Dr. Vahram Martirosyan

Hosted author manuscripts →

• Martirosyan, V. “Accuracy and Interpretability from a Single Solver: Polynomial Regression up to Degree 5 for Concrete Compressive Strength Prediction.” PDF
  ICACTE 2026 (IEEE-indexed conference), 2026.

  Degree-5 R² = 0.859, RMSE 6.25 MPa; 88% term reduction with 74% of the retained terms aligning with known concrete physics; single closed-form solve, no gradient descent.
• Martirosyan, V. “Efficient and Interpretable On-Device Polynomial Inference for Multi-Variable Ocean Profiling.” PDF
  eICocean 2027 (IEEE-indexed conference, accepted), 2027.

  R² = 0.973 temperature, 0.794 salinity, 0.895 dissolved oxygen; 4.4 µs desktop inference; 0.88 ms on ESP32 and 1.12 ms on RP2350; 180 µJ/96 µJ per inference.
• Martirosyan, V. “Automated Multi-Class ECG Arrhythmia Classification Using an Ensemble of Polynomial Regression Models.” PDF
  IEEE ICCA 2026 (accepted), 2026.

  78.32% balanced accuracy, 68.23% macro-F1; sparsification reduces 123.1M terms to 330K — 1.3 MB, a 99.7% reduction — within ≈1 pp.
• Martirosyan, V. “LEAP: Lifting Legacy x86-64 Binaries to Recompilable IR for Automatic Parallelization.” PDF
  ITCA 2026 (accepted), Dec 2026.

  43-program liftability study: 100% at -O0 and 98% at -O1/-O2/-O3; 3.7× geomean from source IR; uplifted binary 10.4× over -O0 and 2.2× over -O2; 958× peak GPU offload; 377/377 tests pass.
• Martirosyan, V. “Deterministic and Auditable On-Device Polynomial Inference for On-Board SAR Oil-Spill Screening.” PDF
  IEEE EIECS 2026 (accepted), 2026.

  657 KB → 82 KB int8 model at G-mean 0.822; 15 KB model at G-mean 0.803 fits an ATmega328P; 3.76-9.2 ms/tile on real microcontrollers.
• Tsyvarev, A. V., Martirosyan, V. A. “Testing of Linux File System Drivers.” DOI
  Proceedings of the Institute for System Programming of the RAS, vol. 23, pp. 413-426, 2012.
• Tsirunyan, K., Martirosyan, V., Tsyvarev, A. “The Spruce System: Quality Verification of Linux File Systems Drivers.” DOI
  SYRCoSE 2012, May 2012.
• Martirosyan, V., Gishyan, S., Shatokhin, E. “Dynamic Verification of Linux File System Drivers.” Link
  Computer Science and Information Technologies Conference (CSIT 2011), Yerevan, Sep 2011.
• Martirosyan, V. et al. “The Spruce System: Linux Kernel Module Verification Automation (in Russian).”
  7th Annual Conference of RAU, Dec 2012.
• Martirosyan, V. et al. “Fault Simulation in Linux Kernel Module Verification (in Russian).”
  7th Annual Conference of RAU, Dec 2012.
• Martirosyan, V. “Linux Verification - Challenge of Today.” Link
  7th Annual Conference of RAU, Dec 2012.
• Martirosyan, V. et al. “X2C - XML To Code: Advantages of Standards.”
  4th Scientific Annual Conference of RAU, Dec 2009.
• Martirosyan, V. et al. “X2C: Code Generation Technology.”
• Martirosyan, V. “Programming via Operating Systems.”
  Hilfmann Press, Feb 2022.
• Martirosyan, V. “Programming via Operating Systems, Second Edition.” Link
  Hilfmann Press, Sep 2024.
• Martirosyan, V., Shatokhin, E., Rubanov, V. “Basics of Programming in OS Linux Environment (in Russian).” Link
  MIPT, Apr 2011.
• Martirosyan, V. et al. “Tutorial: Programming in GNU/Linux Environment - From System Calls to the GUI (in Russian).” Link
  RAU Publishing, Dec 2010.
• Martirosyan, V. “Parallelization of the Moore-Penrose Inverse Algorithm for a Special Bidiagonal Matrix (in Russian).”
  Herald of Russian-Armenian (Slavonic) University, 2009.
• Hakopian, Yu. R., Martirosyan, V. “On the Parallelization of the Matrix Pseudoinversion Algorithm (in Russian).”
  2009.
• Martirosyan, V. “An Algorithm for the Pseudoinversion of Bidiagonal Matrices (in Russian).”
  Herald of Russian-Armenian (Slavonic) University, 2008.
• Martirosyan, V. “On the Pseudoinversion of Bidiagonal Matrices (in Russian).”
  Annual Scientific Conference of Russian-Armenian University, 2007.

• Martirosyan, V. “Interpretable Polynomial Zone Ensembles for Global Ocean Temperature and Salinity Profile Reconstruction.”
  Under review, Q1 journal (TGRS / JAMES / Comput. Geosci.), 2026.

  WOA 2018, 534,090 samples; pooled R² = 0.9703 temperature / 0.8376 salinity; sub-millisecond inference, sub-megabyte footprint.
• Martirosyan, V. “Depth-Zone Polynomial Regression for Global Ocean Climatology.”
  Under review, IEEE journal (Q1 attempt), 2026.

  IAP T/S Climatology 2024; elastic-net degree-6 models per zone: mixed layer, thermocline, deep ocean; spatially adaptive subsampling 534K → 177K rows.
• Martirosyan, V. “Efficient Multi-Class ECG Classification Using Polynomial Regression with an Analytical Solver.”
  Applied Computational Intelligence and Soft Computing (major revision), 2026.

  82.75% ± 1.03% six-class accuracy, macro-F1 0.758; 12 µs inference; outperforms CNN, random forest, XGBoost and SVM baselines.
• Martirosyan, V. “Sparse Polynomial Regression for Interpretable Yield-Strength Prediction in High-Entropy Alloys.”
  Under review, IEEE ITT 2026, 2026.

  Degree-3 R² = 0.738 on n=1,014, statistically tied with random forest and XGBoost and significantly above a neural network (0.618); 8 of 11 retained terms map to known hardening mechanisms; BCC-only R² = 0.841.
• Martirosyan, V. “Efficient Crop Type Mapping via Degree-2 Polynomial Feature Expansion of Multi-Temporal Optical and SAR Data.”
  Submitted to IEEE TGRS, 2026.

  98.67% test accuracy and macro-F1; 2.3 s training on 16 cores; 17× SIMD speedup.
• Martirosyan, V. and N. Martirosyan “Physics-Informed Polynomial Regression for Certified Concrete Mix Design.”
  Under review, ICACTE 2027, 2026.

  Physics-informed degree-3 SRPR: R² = 0.919, RMSE 4.73 MPa; certified global inverse design with 10.6-14.5% cost and 16.7% CO_2 savings versus the best measured mixes; interpretable SCM-substitution design rules.
• Martirosyan, V. “High accuracy, no signal: an identifiability and external-validation audit of climate and nitrogen response in China's cereal yield data.”
  Computers and Electronics in Agriculture (submitted), 2026.

  5,316-row province panel for maize, rice and wheat in China (1961–2024) plus a 12,534-row 0.5° cropland panel; level R² 0.50–0.92 but anomaly R² mostly −0.09 to 0.08; external validation does not support temporal climate or nitrogen-response claims.
• Martirosyan, V. and N. Martirosyan “Interpretable polynomial surrogates for halide-perovskite screening: domain-cleaned bandgaps, ideal tandem limits, and the limits of forward prediction.”
  Solar Energy Materials and Solar Cells (submitted), 2026.

  31,618 domain-cleaned halide perovskites; coefficient-shrunk cubic SRPR reaches 21.0 meV MAE / R² 0.794 under grouped CV; a 101,871-point grid yields 34,014 candidates and a 362-point Pareto front; best ideal two-terminal tandem 43.73% (not a device claim).

• Martirosyan, V. “Interpretable Polynomial Zone Ensembles for Global Ocean Temperature and Salinity Profile Reconstruction.”
  Prepared for IEEE JSTARS, Oct 2026.
• Martirosyan, V. “hpc++: An LLVM-Based Automatic Parallelization System for C++ with GPU Acceleration.” PDF
  arXiv, Dec 2025.

  7 parallelisation strategies, zero source modification; peak 2009.4× on GPU and 32.1× on CPU; numerical correctness across 134 unit + 18 integration tests.
• Martirosyan, V. “Auditable Kilobyte-Scale Surrogates for the Global Significant-Wave-Height Climatology.” PDF
  Prepared for IEEE JSTARS, 2027.

  R² = 0.945, RMSE 0.262 m; cell-holdout R² = 0.933; coastal R² improves from 0.34 to 0.52 with in-situ buoys.
• Martirosyan, V. “Polynomial Regression Models for Interpretable Machine Learning: Theory, Optimization, and Medical Applications.”
  Prepared manuscript, Feb 2026.
• Martirosyan, V. “Stagewise Regularized Polynomial Regression: A Hybrid Optimization Strategy for Medical Diagnosis on Small Datasets.” PDF
  Prepared manuscript, Feb 2026.

  86.7% accuracy on UCI Cleveland Heart Disease; 93% model compression; ≈36 FLOPs/prediction.
• Martirosyan, V. “Heart Disease Diagnosis Using Third-Order Polynomial Regression Models with L2 Regularization.” PDF
  Prepared manuscript, Feb 2026.

  81.8% accuracy, 10-fold CV; full interpretability via explicit polynomial coefficients.
• Martirosyan, V. “Explainable and Efficient Digit Classification via Polynomial Regression Networks: A Transparent Alternative to Deep Learning.” PDF
  arXiv, Jan 2026.

  96.85% average MNIST accuracy; 84% parameter reduction at <1% accuracy loss.
• Martirosyan, V. “Blended Ocean Visualization: Combining Sparse Polynomial Regression with Residual Fields.” PDF
  Working paper, 2026.
• Martirosyan, V. “Portability and Limits of an Interpretable Polynomial Framework: A Global Significant-Wave-Height Case Study.” PDF
  Working paper, 2026.

  Anomaly R² = 0.875; offshore buoy RMSE ≈0.5 m; negative result reported for a distance-to-coast feature.
• Martirosyan, V. “Deep-Ocean Temperature Profile Reconstruction via Interpretable Polynomial Zone Ensembles.”
  Working paper, 2026.

  Stagewise-Rank-Polynomial-Regression; 5,005-term basis; 2.6 µs per-profile inference in C++; edge-deployment analysis.
• Martirosyan, V. “Global Ocean Salinity Profile Reconstruction with Interpretable Polynomial Zone Ensembles.”
  Working paper, 2026.

  WOA dataset; degree-8 expansion, 2,001 coefficients per zone; 23 µs inference; river-mouth salinity correction; residual ensemble.

• Jinnetic Engineering by Richard Stallman, translated into ArmenianLink
  Dec 2010.