The first topic that uses the resources of this project is Pre-trained Large Time Series Model for Fixational Eye Movements in Health and Disease. Fixational eye movements, including microsaccades, drift, and tremor, provide a high-frequency and largely involuntary signal of oculomotor control, with potential relevance for understanding neurological function. Although large pre-trained time-series foundation models have recently shown strong transfer potential across domains such as energy, weather, finance, and general sensor data, their applicability to fixation-based eye-tracking signals remains underexplored. This proposed project will address this gap by evaluating whether large pre-trained time-series models can be adapted to fixational eye-movement data and whether such adaptation can support clinically motivated downstream analysis. The expected contribution of the paper is to establish fixational eye movements as a novel and biologically meaningful application domain for time-series foundation models. Rather than presenting a deployable diagnostic system, the work will provide evidence, methodology, and practical guidance for adapting large pre-trained time-series models to small, clinically relevant eye-tracking datasets. The proposed study may open a path toward future foundation-model-based analysis of neurological signals captured through non-invasive eye tracking. In the meantime, the resources of this project will be used for another project: Systematic evaluation of anomaly detection algorithms on fixational eye movements. It remains unclear which anomaly detection methods are most suitable for analysing FEM time series, and how algorithmic performance depends on preprocessing choices, data representation, and anomaly type. This project will systematically evaluate time-series anomaly detection methods for FEM analysis using the TimeEval benchmarking framework. The project will benchmark more than 50 anomaly detection algorithms on a large collection of real-world and synthetic FEM datasets. Real-world eye-movement recordings will be transformed into multiple dataset variants by applying different preprocessing pipelines and signal representations. In particular, the project will evaluate how blink-removal strategies and normalization methods affect downstream anomaly detection performance, since these preprocessing steps can strongly influence the structure of FEM signals. Both eye-position and eye-velocity representations will be tested as time-series inputs in order to determine which representation is more informative for detecting abnormal FEM patterns. To assess algorithm robustness under controlled conditions, different anomaly types will be introduced into real-world FEM datasets. In addition, synthetic datasets will be generated both with and without anomalies, enabling evaluation under known ground-truth conditions. This design will make it possible to compare algorithms across anomaly types, signal representations, preprocessing methods, and dataset origins. The resulting benchmark will provide a systematic overview of which anomaly detection approaches are reliable for FEM data and under what experimental settings. The expected outcome of the project is a reproducible benchmark and methodological guidance for applying anomaly detection to fixational eye-movement data. The doctoral student carrying out this project is supervised by Dr.Mattias Nilsson, Karolinska Institutet.