arXiv subjects

Abed Hammoud

Publications and source records attributed to Abed Hammoud.

4 recordsLinked to original sources

arXiv2026

From Heat Stress to Perception: Interpretable Data-Driven Models of Human Thermal Sensation

Heat stress indices are designed to quantify physiological thermal stress, but their relevance for inferring the thermal perception of individuals remains unclear. In this study, we show that thermal stress and thermal sensation often diverge, as evidenced by distinct global sensitivity patterns with respect to environmental drivers. Using thermal sensation vote survey data, we demonstrate that the dominant sensitivities of stress-based metrics do not align with those governing reported human thermal sensation. Given the multitude of globally-applicable thermal stress indices and the lack of comparable general thermal sensation metrics, we develop two complementary data-driven modeling frameworks for thermal sensation. First, we construct polynomial chaos expansion (PCE) surrogates to represent thermal sensation as a function of meteorological variables, enabling efficient variance-based sensitivity analysis and explicit identification of influential inputs and interactions. Second, we develop multilayer perceptron (MLP) classifiers that capture the nonlinear and subjective nature of thermal perception, while achieving high predictive accuracy. The PCE models provide physically interpretable sensitivities that can explain the drivers of thermal sensation, while the MLPs offer flexible predictive capability suited to complex environments. We apply both modeling approaches at city- and continent-scales, revealing systematic differences in sensitivity structure and performance across climates. In particular, we find that the sensitivity of TSV-based models to the variability of meteorological conditions across geoclimatic zone encodes distinct dependencies on temperature, radiation, humidity, and wind that vary geographically, and are generally different from those of heat stress indices.

physics.ao-ph↗

arXiv2026

Data-Driven Flux Parameterization for the Atmospheric Boundary Layer

Turbulent fluxes in the atmospheric boundary layer (ABL) govern exchanges of momentum, heat, and mass between the surface and atmosphere, shaping boundary layer structure and influencing weather, climate, and engineering applications. Yet their representation in coarse resolution models remains challenging, particularly under unstable conditions with strongly nonlocal transport and stable conditions with intermittent turbulence. Here, we develop a data driven turbulent flux parameterization in which nondimensional fluxes are represented by a linearized convolution operator acting on nondimensional mean state profiles. We train and evaluate the closure using high resolution large eddy simulations (LES) of idealized flow over homogeneous surfaces spanning multiple stability regimes. Several first order closure variants are constructed from different combinations of mean temperature and velocity profiles to predict heat and momentum fluxes, and the best model is selected by minimizing mean squared error across training and unseen test cases. The resulting parameterization improves predictive skill relative to a standard K-profile closure while retaining an interpretable operator form. Its learned kernels expose the locality and nonlocality of turbulent transport across stability regimes, linking empirical performance to physically inspectable flux--profile relationships. In a posteriori single column simulations, the closure remains stable and produces state profiles that closely match LES, demonstrating its potential as an accurate and transparent ABL flux parameterization.

physics.flu-dyn↗

arXiv2026

Reinforcement Learning, Optimal Control, and Bayesian Filtering in Data Assimilation

We give a finite-horizon variational formulation that places Bayesian filtering and smoothing, variational data assimilation, KL-regularized control, and Kalman-type methods inside one mathematically explicit hierarchy. For a discrete-time hidden Markov model and any admissible one-step candidate law $q_t$, We prove $J_t(q_t)=\mathbb{E}_{q_t}\!\left[-\log p(y_t\mid X_t)\right] +\mathrm{KL}\!\left(q_t\|p_t^f\right) =\mathrm{KL}\!\left(q_t\|p_t^a\right)-\log p(y_t\mid y_{0:t-1})$, and, for any admissible path law $q$, $J_{\mathrm{path}}(q)=\mathbb{E}_{q}\!\left[-\sum_{t=0}^{T}\log p(y_t\mid X_t)\right] +\mathrm{KL}\!\left(q\|p(x_{0:T})\right) =\mathrm{KL}\!\left(q\|p(x_{0:T}\mid y_{0:T})\right)-\log p(y_{0:T})$. These identities determine the evidence as the global infimum and make the analysis and smoothing posteriors the unique minimizers whenever those posterior laws belong to the admissible classes. This separates targets that are often conflated: strong- and weak-constraint 4D-Var are MAP estimators under the stated Gaussian assumptions; KL-regularized control recovers the Bayesian posterior only when the passive dynamics, likelihood cost, temperature, and a restrictive representability condition on the policy class are all matched correctly; and the linear-Gaussian specialization yields the Kalman analysis exactly. The ensemble Kalman filter then appears as a Gaussian and finite-ensemble approximation to the forecast-to-analysis map, exact only in the linear-Gaussian infinite-ensemble limit. This framework also clarifies RMSE-based RL data assimilation: such rewards may define effective estimators or pseudo-posteriors, but not exact posterior recovery unless they realize the likelihood-plus-KL objective.

math.DS↗

arXiv2025

Train Long, Think Short: Curriculum Learning for Efficient Reasoning

Recent work on enhancing the reasoning abilities of large language models (LLMs) has introduced explicit length control as a means of constraining computational cost while preserving accuracy. However, existing approaches rely on fixed-length training budgets, which do not take advantage of the natural progression from exploration to compression during learning. In this work, we propose a curriculum learning strategy for length-controlled reasoning using Group Relative Policy Optimization (GRPO). Our method starts with generous token budgets and gradually tightens them over training, encouraging models to first discover effective solution strategies and then distill them into more concise reasoning traces. We augment GRPO with a reward function that balances three signals: task correctness (via verifier feedback), length efficiency, and formatting adherence (via structural tags). Experiments on GSM8K, MATH500, SVAMP, College Math, and GSM+ demonstrate that curriculum-based training consistently outperforms fixed-budget baselines at the same final budget, achieving higher accuracy and significantly improved token efficiency. We further ablate the impact of reward weighting and decay schedule design, showing that progressive constraint serves as a powerful inductive bias for training efficient reasoning models. Our code and checkpoints are released at: https://github.com/hammoudhasan/curriculum_grpo.

cs.CL↗