About

Deepak Agarwal, MD, is an Assistant Professor at the University of Minnesota specializing in the endoscopic management of benign prostatic hyperplasia and urinary tract stones. He attended medical school at Indiana University School of Medicine and completed a residency in Urology at the Mayo Clinic in Rochester, MN. He also completed a fellowship at Indiana University Department of Urology, with a special emphasis in Holmium Laser Enucleation of the Prostate (HoLEP) and surgeon-guided access for percutaneous nephrolithotomy. Dr. Agarwal's clinical interests include the surgical treatment of enlarged prostate and kidney stones, with a focus on offering patients a variety of options to select the best treatment for their situation. He is engaged in clinical research related to enlarged prostate and kidney stones, aiming to continue the rich history of endourology at the University of Minnesota and to provide patients with innovative and effective treatments.

Research topics

• Medicine
• Surgery
• Internal medicine
• Computer Science
• General surgery
• Radiology
• Gerontology
• Physical therapy
• Psychiatry
• Law
• Urology
• World Wide Web

Selected publications

• IP63-15 WHO IS MOST AT RISK WHEN TEMPERATURES RISE? STRATIFIED EFFECTS OF HEAT ON KIDNEY STONE EMERGENCY VISITS

  The Journal of Urology · 2026-04-27

  articleSenior author
  PublisherDOI

• IP67-15 BEYOND INFECTION: SOCIODEMOGRAPHIC DRIVERS OF ANTIBIOTIC USE IN PATIENTS WITH KIDNEY STONES IN THE EMERGENCY DEPARTMENT

  The Journal of Urology · 2026-04-27

  articleSenior author
  PublisherDOI

• Precision Tracked Transformer via Kalman Filtering, Kriging and Process Noise

  ArXiv.org · 2026-05-12

  articleOpen access

  The Transformer is the foundational building block of modern AI, yet offers no principled handling of \emph{uncertainty}, which is prevalent in real applications: cold-start tokens with sparse histories in sequential recommendation, heterogeneous signal quality in language models, and attention sinks induced by unconstrained softmax. Every token is treated with uniform confidence. We show this uniformity is a degenerate case of our \emph{Bayesian Filtering Transformer} (BFT): attention becomes precision-weighted kriging, the residual connection becomes a Kalman update with adaptive gain, and the FFN becomes a dynamics model propagating precision via a Jacobian--plus--process-noise rule. Observation precision comes from a parameter-free Restricted Maximum Likelihood (REML) estimator with a conjugate Bayesian prior. BFT replaces any Transformer layer with negligible overhead. On sequential recommendation, BFT applied to three major architectures yields significant gains on six benchmarks, with the largest improvements on cold-start users and rare items where uncertainty is highest. On supervised fine-tuning of large language models with noisy data, BFT improves robustness in two regimes: noisy supervision (token-label corruption in question answering) and noisy context (retrieval-augmented QA with real RAG distractors). A single principled modification -- restoring precision -- unlocks substantial headroom across both classical sequence-modeling and modern LLM regimes.

  PublisherOA PDF

• Precision Tracked Transformer via Kalman Filtering, Kriging and Process Noise

  arXiv (Cornell University) · 2026-05-12

  preprintOpen access

  The Transformer is the foundational building block of modern AI, yet offers no principled handling of \emph{uncertainty}, which is prevalent in real applications: cold-start tokens with sparse histories in sequential recommendation, heterogeneous signal quality in language models, and attention sinks induced by unconstrained softmax. Every token is treated with uniform confidence. We show this uniformity is a degenerate case of our \emph{Bayesian Filtering Transformer} (BFT): attention becomes precision-weighted kriging, the residual connection becomes a Kalman update with adaptive gain, and the FFN becomes a dynamics model propagating precision via a Jacobian--plus--process-noise rule. Observation precision comes from a parameter-free Restricted Maximum Likelihood (REML) estimator with a conjugate Bayesian prior. BFT replaces any Transformer layer with negligible overhead. On sequential recommendation, BFT applied to three major architectures yields significant gains on six benchmarks, with the largest improvements on cold-start users and rare items where uncertainty is highest. On supervised fine-tuning of large language models with noisy data, BFT improves robustness in two regimes: noisy supervision (token-label corruption in question answering) and noisy context (retrieval-augmented QA with real RAG distractors). A single principled modification -- restoring precision -- unlocks substantial headroom across both classical sequence-modeling and modern LLM regimes.

  PublisherDOI

• Sampling for Quality: Training-Free Reward-Guided LLM Decoding via Sequential Monte Carlo

  arXiv (Cornell University) · 2026-04-07

  articleOpen accessSenior author

  We introduce a principled probabilistic framework for reward-guided decoding in large language models, addressing the limitations of standard decoding methods that optimize token-level likelihood rather than sequence-level quality. Our method defines a reward-augmented target distribution over complete sequences by combining model transition probabilities with prefix-dependent reward potentials. Importantly, the approach is training-free: it leaves model weights unchanged and instead modifies the inference distribution via reward potentials, with all gains arising purely from inference-time sampling. To sample from this distribution, we develop Sequential Monte Carlo algorithms, including a computationally efficient prefix-only variant and a lookahead variant whose intermediate targets match the exact marginals of the full sequence distribution. The framework also integrates resample-move updates with Metropolis-Hastings rejuvenation and supports block-wise generation, subsuming common decoding strategies such as temperature sampling and power-tempered objectives. Empirical results across three 7B models show significant gains. On code generation (HumanEval), our method improves base performance by up to 54.9% and surpasses the strongest sampling baselines by 9.1%-15.3%. On mathematical reasoning (MATH500), it achieves gains of up to 8.8%. Notably, it reaches 87.8% on HumanEval and 78.4% on MATH500 with Qwen2.5-7B, consistently outperforming the reinforcement learning method GRPO.

  PublisherOA PDF

• Sampling for Quality: Training-Free Reward-Guided LLM Decoding via Sequential Monte Carlo

  arXiv (Cornell University) · 2026-04-07

  preprintOpen accessSenior author

  We introduce a principled probabilistic framework for reward-guided decoding in large language models, addressing the limitations of standard decoding methods that optimize token-level likelihood rather than sequence-level quality. Our method defines a reward-augmented target distribution over complete sequences by combining model transition probabilities with prefix-dependent reward potentials. Importantly, the approach is training-free: it leaves model weights unchanged and instead modifies the inference distribution via reward potentials, with all gains arising purely from inference-time sampling. To sample from this distribution, we develop Sequential Monte Carlo algorithms, including a computationally efficient prefix-only variant and a lookahead variant whose intermediate targets match the exact marginals of the full sequence distribution. The framework also integrates resample-move updates with Metropolis-Hastings rejuvenation and supports block-wise generation, subsuming common decoding strategies such as temperature sampling and power-tempered objectives. Empirical results across three 7B models show significant gains. On code generation (HumanEval), our method improves base performance by up to 54.9% and surpasses the strongest sampling baselines by 9.1%-15.3%. On mathematical reasoning (MATH500), it achieves gains of up to 8.8%. Notably, it reaches 87.8% on HumanEval and 78.4% on MATH500 with Qwen2.5-7B, consistently outperforming the reinforcement learning method GRPO.

  PublisherDOI

• A prospective survey evaluating the visual quality of KARL STORZ fiberoptic, digital, and disposable flexible ureteroscopes

  World Journal of Urology · 2025-02-12

  article
  PublisherDOI

• Multicenter Review of Obstetric Management and Complications of Ureteroscopy During Pregnancy

  Journal of Endourology · 2025-10-01 · 1 citations

  article

  Introduction and Objectives: Managing nephrolithiasis during pregnancy requires collaboration between obstetricians and urologists. When surgical intervention is needed, ureteroscopy (URS) is a safe and effective; however, limited data exist on obstetric (OB) management and complications of URS during pregnancy. This multicenter study aimed to clarify OB practice patterns and complications of URS in pregnant patients. Methods: A multicenter retrospective review of pregnant patients who underwent URS with fellowship-trained endourologists at seven institutions from 2015 to 2024 was performed. We recorded patient demographics, indications for URS, preoperative workup, intraoperative details, perioperative OB involvement, fetal monitoring (FM) practices, and postoperative OB complications within 30 days. Results: We identified 72 cases of URS during pregnancy. Mean maternal age was 30 years, and mean gestational age was 23.5 weeks. Preoperative imaging was performed in all cases, with ultrasound used in 70/72 (97%). No intraoperative complications were noted. OB involvement and FM varied among institutions—one site required perioperative OB presence with FM, five sites performed FM case-by-case, and one site required pre- and postoperative nonstress tests (NST). FM was done in 11/72 cases, while intraoperative FM was performed in 16/72 cases. OB was present in 13/72 cases. OB complications occurred in eight cases. There were no cases of preterm labor in the immediate perioperative; however, three cases had postoperative admissions with spontaneously resolving contractions and abnormal NST. The two cases of preterm labor requiring C-section at 27 weeks were linked to OB comorbidities. Conclusions: URS is safe in pregnancy in the care of urologists who are experienced in stone disease. OB management and FM during URS are heterogeneous across institutions. In this multicenter series, there were no cases of preterm labor. Cases of early contractions and abnormal NST resolved spontaneously. Standardization of OB practice patterns during URS could be beneficial and incorporated in further guideline development.

  PublisherDOI

• IP06-23 MULTICENTER REVIEW OF OBSTETRIC MANAGEMENT AND COMPLICATIONS OF URETEROSCOPY DURING PREGNANCY

  The Journal of Urology · 2025-04-08

  article
  PublisherDOI

• Construction Barge Performance Management – Artificial Intelligence Based Software

  2025-04-21

  articleSenior author

  Abstract This paper presents a comprehensive solution for enhancing the performance management of construction barges in offshore oil and gas operations. The proposed system leverages artificial intelligence (AI) and machine learning (ML) to optimize workforce allocation, improve human efficiency, and reduce non-productive time (NPT). The solution aims to provide data-driven insights and recommendations for better decision-making in barge management.

  PublisherDOI

Frequent coauthors

• Tim Large

  133 shared

• Marcelino Rivera

  104 shared

• Amy E. Krambeck

  94 shared

• Mark A. Assmus

  73 shared

• Matthew Lee

  The Ohio State University

  50 shared

• Charles U. Nottingham

  44 shared

• Bee-Chung Chen

  33 shared

• Yunku Yeu

  18 shared

Labs

• Timm-Nelson LabPI