Andrei Mogoutov | Searcharxiv

arXiv2025

Brain states analysis of EEG predicts multiple sclerosis and mirrors disease duration and burden

Background: Any treatment of multiple sclerosis should preserve mental function, considering how cognitive deterioration interferes with quality of life. However, mental assessment is still realized with neuro-psychological tests without monitoring cognition on neuro-biological grounds whereas the ongoing neural activity is readily observable and readable. Objective: The proposed method deciphers electrical brain states which as multi-dimensional cognetoms quantitatively discriminate normal from pathological patterns in an EEG. Method: Baseline recordings from a prior EEG study of 88 subjects, 36 with MS, were analyzed. Spectral bands served to compute cognetoms and categorize subsequent feature combination sets. Result: The brain states predictor correlates with disease burden and duration. Using cognetoms and spectral bands, a cross-sectional comparison separated patients from controls with a precision of 85% while using bands alone arrived at 79%. Conclusion: We demonstrate the efficiency of the quantitative data-driven method based on brain states analysis by contrasting EEG data of patients with MS and healthy subjects. The congruity with disease severity and duration is a neurophysiological indicator for disease accumulation over time. We discuss potential applications of the approach for the monitoring of disease time course and treatment efficacy in longitudinal clinical studies in psychiatry and neurology.

q-bio.NC↗

arXiv2024

Towards Transparency: Exploring LLM Trainings Datasets through Visual Topic Modeling and Semantic Frame

LLMs are now responsible for making many decisions on behalf of humans: from answering questions to classifying things, they have become an important part of everyday life. While computation and model architecture have been rapidly expanding in recent years, the efforts towards curating training datasets are still in their beginnings. This underappreciation of training datasets has led LLMs to create biased and low-quality content. In order to solve that issue, we present Bunka, a software that leverages AI and Cognitive Science to improve the refinement of textual datasets. We show how Topic Modeling coupled with 2-dimensional Cartography can increase the transparency of datasets. We then show how the same Topic Modeling techniques can be applied to Preferences datasets to accelerate the fine-tuning process and increase the capacities of the model on different benchmarks. Lastly, we show how using Frame Analysis can give insights into existing biases in the training corpus. Overall, we argue that we need better tools to explore and increase the quality and transparency of LLMs training datasets.

cs.CL↗