Global ETD Search

Search theses and dissertations gathered from participating repositories worldwide. Every result links back to the library that holds it. No account is needed.

Results

Showing 1 to 20 of 25 for “"Preference Optimization"”.

  1. Driving Image Generation Using Interactive Preference Optimization

    … replicate the mental model of their subjective preferences and handle complex data effortlessly. Results from our first application demonstrate the effectiveness of our tool when ranking individual samples from extensive image collections. We are further able to simplify the process for users in …

    cornell Repository record for Driving Image Generation Using Interactive Preference Optimization (opens in a new tab)

  2. Magpie: Generating High-Quality Synthetic Data with Open-Source Large Language Models

    … of Magpie for filtering, generating multi-turn, preference optimization, domain-specific and multilingual datasets. We perform a comprehensive analysis of the Magpie-generated data. To compare Magpie-generated data with other public instruction datasets (e.g., ShareGPT, WildChat, Evol-Instruct, …

    washington Repository record for Magpie: Generating High-Quality Synthetic Data with Open-Source Large Language Models (opens in a new tab)

  3. Personalized and adaptive therapeutic music generation from biosignals using knowledge-guided multimodal large language models

    … evidence-grounded reasoning, and stabilized preference alignment. At the representation layer, the dissertation develops a family of tokenizers that convert continuous biomedical and media signals into compact discrete sequences for transformer-based modeling. Harmonizer provides …

    utc Repository record for Personalized and adaptive therapeutic music generation from biosignals using knowledge-guided multimodal large language models (opens in a new tab)

  4. Optimization methods for reinforcement learning: theory and applications

    … in the development and application of efficient optimization methods that enable agents to learn complex behaviors. This thesis contributes to this area by introducing novel optimization techniques and by analyzing their applications in various RL settings, encompassing cooperative multi-agent …

    oxford Repository record for Optimization methods for reinforcement learning: theory and applications (opens in a new tab)

  5. Explainable Emotion Classification in Social Media

    … model that uses generative explanations and preference alignment. Notably, we constructed a novel preference alignment dataset using GPT-4o with chain-of-thought prompting, where human annotators assessed model outputs for correctness, clarity, helpfulness, and verbosity. Utilizing this …

    ottawa-retro Repository record for Explainable Emotion Classification in Social Media (opens in a new tab)

  6. Across the Reinforcement Learning and Large Language Model Spectrum

    … sampler-dependent convergence for online direct preference optimization (DPO) and last-iterate convergence guarantees for Nash learning under non-transitive preferences. The final part turns to RL for LLMs, presenting reflection-assisted online fine-tuning for agentic tasks, and a large-scale …

    washington Repository record for Across the Reinforcement Learning and Large Language Model Spectrum (opens in a new tab)

  7. PREFERENCE LEARNING AT THE INTERFACE OF AI AND OPERATIONS

    This thesis studies preference learning at the interface between artificial intelligence and operations. The first study brings a probabilistic ranking perspective widely used in AI to operational problems. It develops a novel, tractable Mallows-type ranking model that enables preferences to be …

    nus Repository record for PREFERENCE LEARNING AT THE INTERFACE OF AI AND OPERATIONS (opens in a new tab) Download file for PREFERENCE LEARNING AT THE INTERFACE OF AI AND OPERATIONS (opens in a new tab)

  8. Analysis of artificial neural networks for solving reasoning tasks /

    … trains an error­conditioned refiner using Direct Preference Optimization on real Lean 4 compiler error pairs, achieving 45.9% Pass@1 with a 16.3% correction rate.

    vilnius Repository record for Analysis of artificial neural networks for solving reasoning tasks / (opens in a new tab)

  9. MODELING, LEARNING AND REASONING ABOUT PREFERENCE TREES OVER COMBINATORIAL DOMAINS

    … studied problems arising in various aspects of preferences: preference modeling, preference learning, and preference reasoning, when preferences concern outcomes ranging over combinatorial domains. Preferences is a major research component in artificial intelligence (AI) and decision theory, and …

    uky-thes Repository record for MODELING, LEARNING AND REASONING ABOUT PREFERENCE TREES OVER COMBINATORIAL DOMAINS (opens in a new tab)

  10. Secure internet of things using deep learning: applying transformer-based NLP techniques for enhanced IoT security

    … alignment techniques, including Contrastive Preference Optimization (CPO) and Kahneman-Tversky Optimization (KTO), these models demonstrated enhanced predictive capabilities and contextual accuracy. Their integration with static analysis workflows resulted in a hybrid framework, improving …

    qu-belfast Repository record for Secure internet of things using deep learning: applying transformer-based NLP techniques for enhanced IoT security (opens in a new tab) Download PDF for Secure internet of things using deep learning: applying transformer-based NLP techniques for enhanced IoT security (opens in a new tab)

  11. Aligning language models to professional domains using preference training

    Recent research has shown that utilizing preference training methods, such as reinforcement learning from human feedback (RLHF) or direct preference optimization (DPO), can significantly improve the alignment of models with user intent and linguistic requirements. By using these training methods, …

    reykjavik Repository record for Aligning language models to professional domains using preference training (opens in a new tab)

  12. TOWARDS PHYSICALLY-GROUNDED AND SCALABLE 3D VISUAL MODELING

    … This approach eliminates costly per-scene optimization and significantly improves scalability. Finally, we extend our framework from static scene modeling to dynamic visual world modeling. Furthermore, we develop video generation frameworks that incorporate controllability and …

    jhu Repository record for TOWARDS PHYSICALLY-GROUNDED AND SCALABLE 3D VISUAL MODELING (opens in a new tab) Embargoed until 2028-08-01

  13. Methods for Language Model Alignment: Inference-Time Sampling and Preference Fine-Tuning

    … systems, aligning their outputs with human preferences and objectives has become a central challenge. While pretrained LLMs generate fluent text, they often fail to follow instructions, meet users’ objectives, or avoid harmful content. This thesis investigates methods for aligning language …

    ethz Repository record for Methods for Language Model Alignment: Inference-Time Sampling and Preference Fine-Tuning (opens in a new tab)

  14. Efficient Algorithm for Learning and Generation with Reinforcement Learning

    … computational intractability of multi-objective preference optimization by introducing Diffusion Blend, a framework that enables inference-time multi-preference alignment without additional fine-tuning. Our theoretical insight reveals that backward diffusion processes corresponding to arbitrary …

    tamu Repository record for Efficient Algorithm for Learning and Generation with Reinforcement Learning (opens in a new tab) Embargoed until 2028-05-01

  15. Efficient Algorithm for Learning and Generation with Reinforcement Learning

    … computational intractability of multi-objective preference optimization by introducing Diffusion Blend, a framework that enables inference-time multi-preference alignment without additional fine-tuning. Our theoretical insight reveals that backward diffusion processes corresponding to arbitrary …

    tamu Repository record for Efficient Algorithm for Learning and Generation with Reinforcement Learning (opens in a new tab) Embargoed until 2028-05-01

  16. Bias, Memorization, and Identity Fidelity in Generative Models

    … further advances debiasing by adapting Direct Preference Optimization to treat underrepresented demographic classes as preferred samples, achieving state-of-the-art fairness while preserving image quality and generalizing across multiple datasets and demographic attributes. We then examine …

    jhu Repository record for Bias, Memorization, and Identity Fidelity in Generative Models (opens in a new tab)

  17. Reliable Medical LLM and Vision-Language RAG through Multi-Agent Orchestration and Single-Step Preference Alignment

    … these challenges through a unified Triple Preference Optimization framework with three-stage preference dataset curation, focusing on cross-modal alignment, retrieval balance, and a dual staged visual feedback agent. Unlike existing solutions, our method employs a single-step optimization …

    rice Repository record for Reliable Medical LLM and Vision-Language RAG through Multi-Agent Orchestration and Single-Step Preference Alignment (opens in a new tab) Embargoed until 2030-12-01

  18. Models and Application of Question Retrieval for Natural Language Processing

    … distilled into model parameters through Direct Preference Optimization (DPO) and Supervised Fine-Tuning, producing standalone models with improved coherence that surpass the inference-time approach. For retrieval systems, we apply clusters to train models for consistency: the Coherence Ranking …

    trento Repository record for Models and Application of Question Retrieval for Natural Language Processing (opens in a new tab)

  19. Defending the Dialog: Characterizing and Mitigating Harmful Fine-Tuning Attacks in Conversational AI

    … and combines synthetic healing data with direct preference optimization (DPO) to steer the chatbot toward safe responses. Optimus remains effective even when the toxicity classifier is biased or imperfect. Optimus also remains resilient against adaptive adversarial attacks that target individual …

    vt Repository record for Defending the Dialog: Characterizing and Mitigating Harmful Fine-Tuning Attacks in Conversational AI (opens in a new tab)

Page 1 of 2