Publications

You can also find my articles on my Google Scholar profile.

Books

Research Data that Can Be Trusted

Published in SpringerBriefs in Computer Science, Springer, 2026

A SpringerBriefs in Computer Science book on data provenance for reproducible research: descriptive dataflow operators that integrate workflow languages with a data-modeling DSL, with the Dorieh Data Platform as the exemplar and healthcare claims data as the case study.

Recommended citation: Bouzinier, M., Etin, D., Khoshnevis, N., Shad, M., & Yockel, S. (2026). Research Data that Can Be Trusted. SpringerBriefs in Computer Science. Springer. https://link.springer.com/book/9783032210319
Download Paper

Journal Articles

Comparing traditional and causal inference methodologies for evaluating impacts of long-term air pollution exposure on hospitalization with Alzheimer disease and related dementias

Published in American Journal of Epidemiology, 2025

This study builds on previous research by applying modern statistical causal inference methodologies—generalized propensity score (GPS) weighting and matching—on a large, longitudinal dataset of 50 million Medicare enrollees to investigate impacts of air pollutants (PM2.5, NO2, and O3) on elderly patients rate of first hospitalization with an ADRD diagnosis.

Recommended citation: Qin, M. M., Khoshnevis, N., Dominici, F., Braun, D., Zanobetti, A., & Mork, D. (2025). Comparing traditional and causal inference methodologies for evaluating impacts of long-term air pollution exposure on hospitalization with Alzheimer disease and related dementias. American Journal of Epidemiology, 194(1), 64-72. https://doi.org/10.1093/aje/kwae427
Download Paper

SpaCE: The Spatial Confounding Environment

Published in International Conference on Learning Representations (ICLR), 2024

We introduce SpaCE: The Spatial Confounding Environment, the first toolkit to provide realistic benchmark datasets and tools for systematically evaluating causal inference methods designed to alleviate spatial confounding. The toolkit includes training data, true counterfactuals, spatial graphs with coordinates, and realistic semi-synthetic outcomes generated using state-of-the-art machine learning ensembles.

Recommended citation: Tec, M., Trisovic, A., Audirac, M., Woodward, S., Hu, J., Khoshnevis, N., & Dominici, F. (2024). SpaCE: The Spatial Confounding Environment. In International Conference on Learning Representations (ICLR 2024).
Download Paper

GPCERF-An R package for implementing Gaussian processes for estimating causal exposure response curves

Published in Journal of Open Source Software, 2024

We present the GPCERF R package, which employs a novel Bayesian approach based on Gaussian Process (GP) to estimate the causal exposure-response function (CERF) for continuous exposures, along with associated uncertainties. The package provides a two-step end-to-end solution for causal inference with continuous exposures equipped with automatic and efficient uncertainty quantification.

Recommended citation: Khoshnevis, N., Ren, B., & Braun, D. (2024). GPCERF-An R package for implementing Gaussian processes for estimating causal exposure response curves. Journal of Open Source Software, 9(95), 5465. https://doi.org/10.21105/joss.05465
Download Paper

CRE: An R package for interpretable discovery and inference of heterogeneous treatment effects

Published in Journal of Open Source Software, 2023

CRE is an R package providing a flexible implementation of Causal Rule Ensemble, a new method for interpretable heterogeneous treatment effect (HTE) characterization in terms of decision rules, via an extensive exploration of heterogeneity patterns by an ensemble-of-trees approach. The package addresses limitations of single-tree heterogeneity discovery including instability and reduced exploration of potential heterogeneity.

Recommended citation: Cadei, R., Khoshnevis, N., Lee, K., Garcia, D. M., & Bargagli-Stoffi, F. J. (2023). CRE: An R package for interpretable discovery and inference of heterogeneous treatment effects. Journal of Open Source Software, 8(92), 5587. https://doi.org/10.21105/joss.05587
Download Paper

Application of pool-based active learning in reducing the number of required response history analyses

Published in Computers & Structures, 2020

A step by step method is presented for reducing the need for a large number of response history analyses (RHAs) in developing surrogates to predict structural responses. We apply a pool-based query-by-committee active learning algorithm to choose the most informative data samples while preserving the accuracy of prediction models used in deriving fragility curves.

Recommended citation: Kiani, J., Camp, C., Pezeshk, S., & Khoshnevis, N. (2020). Application of pool-based active learning in reducing the number of required response history analyses. Computers & Structures, 241, 106355. https://doi.org/10.1016/j.compstruc.2020.106355
Download Paper

Application of pool‐based active learning in physics‐based earthquake ground‐motion simulation

Published in Seismological Research Letters, 2019

We present the results of the application of the active learning method in developing surrogates as physics‐based earthquake ground‐motion simulators. The surrogates map input parameters into output values without demanding intensive computations, essential for parameter optimization, sensitivity, and uncertainty analysis. A step‐by‐step learning method is employed to reduce the need for generating unnecessary training data from computationally challenging numerical simulations.

Recommended citation: Khoshnevis, N., & Taborda, R. (2019). Application of pool‐based active learning in physics‐based earthquake ground‐motion simulation. Seismological Research Letters, 90(2A), 614-622. https://doi.org/10.1785/0220180144
Download Paper

Prioritizing ground‐motion validation metrics using semisupervised and supervised learning

Published in Bulletin of the Seismological Society of America, 2018

We present an analysis that looks into 11 ground‐motion validation metrics using semisupervised and supervised machine learning techniques to help label and classify goodness‐of‐fit results with the objective of prioritizing and narrowing the choice of validation metrics. This addresses the subjective nature of metric selection in ground-motion simulation validation.

Recommended citation: Khoshnevis, N., & Taborda, R. (2018). Prioritizing ground‐motion validation metrics using semisupervised and supervised learning. Bulletin of the Seismological Society of America, 108(4), 2248-2264. https://doi.org/10.1785/0120180043
Download Paper

Evaluation of the southern California seismic velocity models through simulation of recorded events

Published in Geophysical Journal International, 2016

This study compares seismic velocity models for southern California, focusing on CVM-S, CVM-H, and their variations. Using simulations of 30 moderate earthquakes (1998–2014) against observed data, results show that CVM-S4.26 provides the most accurate predictions of ground motion in the Los Angeles region.

Recommended citation: Ricardo Taborda, Shima Azizzadeh-Roodpish, Naeem Khoshnevis, Keli Cheng, Evaluation of the southern California seismic velocity models through simulation of recorded events, Geophysical Journal International, Volume 205, Issue 3, June 2016, Pages 1342–1364, https://doi.org/10.1093/gji/ggw085
Download Paper

Conference Papers

Fast Recovery Is Not Enough: Correct Goodput and Trajectory Fidelity for Foundation-Model Training on Preemptible Clusters

Published in IEEE High Performance Extreme Computing Conference (HPEC) 2026, poster paper; to appear in IEEE Xplore, 2026

Fast recovery is not the same as correct recovery. We define trajectory fidelity and correct goodput, implement and harden a correctness-aware recovery path in KempnerForge (FSDP2 x tensor parallel on PyTorch Distributed Checkpoint and SLURM requeue), and report the first at-scale measurement of what correct fault tolerance costs on dense 13B and 70B training to 160 NVIDIA H200 GPUs.

Recommended citation: Khoshnevis, N., Mazloumi, Y., Brown-Pinilla, C., Ngotiaoco, T., Balamurugan, D., Leinicke, S., & Shad, M. (2026). Fast Recovery Is Not Enough: Correct Goodput and Trajectory Fidelity for Foundation-Model Training on Preemptible Clusters. In 2026 IEEE High Performance Extreme Computing Conference (HPEC). IEEE.

Datasets and Benchmarks

MLPerf Inference v6.1: Llama 3.1-8B in FP8 on a single NVIDIA H200 (Closed Division, Datacenter)

Published in MLCommons MLPerf Inference v6.1 results (Closed Division, Datacenter), individual submitter, 2026

An audited, peer-reviewed MLPerf Inference v6.1 submission: Llama 3.1-8B with FP8 weights, activations and KV cache through TensorRT-LLM on one NVIDIA H200 SXM 141 GB, measuring 7,796.59 output tokens/s Offline and 7,376.60 output tokens/s Server within the p99 latency limits, with the full configuration and accuracy artifacts published in the MLCommons results repository.

Recommended citation: Khoshnevis, N. (2026). MLPerf Inference v6.1 results: Llama 3.1-8B, FP8, 1x NVIDIA H200 SXM 141 GB, Closed Division, Datacenter. MLCommons. https://mlcommons.org/2026/09/mlperf-inference-v6-1-results/
Download Paper

FLUX.2-dev Synthetic 2M: a large-scale synthetic text-to-image dataset

Published in Hugging Face Datasets, KempnerInstituteAI (DOI 10.57967/hf/8311), 2026

A citable Hugging Face dataset of about 2.28 million image-caption pairs generated with the FLUX.2-dev diffusion model on the Kempner AI cluster, packaged as 571 WebDataset shards with per-sample generation metadata.

Recommended citation: Khoshnevis, N., Guo, G., Vanden-Eijnden, E., Boffi, N., & Albergo, M. S. (2026). FLUX.2-dev Synthetic 2M [Data set]. Hugging Face. https://doi.org/10.57967/hf/8311
Download Paper