Daniel Perry, Ph.D.

Research

Research publications

Efficient language models, practical evaluation, and the mathematical foundations of machine learning.

Complete Scholar profile

2026 · EACL Industry

  • Conference
  • Language models & NLP

Improving Training Efficiency and Reducing Maintenance Costs via Language Specific Model Merging

Updating one language should not require starting again with every language. Our coauthored study evaluates language-specific model merging across three tasks. It reports up to 50% less initial training time and more than 60% lower training cost for individual-language updates, while maintaining quality parity in the evaluated settings.

Read the paper

2023 · IAAI / AAAI

  • Conference
  • Language models & NLP

Compressing Cross-Lingual Multi-Task Models at Qualtrics

How do you bring multilingual NLP into production without multiplying deployment overhead? Our work combines cross-lingual, multi-task modeling with distillation. The published study covers classification in 12 languages and reports estimated hardware-cost savings of 44% compared with the original full-size model. Those are study-specific estimates, not a general guarantee.

Read the paper

2023 · EMNLP · GEM Workshop

  • Workshop
  • Language models & NLP

Effective Proxy for Human Labeling: Ensemble Disagreement Scores in Large Language Models for Industrial NLP

Evaluating language models on production data can be costly when every assessment requires fresh human labels. This coauthored study explores ensemble disagreement as a proxy for model error on keyphrase extraction across studied languages and domains. It provides evidence for a lower-cost evaluation signal, not a universal replacement for human judgment.

Read the paper

2020 · arXiv preprint

  • Preprint
  • Language models & NLP

Optimal Subarchitecture Extraction For BERT

Architecture search and knowledge distillation meet in BORT, a compact language-model architecture. This work connects an enduring research interest in selecting useful structure with practical model efficiency.

Read the paper

2020 · arXiv preprint

  • Preprint
  • Optimization

Visualization of Topology Optimization Designs with Representative Subset Selection

D. Perry, V. Keshavarzzadeh, S. Elhabian, R. Kirby, M. Gleicher, and R. Whitaker. Originally prepared for submission in 2017; posted as a preprint in December 2020.

Read the paper

2019 · SIAM/ASA Journal on Uncertainty Quantification

  • Journal
  • Matrix methods & scientific computing

Allocation Strategies for High Fidelity Models in the Multifidelity Regime

Resource allocation for expensive simulations is formulated as a regularized subset-selection problem, with comparisons against alternative allocation methods. Accepted in December 2018 and published in February 2019.

Read the paper

2017 · ECML-PKDD

  • Conference
  • Matrix methods & scientific computing

Nyström Sketches

Optimized out-of-sample sketches provide a compact representation for Nyström approximation of kernel matrices, with experiments on online learning and robustness to outliers.

Read the paper

2017 · Ph.D. thesis · School of Computing, University of Utah

  • Thesis / dissertation
  • Matrix methods & scientific computing

Data Analysis and Visualization using Basis Selection for Matrix Approximation

D. Perry.

2016 · ECML-PKDD

  • Conference
  • Matrix methods & scientific computing

Augmented Leverage Score Sampling with Bounds

A scale-aware modification of leverage-score sampling, with an error bound for deterministic sampling.

Read the paper

2016 · AISTATS · PMLR 51

  • Conference
  • Matrix methods & scientific computing

Streaming Kernel Principal Component Analysis

A streaming approach to kernel PCA combines random feature maps and matrix sketching to maintain a compact basis, with spectral-norm approximation guarantees.

Read the paper Poster Preprint Code

2013 · Journal of Cardiovascular Magnetic Resonance · 15:105

  • Journal
  • Medical imaging

Evaluation of Current Algorithms for Segmentation of Scar Tissue from Late Gadolinium Enhancement Cardiovascular Magnetic Resonance of the Left Atrium: An Open-Access Grand Challenge

A multi-institutional comparison of algorithms for left-atrial scar segmentation, coauthored with Rashed Karim and collaborators.

Read the paper

2012 · SPIE Medical Imaging

  • Conference
  • Medical imaging

Automatic Classification of Scar Tissue in Late Gadolinium Enhancement Cardiac MRI for the Assessment of Left-Atrial Wall Injury after Radiofrequency Ablation

An automated method for classifying scar tissue in left-atrial MRI, evaluated against expert annotations for overlap and estimated scar extent.

Read the paper

2012 · European Society of Cardiology Congress · Poster presentation

  • Poster / abstract
  • Medical imaging

Automatic Left-Atrial Scar Assessment in Late Gadolinium Enhancement MRI

D. Perry, A. Morris, N. Burgon, C. McGann, R. MacLeod, N. Marrouche, and J. Cates.

2012 · European Society of Cardiology Congress · Poster presentation

  • Poster / abstract
  • Medical imaging

Intensity Inhomogeneity Correction in LGE-MRI of AF Patients Is Important for Accurate LA Wall Tissue Characterization

A. Morris, J. Cates, D. Perry, N. Burgon, N. Marrouche, and E. Kholmovski.

2011 · University of Utah Undergraduate Research Abstracts · Vol. 11, p. 139 · Abstract and poster presentation

  • Poster / abstract
  • Medical imaging

Evaluation of Inter-observer and Intra-observer Agreement in Left Atrial Segmentation Using Delayed Enhancement MRI for Patients with Atrial Fibrillation

P. Anderson, W. Chao, D. Perry, T. Haslam, A. Au, and J. Cates.

2008 · Master's thesis · School of Computing, University of Utah

  • Thesis / dissertation
  • Optimization

Solving Geometric Constraint Problems with Monte Carlo Optimization

D. Perry.

2007 · Honors bachelor's thesis · School of Computing, University of Utah

  • Thesis / dissertation
  • Computer graphics

Cloud Modeling and Rendering

D. Perry.