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 5 of 5 for “"Controlled Decoding"”.
-
STEERING GENERATIVE AI ON THE FLY: INFERENCE-TIME APPROACHES FOR SAFE, RELIABLE, AND INCLUSIVE LANGUAGE MODELS
… four complementary challenges: principled decoding for alignment, safety against adversarial attacks, effective test-time reasoning, and inclusive adaptation for low-resource languages. We first establish theoretical foundations for inference-time alignment. Transfer-Q* resolves the central …
-
PROBING PROFICIENCY-RELATED NEURAL REPRESENTATIONS WITH PCA AND ICA
… those representations under matched leakage-controlled decoding pipelines. Across analyses, the central comparison is between representational frameworks rather than between feature sets alone. PCA produces broad, overlapping variance-maximizing components whose component-wise summaries are …
-
PROBING PROFICIENCY-RELATED NEURAL REPRESENTATIONS WITH PCA AND ICA
… those representations under matched leakage-controlled decoding pipelines. Across analyses, the central comparison is between representational frameworks rather than between feature sets alone. PCA produces broad, overlapping variance-maximizing components whose component-wise summaries are …
-
PRINCIPLED FRAMEWORKS FOR AI ALIGNMENT: FROM POST-TRAINING TO INFERENCE
… this end, we propose Transfer Q*, a principled controlled decoding algorithm that leverages aligned base models to estimate optimal value functions for new reward specifications, enabling provably efficient and high-quality alignment with strong personalization on real-world tasks. 2. …
-
Efficient Inference-time Control and Alignment
… alignment. We show that stable reward-guided decoding requires carefully designed token-level reward models and propose two algorithms, PARGS and FaRMA, that enable effective reward-guided generation. Third, we address the computational cost of RGTG and propose an efficient algorithm that adds …