{"id":{"repo_id":"nus","oai_identifier":"oai:scholarbank.nus.edu.sg:10635/319151"},"canonical_url":"https://search.dev.ndltd.org/etd/nus/oai:scholarbank.nus.edu.sg:10635/319151","repository":{"repo_id":"nus","name":"National University of Singapore","base_url":"https://scholarbank.nus.edu.sg/oai/request"},"display":{"title":"CLOSED-LOOP SCALING: AUTONOMOUS IMPROVEMENT OF LLM AND LVLM REASONING","abstract":"As human-curated data approaches exhaustion, sustaining the improvement of large language models (LLMs) and large vision--language models (LVLMs) demands a paradigm shift. This thesis proposes automatic scaling: a closed-loop framework in which models autonomously improve through their own computation via three layers. Inference-time scaling treats reasoning as search guided by self-evaluation. Training-time scaling internalizes search-discovered knowledge into parameters through iterative preference alignment. Architectural grounding provides structural foundations for sustainable scaling. Through critical analysis, we identify the coherence--correctness gap as a systemic limitation of self-referential scaling and present MVP-Bench, a diagnostic benchmark revealing significant deficits in multi-level visual perception. We propose future directions in dynamic evaluation, agentic scaling, pre-linguistic reasoning foundations, and native multimodal scaling, delineating both the promise and boundaries of autonomous model improvement.","abstract_html":"As human-curated data approaches exhaustion, sustaining the improvement of large language models (LLMs) and large vision--language models (LVLMs) demands a paradigm shift. This thesis proposes automatic scaling: a closed-loop framework in which models autonomously improve through their own computation via three layers. Inference-time scaling treats reasoning as search guided by self-evaluation. Training-time scaling internalizes search-discovered knowledge into parameters through iterative preference alignment. Architectural grounding provides structural foundations for sustainable scaling. Through critical analysis, we identify the coherence--correctness gap as a systemic limitation of self-referential scaling and present MVP-Bench, a diagnostic benchmark revealing significant deficits in multi-level visual perception. We propose future directions in dynamic evaluation, agentic scaling, pre-linguistic reasoning foundations, and native multimodal scaling, delineating both the promise and boundaries of autonomous model improvement.","abstract_has_math":false,"creators":["XIE YUXI"],"institution":null,"degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":[],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2025,"date_issued":"2025-10-08","date_published":"2025-10-08","updated_at":"2026-07-24T03:31:26Z","subjects":["Credit Assignment","Order-Agnostic Modeling","Multimodal Reasoning","Automatic Scaling","Alignment","Large Language Models"],"languages":[],"rights":[],"rights_urls":["https://scholarbank.nus.edu.sg/bitstreams/27359c72-2d01-4d8d-998a-2cf09bb4ef1e/download"],"identifier_entries":[]},"links":{"outbound_url":null,"outbound_label":null,"outbound_source":null},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:creator","label":"Author","values":["XIE YUXI"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.issued","label":"Date","values":["2025-10-08"]},{"key":"dc:relation.isreferencedby","label":"Dc Relation Isreferencedby","values":["https://scholarbank.nus.edu.sg/handle/10635/319151"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Credit Assignment","Order-Agnostic Modeling","Multimodal Reasoning","Automatic Scaling","Alignment","Large Language Models"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:rights","label":"Dc Rights","values":["https://scholarbank.nus.edu.sg/bitstreams/27359c72-2d01-4d8d-998a-2cf09bb4ef1e/download"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://scholarbank.nus.edu.sg/bitstreams/bdaa8f7f-a4cc-44fa-842f-448b35f6d90e/download"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["As human-curated data approaches exhaustion, sustaining the improvement of large language models (LLMs) and large vision--language models (LVLMs) demands a paradigm shift. This thesis proposes automatic scaling: a closed-loop framework in which models autonomously improve through their own computation via three layers. Inference-time scaling treats reasoning as search guided by self-evaluation. Training-time scaling internalizes search-discovered knowledge into parameters through iterative preference alignment. Architectural grounding provides structural foundations for sustainable scaling. Through critical analysis, we identify the coherence--correctness gap as a systemic limitation of self-referential scaling and present MVP-Bench, a diagnostic benchmark revealing significant deficits in multi-level visual perception. We propose future directions in dynamic evaluation, agentic scaling, pre-linguistic reasoning foundations, and native multimodal scaling, delineating both the promise and boundaries of autonomous model improvement."]},{"key":"dc:format.checksum.md5","label":"Dc Format Checksum Md5","values":["9f3c6f2aa8f96291ef77b0a18484521d","c08463abdaa902a267049cec6194a58f","a40eae9c1bdaf3d9822c640ecf3c72bf"]},{"key":"dc:title","label":"Title","values":["CLOSED-LOOP SCALING: AUTONOMOUS IMPROVEMENT OF LLM AND LVLM REASONING"]}]}],"canonical_facts":{"dc:creator":["XIE YUXI"],"dc:date.issued":["2025-10-08"],"dc:description.abstract":["As human-curated data approaches exhaustion, sustaining the improvement of large language models (LLMs) and large vision--language models (LVLMs) demands a paradigm shift. This thesis proposes automatic scaling: a closed-loop framework in which models autonomously improve through their own computation via three layers. Inference-time scaling treats reasoning as search guided by self-evaluation. Training-time scaling internalizes search-discovered knowledge into parameters through iterative preference alignment. Architectural grounding provides structural foundations for sustainable scaling. Through critical analysis, we identify the coherence--correctness gap as a systemic limitation of self-referential scaling and present MVP-Bench, a diagnostic benchmark revealing significant deficits in multi-level visual perception. We propose future directions in dynamic evaluation, agentic scaling, pre-linguistic reasoning foundations, and native multimodal scaling, delineating both the promise and boundaries of autonomous model improvement."],"dc:format.checksum.md5":["9f3c6f2aa8f96291ef77b0a18484521d","c08463abdaa902a267049cec6194a58f","a40eae9c1bdaf3d9822c640ecf3c72bf"],"dc:identifier.uri":["https://scholarbank.nus.edu.sg/bitstreams/bdaa8f7f-a4cc-44fa-842f-448b35f6d90e/download"],"dc:relation.isreferencedby":["https://scholarbank.nus.edu.sg/handle/10635/319151"],"dc:rights":["https://scholarbank.nus.edu.sg/bitstreams/27359c72-2d01-4d8d-998a-2cf09bb4ef1e/download"],"dc:subject":["Credit Assignment","Order-Agnostic Modeling","Multimodal Reasoning","Automatic Scaling","Alignment","Large Language Models"],"dc:title":["CLOSED-LOOP SCALING: AUTONOMOUS IMPROVEMENT OF LLM AND LVLM REASONING"],"dc:type":["Thesis"]},"updated_at":"2026-07-24T03:31:26Z"}