aiqbench analysis
D5P4 introduces a novel decoding method for discrete diffusion models that addresses the limitations of standard autoregressive search by treating beam selection as MAP inference under a partitioned Determinantal Point Process. This approach effectively balances generation quality with diversity and coverage without relying on external verifiers, showing improved performance in open-ended generation and reasoning tasks. The work highlights a critical advancement in decoding strategies for non-autoregressive models, offering a more robust framework for parallel text generation.
D5P4: Partition Determinantal Point Process for Diversity in Parallel Discrete Diffusion Decoding
arXiv:2603.19146v2 Announce Type: replace-cross Abstract: Discrete diffusion models are promising alternatives to autoregressive approaches for text generation, yet their decoding methods remain under-studied. Standard autoregressive search procedures, such as beam search, do not directly apply to iterative denoising, where hypotheses are complete intermediate sequences rather than left-to-right prefixes. Furthermore, existing diffusion decoding procedures only provide limited control over the diversity and coverage of retained hypotheses. In this work, we introduce D5P4, a beam-style decoding method tailored to discrete diffusion models, which casts intermediate beam selection as MAP inference under a partitioned Determinantal Point Process. This yields a model-internal batch objective that balances quality and diversity without external verifiers. Experiments on open-ended generation, question answering, and mathematical reasoning show that D5P4 improves diversity and pass@$k$ coverage while matching or surpassing baseline quality and fidelity
Abstract:Discrete diffusion models are promising alternatives to autoregressive approaches for text generation, yet their decoding methods remain under-studied. Standard autoregressive search procedures, such as beam search, do not directly apply to iterative denoising, where hypotheses are complete intermediate sequences rather than left-to-right prefixes. Furthermore, existing diffusion decoding procedures only provide limited control over the diversity and coverage of retained hypotheses. In this work, we introduce D5P4, a beam-style decoding method tailored to discrete diffusion models, which casts intermediate beam selection as MAP inference under a partitioned Determinantal Point Process. This yields a model-internal batch objective that balances quality and diversity without external verifiers. Experiments on open-ended generation, question answering, and mathematical reasoning show that D5P4 improves diversity and pass@$k$ coverage while matching or surpassing baseline quality and fidelity
Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
arXiv:2603.19146 [cs.AI]
(or arXiv:2603.19146v2 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2603.19146 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 223 words