Justin Lovelace

Gates Hall

Cornell University

Ithaca, NY

My research focuses on building generative AI systems that are efficient and controllable. I develop diffusion models that operate in semantic latent spaces, enabling transparent planning, fine-grained steering, and reliable grounding across language, speech, and image domains. I am also broadly interested in understanding how to train large language models effectively, including under real-world data constraints. I am fortunate to be advised by Prof. Kilian Q. Weinberger.

I have had the opportunity to intern with several industry research labs. I am currently interning at Databricks Mosaic Research, hosted by Wen Sun and Jonathan Frankle. Previously, I interned with the Audiobox team at Meta FAIR, hosted by Andros Tjandra; the Speech AI group at Adobe Research, hosted by Rithesh Kumar and Zeyu Jin; and ASAPP Research, hosted by Felix Wu.

Before my PhD, I earned my M.S. in Language Technologies from Carnegie Mellon University, where I conducted NLP research under the advisement of Dr. Carolyn Rosé. I completed my undergraduate studies at Texas A&M University, where I engaged in Clinical NLP research with Dr. Bobak Mortazavi.

selected publications

Preprint

Prescriptive Scaling Laws for Data Constrained Training

Justin Lovelace, Christian Belardi, Srivatsa R Kundurthy, and 2 more authors

arXiv Preprint, 2026

Bib Paper Blog

@article{lovelace2026prescriptive,
  title = {Prescriptive Scaling Laws for Data Constrained Training},
  author = {Lovelace, Justin and Belardi, Christian and Kundurthy, Srivatsa R and Sudhakar, Shriya and Weinberger, Kilian Q},
  journal = {arXiv Preprint},
  year = {2026},
  paper = {https://arxiv.org/abs/2605.01640},
}

ICLR

SpeechOp: Inference-Time Task Composition for Generative Speech Processing

Justin Lovelace, Rithesh Kumar, Jiaqi Su, and 3 more authors

International Conference on Learning Representations (ICLR), 2026

Bib Paper Demo

@article{lovelace2025speechop,
  title = {SpeechOp: Inference-Time Task Composition for Generative Speech Processing},
  author = {Lovelace, Justin and Kumar, Rithesh and Su, Jiaqi and Chen, Ke and Weinberger, Kilian Q and Jin, Zeyu},
  journal = {International Conference on Learning Representations (ICLR)},
  year = {2026},
  paper = {https://arxiv.org/abs/2509.14298},
  demo = {/projects/speechop/}
}

ICLR

Adaptive Moments are Surprisingly Effective for Plug-and-Play Diffusion Sampling

Christian Belardi, Justin Lovelace, Kilian Q Weinberger, and 1 more author

International Conference on Learning Representations (ICLR), 2026

Bib Paper

@article{belardi2026adaptive,
  title = {Adaptive Moments are Surprisingly Effective for Plug-and-Play Diffusion Sampling},
  author = {Belardi, Christian and Lovelace, Justin and Weinberger, Kilian Q and Gomes, Carla P},
  journal = {International Conference on Learning Representations (ICLR)},
  year = {2026},
  paper = {https://arxiv.org/abs/2603.16797}
}

COLM

Stop-Think-AutoRegress: Language Modeling with Latent Diffusion Planning

Justin Lovelace, Christian Belardi, Sofian Zalouk, and 3 more authors

Conference on Language Modeling (COLM), 2025

Bib Paper

@article{lovelace2025stopthink,
  title = {{Stop-Think-AutoRegress}: Language Modeling with Latent Diffusion Planning},
  author = {Lovelace, Justin and Belardi, Christian and Zalouk, Sofian and Polavaram, Adhitya and Kundurthy, Srivatsa and Weinberger, Kilian Q},
  journal = {Conference on Language Modeling (COLM)},
  year = {2025},
  paper = {https://arxiv.org/abs/2602.20528}
}

NeurIPS

Latent Diffusion for Language Generation

Justin Lovelace, Varsha Kishore, Chao Wan, and 2 more authors

Conference on Neural Information Processing Systems (NeurIPS), 2023

Bib Paper

@article{lovelace2023latent,
  title = {Latent Diffusion for Language Generation},
  author = {Lovelace, Justin and Kishore, Varsha and Wan, Chao and Shekhtman, Eliot and Weinberger, Kilian Q},
  journal = {Conference on Neural Information Processing Systems (NeurIPS)},
  year = {2023},
  paper = {https://arxiv.org/abs/2212.09462}
}