arXiv subjects

Mohamed Hendy

Publications and source records attributed to Mohamed Hendy.

5 recordsLinked to original sources

arXiv2026

Flattery in Motion: Benchmarking and Analyzing Sycophancy in Video-LLMs

As video large language models (Video-LLMs) become increasingly integrated into real-world applications that demand grounded multimodal reasoning, ensuring their factual consistency and reliability is of critical importance. However, sycophancy, the tendency of these models to align with user input even when it contradicts the visual evidence, undermines their trustworthiness in such contexts. Current sycophancy research has largely overlooked its specific manifestations in the videolanguage domain, resulting in a notable absence of systematic benchmarks and targeted evaluations to understand how Video-LLMs respond under misleading user input. To fill this gap, we propose VISE(Video-LLM Sycophancy Benchmarking and Evaluation), the first benchmark designed to evaluate sycophantic behavior in state-of-the-art Video-LLMs across diverse question formats, prompt biases, and visual reasoning tasks. Specifically, VISEpioneeringly brings linguistic perspectives on sycophancy into the video domain, enabling fine-grained analysis across multiple sycophancy types and interaction patterns. Furthermore, we propose two potential training-free mitigation strategies revealing potential paths for reducing sycophantic bias: (i) enhancing visual grounding through interpretable key-frame selection and (ii) steering model behavior away from sycophancy via targeted, inference-time intervention on its internal neural representations. Our code is available at https://anonymous.4open.science/r/VideoSycophancy-567F.

cs.CL↗

arXiv2024

GAP-DFT: A graph-based alchemical perturbation density functional theory for catalytic high-entropy alloys

High-entropy alloys (HEAs) exhibit exceptional catalytic performance due to their complex surface structures. However, the vast number of active binding sites in HEAs, as opposed to conventional alloys, presents a significant computational challenge in catalytic applications. To tackle this challenge, robust methods must be developed to efficiently explore the configurational space of HEA catalysts. Here, we introduce a novel approach that combines alchemical perturbation density functional theory (APDFT) with a graph-based correction scheme to explore the binding energy landscape HEAs. Our results demonstrate that APDFT can accurately predict binding energies for isoelectronic permutations in HEAs at minimal computational cost, significantly accelerating configurational space sampling. However, APDFT errors increase substantially when permutations occur near binding sites. To address this issue, we developed a graph-based Gaussian process regression model to correct discrepancies between APDFT and conventional density functional theory values. Our approach enables the prediction of binding energies for hundreds of thousands of configurations with a mean average error of 30 meV, requiring a handful of ab initio simulations.

cond-mat.mtrl-sci↗

arXiv2022

Electronic effects on the radiation damage in high-entropy alloys

High-entropy alloys (HEAs) are exceptional candidates for radiation-resistant materials due to their complex local chemical environment and slow defect migration. Despite commonly overlooked, electronic effects on defects evolution in radiation environments also play a crucial role by dissipating excess energy through electron-phonon coupling and electronic heat conduction during cascade events. We present a systematic study on electronic properties in random-solid solutions (RSS) in four and five principal elements HEAs and their effect on defect formation, clustering, and recombination. Electronic properties, including electron-phonon coupling factor, the electronic specific heat, and the electronic thermal conductivity, are computed within first-principles calculations. Using the two-temperature molecular dynamics simulations, we show that the electron-phonon coupling factor and electronic specific heat play a critical role in Frenkel pairs formation. Specifically, the electron-phonon coupling factor quickly dissipates the kinetic energy during primary knock-on atom events via plasmon excitations and is subsequently dissipated via the free-electrons conduction. We show that these effects are more critical than the elastic distortion effects produced by the atomic mismatch. Of tremendous interest, we show that including lighter elements helps to increase the electron-phonon coupling factor, suggesting the possibility to improve radiation resistance in HEA through optimal composition.

cond-mat.mtrl-sci↗

arXiv2022

A multiscale and multiphysics framework to simulate radiation damage in nano-crystalline materials

This work presents a multiscale and multiphysics framework to investigate the radiation-induced damage in nano-crystalline materials. The framework combines two methodologies, including molecular dynamics simulations with electronic effects and long-term atomistic diffusion simulations in nano-crystalline materials. Using this framework, we investigated nano-crystalline materials' self-healing behavior under radiation events. We found that the number of defects generated in nano-crystals during the cascade simulations was less than in single crystals. This behavior was due to the fast absorption of interstitial atoms in the grain boundary network during the cascade simulations, while vacancies migrated to the boundaries in a much longer time scale than interstitial atoms. Thus, nano-crystalline materials showed a self-healing behavior where the number and size of the defects are drastically reduced with time. We found that the self-healing behavior of nano-crystalline materials is limited, and about 50% of vacancies survived. This effect resulted from clusters of vacancies' collective behavior, which are much more stable than individual vacancies.

cond-mat.mes-hall↗

arXiv2022

Effects of the local chemical environment on vacancy diffusion in multi-principal element alloys

Multi-principal element alloys (MPEAs) are exciting systems showing remarkable properties compared to conventional materials due to their exceedingly large compositional space and spatially varying chemical environment. However, predicting fundamental properties from the local chemical environment is challenging due to the large scale of the problem. To investigate this fundamental problem, we employ a combination of atomistic simulations (using ab-initio and molecular dynamics) and convolutional neural networks (CNNs) to evaluate point defect and migration energies in an equimolar CoFeCrNi MPEA. We show how energies of point defects can be predicted with reasonable accuracy using a small subset of local chemical environments. Using the CNNs, we develop a lattice Monte Carlo simulation that computes the migration path and diffusivities of vacancies. Remarkably, our work illustrates how the local chemical environment leads rise to a distribution function of the point defect energies, which is responsible for the macroscopic diffusivity of vacancies. In particular, we observed that vacancies get trapped in super basins surrounded by large migration and connected with low migration energy states. As a result, vacancy diffusivity is highly dependent on the environment and could change several orders of magnitude for a given temperature. Our works illustrate the importance of understanding properties in MPEAs depending on the local chemical environment and the ability of CNN to provide a model for computing energies in high-dimensional spaces, which can be used to scale things up to higher-order models.

cond-mat.mtrl-sci↗