that the proposed methods have robust error control and can increase power. Further, we showcase how weights can be used to identify regions that are inconclusive due to lack of power.

stat.AP↗

arXiv2023

Network-based Statistics Distinguish Anomic and Broca Aphasia

Aphasia is a speech-language impairment commonly caused by damage to the left hemisphere. Due to the complexity of speech-language processing, the neural mechanisms that underpin various symptoms between different types of aphasia are still not fully understood. We used the network-based statistic method to identify distinct subnetwork(s) of connections differentiating the resting-state functional networks of the anomic and Broca groups. We identified one such subnetwork that mainly involved the brain regions in the premotor, primary motor, primary auditory, and primary sensory cortices in both hemispheres. The majority of connections in the subnetwork were weaker in the Broca group than the anomic group. The network properties of the subnetwork were examined through complex network measures, which indicated that the regions in the superior temporal gyrus and auditory cortex bilaterally exhibit intensive interaction, and primary motor, premotor and primary sensory cortices in the left hemisphere play an important role in information flow and overall communication efficiency. These findings underlied articulatory difficulties and reduced repetition performance in Broca aphasia, which are rarely observed in anomic aphasia. This research provides novel findings into the resting-state brain network differences between groups of individuals with anomic and Broca aphasia. We identified a subnetwork of, rather than isolated, connections that statistically differentiate the resting-state brain networks of the two groups, in comparison with standard lesion symptom mapping results that yield isolated connections.

q-bio.NC↗