{"id":{"repo_id":"mcmaster","oai_identifier":"oai:macsphere.mcmaster.ca:11375/32482"},"canonical_url":"https://search.ndltd.org/etd/mcmaster/oai:macsphere.mcmaster.ca:11375/32482","repository":{"repo_id":"mcmaster","name":"McMaster University","base_url":"https://macsphere.mcmaster.ca/server/oai/request"},"display":{"title":"Geometric Deep Learning For Financial Time Series and Efficient Fine-Tuning of Foundation Models","abstract":"This thesis presents two significant research contributions: one focuses on improving the adaptation of large language models (LLMs) using parameter-efficient fine-tuning (PEFT), and the other addresses the effective modelling of history-dependent stochastic processes—specifically Volterra processes, which are commonly applied in quantitative finance. In the first part, I introduce a user-friendly adaptation pipeline that boosts the performance of a standard foundation model, bringing it much closer to a fully fine-tuned, task-specific version. Remarkably, it achieves this while using significantly less compute and memory, all while keeping data private. The pipeline leverages existing learnable low-rank adapters (LoRA) for known datasets and predicts adapter values for new datasets using this readily available information. Its main advantage is that it can run on a standard laptop without requiring GPU power, ensuring that data remains local. This method effectively closes about half of the performance gap between an untuned base model and a fully fine-tuned one, making specialized models more accessible to researchers, practitioners, and everyday users who lack expensive infrastructure or work with sensitive data on devices like smartphones. The second part addresses a computational challenge in translating the non-Markovian Volterra process into a format suitable for computation. This translation is difficult because the data history dimension affecting the current state grows with the length of the path. I propose a two-step approach to make this process manageable: first, the Volterra process is mapped onto a simpler, lower-dimensional manifold; then, a geometric deep learning model—a \"hypernetwork\"—is applied, specifically designed for the manifold’s structure. We provide both mathematical and computational evidence demonstrating the model’s effectiveness and practicality (with proofs developed by co-authors available in the main paper), along with extensive testing of each parameter to validate our approach.","abstract_html":"This thesis presents two significant research contributions: one focuses on improving the adaptation of large language models (LLMs) using parameter-efficient fine-tuning (PEFT), and the other addresses the effective modelling of history-dependent stochastic processes—specifically Volterra processes, which are commonly applied in quantitative finance. In the first part, I introduce a user-friendly adaptation pipeline that boosts the performance of a standard foundation model, bringing it much closer to a fully fine-tuned, task-specific version. Remarkably, it achieves this while using significantly less compute and memory, all while keeping data private. The pipeline leverages existing learnable low-rank adapters (LoRA) for known datasets and predicts adapter values for new datasets using this readily available information. Its main advantage is that it can run on a standard laptop without requiring GPU power, ensuring that data remains local. This method effectively closes about half of the performance gap between an untuned base model and a fully fine-tuned one, making specialized models more accessible to researchers, practitioners, and everyday users who lack expensive infrastructure or work with sensitive data on devices like smartphones. The second part addresses a computational challenge in translating the non-Markovian Volterra process into a format suitable for computation. This translation is difficult because the data history dimension affecting the current state grows with the length of the path. I propose a two-step approach to make this process manageable: first, the Volterra process is mapped onto a simpler, lower-dimensional manifold; then, a geometric deep learning model—a &quot;hypernetwork&quot;—is applied, specifically designed for the manifold’s structure. We provide both mathematical and computational evidence demonstrating the model’s effectiveness and practicality (with proofs developed by co-authors available in the main paper), along with extensive testing of each parameter to validate our approach.","abstract_has_math":false,"creators":["Arabpour Dahoei, Reza"],"institution":null,"degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":"Computational Engineering and Science","school":null,"contributors":[],"advisors":["Kratsios, Anastasis"],"committee_chairs":[],"committee_members":[],"year":2025,"date_issued":"2025-11","date_published":"2025-11","updated_at":"2026-11-19T05:57:29Z","subjects":["Machine Learning","Deep Learning","Geometric Deep Learning","Financial Mathematics","Time Series","Financial Time Series","Foundation Models","Large Language Models","LLMs","Hypernetworks","Mathematics","Artificial Intelligence"],"languages":["en"],"rights":[],"rights_urls":[],"identifier_entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://doi.org/10.71548/20013"],"render_values":[{"text":"https://doi.org/10.71548/20013","href":"https://doi.org/10.71548/20013","code":true}]}],"person_identifier_entries":[]},"links":{"outbound_url":"http://hdl.handle.net/11375/32482","outbound_label":"Handle","outbound_source":"dc:identifier.uri","record_url":"http://hdl.handle.net/11375/32482","record_label":"Repository record","record_kind":"handle","record_source":"dc:identifier.uri","download_url":null,"download_label":null,"download_source":null,"access_status":null,"embargo_until":null,"access_note":null},"source_record":{"url":"https://macsphere.mcmaster.ca/server/oai/request?verb=GetRecord&metadataPrefix=dim&identifier=oai%3Amacsphere.mcmaster.ca%3A11375%2F32482","prefix":"dim"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Kratsios, Anastasis"]},{"key":"dc:contributor.department","label":"Department","values":["Computational Engineering and Science"]},{"key":"dc:creator","label":"Author","values":["Arabpour Dahoei, Reza"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2025-10-02T16:16:51Z"]},{"key":"dc:date.available","label":"Dc Date Available","values":["2025-10-02T16:16:51Z"]},{"key":"dc:date.issued","label":"Date","values":["2025-11"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Machine Learning","Deep Learning","Geometric Deep Learning","Financial Mathematics","Time Series","Financial Time Series","Foundation Models","Large Language Models","LLMs","Hypernetworks","Mathematics","Artificial Intelligence"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language.iso","label":"Language (ISO)","values":["en"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["http://hdl.handle.net/11375/32482","https://doi.org/10.71548/20013"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["This thesis presents two significant research contributions: one focuses on improving the adaptation of large language models (LLMs) using parameter-efficient fine-tuning (PEFT), and the other addresses the effective modelling of history-dependent stochastic processes—specifically Volterra processes, which are commonly applied in quantitative finance. In the first part, I introduce a user-friendly adaptation pipeline that boosts the performance of a standard foundation model, bringing it much closer to a fully fine-tuned, task-specific version. Remarkably, it achieves this while using significantly less compute and memory, all while keeping data private. The pipeline leverages existing learnable low-rank adapters (LoRA) for known datasets and predicts adapter values for new datasets using this readily available information. Its main advantage is that it can run on a standard laptop without requiring GPU power, ensuring that data remains local. This method effectively closes about half of the performance gap between an untuned base model and a fully fine-tuned one, making specialized models more accessible to researchers, practitioners, and everyday users who lack expensive infrastructure or work with sensitive data on devices like smartphones. The second part addresses a computational challenge in translating the non-Markovian Volterra process into a format suitable for computation. This translation is difficult because the data history dimension affecting the current state grows with the length of the path. I propose a two-step approach to make this process manageable: first, the Volterra process is mapped onto a simpler, lower-dimensional manifold; then, a geometric deep learning model—a \"hypernetwork\"—is applied, specifically designed for the manifold’s structure. We provide both mathematical and computational evidence demonstrating the model’s effectiveness and practicality (with proofs developed by co-authors available in the main paper), along with extensive testing of each parameter to validate our approach."]},{"key":"dc:description.abstractgeneral","label":"General Abstract","values":["This thesis presents two contributions at the intersection of artificial intelligence and mathematics. First, I introduce a novel method for adapting large language models on widely available hardware. This approach recovers half of the performance lost when using an untuned base model instead of a GPU fine-tuned one, while running on a single laptop with minimal cost and energy consumption. It makes specialized models more accessible, preserves privacy by keeping data local, and promotes environmentally responsible computing. Second, I develop a practical framework for working with history-dependent stochastic processes commonly used in quantitative finance. Such processes are often too large to compute efficiently. The method proposed here compresses them into a low-dimensional representation and then applies a computational model, enabling efficient simulation, estimation, and practical application. Together, these contributions introduce novel algorithms capable of addressing real-world problems from fresh perspectives."]},{"key":"dc:description.degree","label":"Dc Description Degree","values":["Master of Science (MSc)"]},{"key":"dc:title","label":"Title","values":["Geometric Deep Learning For Financial Time Series and Efficient Fine-Tuning of Foundation Models"]},{"key":"dc:title.alternative","label":"Alternative Title","values":["Geometric Deep Learning for Time Series and Foundation Models"]},{"key":"others:access_status","label":"Others Access Status","values":["open.access"]}]}],"canonical_facts":{"dc:contributor.advisor":["Kratsios, Anastasis"],"dc:contributor.department":["Computational Engineering and Science"],"dc:creator":["Arabpour Dahoei, Reza"],"dc:date.accessioned":["2025-10-02T16:16:51Z"],"dc:date.available":["2025-10-02T16:16:51Z"],"dc:date.issued":["2025-11"],"dc:description.abstract":["This thesis presents two significant research contributions: one focuses on improving the adaptation of large language models (LLMs) using parameter-efficient fine-tuning (PEFT), and the other addresses the effective modelling of history-dependent stochastic processes—specifically Volterra processes, which are commonly applied in quantitative finance. In the first part, I introduce a user-friendly adaptation pipeline that boosts the performance of a standard foundation model, bringing it much closer to a fully fine-tuned, task-specific version. Remarkably, it achieves this while using significantly less compute and memory, all while keeping data private. The pipeline leverages existing learnable low-rank adapters (LoRA) for known datasets and predicts adapter values for new datasets using this readily available information. Its main advantage is that it can run on a standard laptop without requiring GPU power, ensuring that data remains local. This method effectively closes about half of the performance gap between an untuned base model and a fully fine-tuned one, making specialized models more accessible to researchers, practitioners, and everyday users who lack expensive infrastructure or work with sensitive data on devices like smartphones. The second part addresses a computational challenge in translating the non-Markovian Volterra process into a format suitable for computation. This translation is difficult because the data history dimension affecting the current state grows with the length of the path. I propose a two-step approach to make this process manageable: first, the Volterra process is mapped onto a simpler, lower-dimensional manifold; then, a geometric deep learning model—a \"hypernetwork\"—is applied, specifically designed for the manifold’s structure. We provide both mathematical and computational evidence demonstrating the model’s effectiveness and practicality (with proofs developed by co-authors available in the main paper), along with extensive testing of each parameter to validate our approach."],"dc:description.abstractgeneral":["This thesis presents two contributions at the intersection of artificial intelligence and mathematics. First, I introduce a novel method for adapting large language models on widely available hardware. This approach recovers half of the performance lost when using an untuned base model instead of a GPU fine-tuned one, while running on a single laptop with minimal cost and energy consumption. It makes specialized models more accessible, preserves privacy by keeping data local, and promotes environmentally responsible computing. Second, I develop a practical framework for working with history-dependent stochastic processes commonly used in quantitative finance. Such processes are often too large to compute efficiently. The method proposed here compresses them into a low-dimensional representation and then applies a computational model, enabling efficient simulation, estimation, and practical application. Together, these contributions introduce novel algorithms capable of addressing real-world problems from fresh perspectives."],"dc:description.degree":["Master of Science (MSc)"],"dc:identifier.uri":["http://hdl.handle.net/11375/32482","https://doi.org/10.71548/20013"],"dc:language.iso":["en"],"dc:subject":["Machine Learning","Deep Learning","Geometric Deep Learning","Financial Mathematics","Time Series","Financial Time Series","Foundation Models","Large Language Models","LLMs","Hypernetworks","Mathematics","Artificial Intelligence"],"dc:title":["Geometric Deep Learning For Financial Time Series and Efficient Fine-Tuning of Foundation Models"],"dc:title.alternative":["Geometric Deep Learning for Time Series and Foundation Models"],"dc:type":["Thesis"],"others:access_status":["open.access"]},"updated_at":"2026-11-19T05:57:29Z"}