{"id":{"repo_id":"nus","oai_identifier":"oai:scholarbank.nus.edu.sg:10635/317970"},"canonical_url":"https://search.dev.ndltd.org/etd/nus/oai:scholarbank.nus.edu.sg:10635/317970","repository":{"repo_id":"nus","name":"National University of Singapore","base_url":"https://scholarbank.nus.edu.sg/oai/request"},"display":{"title":"DYNAMIC NEURAL NETWORKS FOR EFFICIENT VISION MODEL INFERENCE","abstract":"Recent progress in dynamic neural networks is hindered by high training overhead from full-parameter fine-tuning and limited application beyond visual perception. This thesis addresses these challenges through three key contributions: Parameter-Efficient Fine-Tuning (PEFT): We demonstrate that dynamic networks can be developed with negligible trainable parameters, reducing FLOPs by 30% while significantly lowering training costs. Diffusion Transformers (DiT) Acceleration: We introduce a dynamic architecture for visual generation, achieving a 50% FLOPs reduction and 1.7× speedup with improved quality. Vision-Language Model (VLM) Acceleration: By implementing a dynamic small-large model cooperation mechanism, we achieve up to 3× faster inference while maintaining competitive performance. In conclusion, this research provides a comprehensive solution to reduce training costs and extends dynamic neural networks to visual generation and multi-modal domains, unlocking new possibilities for efficient real-world applications.","abstract_html":"Recent progress in dynamic neural networks is hindered by high training overhead from full-parameter fine-tuning and limited application beyond visual perception. This thesis addresses these challenges through three key contributions: Parameter-Efficient Fine-Tuning (PEFT): We demonstrate that dynamic networks can be developed with negligible trainable parameters, reducing FLOPs by 30% while significantly lowering training costs. Diffusion Transformers (DiT) Acceleration: We introduce a dynamic architecture for visual generation, achieving a 50% FLOPs reduction and 1.7× speedup with improved quality. Vision-Language Model (VLM) Acceleration: By implementing a dynamic small-large model cooperation mechanism, we achieve up to 3× faster inference while maintaining competitive performance. In conclusion, this research provides a comprehensive solution to reduce training costs and extends dynamic neural networks to visual generation and multi-modal domains, unlocking new possibilities for efficient real-world applications.","abstract_has_math":false,"creators":["ZHAO WANGBO"],"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":"2025-07-23","date_published":"2025-07-23","updated_at":"2026-07-24T03:30:34Z","subjects":["Dynamic Neural Networks; Vision Models; Efficient Inference"],"languages":[],"rights":[],"rights_urls":["https://scholarbank.nus.edu.sg/bitstreams/1b1e9b35-ac3a-4afb-a1ca-c03f4e1df669/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":["ZHAO WANGBO"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.issued","label":"Date","values":["2025-07-23"]},{"key":"dc:relation.isreferencedby","label":"Dc Relation Isreferencedby","values":["https://scholarbank.nus.edu.sg/handle/10635/317970"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Dynamic Neural Networks; Vision Models; Efficient Inference"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:rights","label":"Dc Rights","values":["https://scholarbank.nus.edu.sg/bitstreams/1b1e9b35-ac3a-4afb-a1ca-c03f4e1df669/download"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://scholarbank.nus.edu.sg/bitstreams/3e466b5e-6828-4490-98f2-275f3c5c6f36/download"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Recent progress in dynamic neural networks is hindered by high training overhead from full-parameter fine-tuning and limited application beyond visual perception. This thesis addresses these challenges through three key contributions: Parameter-Efficient Fine-Tuning (PEFT): We demonstrate that dynamic networks can be developed with negligible trainable parameters, reducing FLOPs by 30% while significantly lowering training costs. Diffusion Transformers (DiT) Acceleration: We introduce a dynamic architecture for visual generation, achieving a 50% FLOPs reduction and 1.7× speedup with improved quality. Vision-Language Model (VLM) Acceleration: By implementing a dynamic small-large model cooperation mechanism, we achieve up to 3× faster inference while maintaining competitive performance. In conclusion, this research provides a comprehensive solution to reduce training costs and extends dynamic neural networks to visual generation and multi-modal domains, unlocking new possibilities for efficient real-world applications."]},{"key":"dc:format.checksum.md5","label":"Dc Format Checksum Md5","values":["9f3c6f2aa8f96291ef77b0a18484521d","43dcf605f6026faa9dcc0a9f733eb9b6","46ab09627f92ffa9e7f37155f714229e"]},{"key":"dc:title","label":"Title","values":["DYNAMIC NEURAL NETWORKS FOR EFFICIENT VISION MODEL INFERENCE"]}]}],"canonical_facts":{"dc:creator":["ZHAO WANGBO"],"dc:date.issued":["2025-07-23"],"dc:description.abstract":["Recent progress in dynamic neural networks is hindered by high training overhead from full-parameter fine-tuning and limited application beyond visual perception. This thesis addresses these challenges through three key contributions: Parameter-Efficient Fine-Tuning (PEFT): We demonstrate that dynamic networks can be developed with negligible trainable parameters, reducing FLOPs by 30% while significantly lowering training costs. Diffusion Transformers (DiT) Acceleration: We introduce a dynamic architecture for visual generation, achieving a 50% FLOPs reduction and 1.7× speedup with improved quality. Vision-Language Model (VLM) Acceleration: By implementing a dynamic small-large model cooperation mechanism, we achieve up to 3× faster inference while maintaining competitive performance. In conclusion, this research provides a comprehensive solution to reduce training costs and extends dynamic neural networks to visual generation and multi-modal domains, unlocking new possibilities for efficient real-world applications."],"dc:format.checksum.md5":["9f3c6f2aa8f96291ef77b0a18484521d","43dcf605f6026faa9dcc0a9f733eb9b6","46ab09627f92ffa9e7f37155f714229e"],"dc:identifier.uri":["https://scholarbank.nus.edu.sg/bitstreams/3e466b5e-6828-4490-98f2-275f3c5c6f36/download"],"dc:relation.isreferencedby":["https://scholarbank.nus.edu.sg/handle/10635/317970"],"dc:rights":["https://scholarbank.nus.edu.sg/bitstreams/1b1e9b35-ac3a-4afb-a1ca-c03f4e1df669/download"],"dc:subject":["Dynamic Neural Networks; Vision Models; Efficient Inference"],"dc:title":["DYNAMIC NEURAL NETWORKS FOR EFFICIENT VISION MODEL INFERENCE"],"dc:type":["Thesis"]},"updated_at":"2026-07-24T03:30:34Z"}