An iterative optimization algorithm with MD simulations in the loop is developed and applied to optimize Lennard-Jones (LJ) parameters specific for liquid tri-n-butyl phosphate (TBP). The optimization loop uses non-dominated sorting genetic algorithms to obtain LJ parameters that reproduce key properties such as mass density, electric dipole moment, heat of vaporization, self-diffusion coefficient (SDC), and shear viscosity. Errors relative to experimentally measured properties lead to a multi-objective function optimization problem stated in terms of a Pareto-optimal set. A systematic application of the optimization algorithm to cases involving single- and multi-objective functions was carried out in this work, establishing a framework for atomistic TBP property predictions. We demonstrate the use of a neural network property model to amortize the high cost of MD simulations in the optimization loop and to allow for large populations and more generations to be used in the genetic algorithms. In our previous study of finding the best force field for TBP property predictions as judged by the aforementioned thermophysical properties, we found the Polarized AMBER-MNDO force field to be the best overall showing a 74% relative deviation from experimental values. However, in this study, we show optimized values of the LJ parameters that improve the overall deviation from experimental data to 23% when using the NN NSGA-III algorithm. Despite this large improvement, the accurate prediction of the transport properties, SDC and shear viscosity, remains difficult since improvements in one of them worsen the other, and vice versa. Finally, finding optimized LJ parameters that further minimize the error of all transport properties is still possible via additional computing power used with larger population of parameter vectors and more generations in the genetic algorithms employed here.
Statistical mechanics describes the behavior of the particles (atoms and molecules) that comprise solids, liquids, and gases. The dynamics of particle motion is encoded in kinetic equations (e.g., Liouville, Boltzmann, Fokker-Planck, and Langevin) that predict the hydrodynamic, magnetic, and dielectric properties of materials, the rates of chemical reactions, and the stability of plasmas. However, interest in multiscale, composite, heterogeneous, and fractal materials has stimulated the development of generalized, and in some cases, fractional calculus models in order to describe the emergence of anomalous transport of mass, momentum, and energy. Extending integer time and space derivatives to fractional order provides a concise way to incorporate memory and non-locality into the force, flow, and flux terms of each equation. This approach recasts the fundamental equations and complements the nonlinear and perturbation models commonly used to address complex dynamics and stochastic processes. In this review, we compare fractional calculus methods with other techniques, such as the assumption of time- or position-dependent diffusion coefficients, and highlight the connections between fractional-order operators, stochastic differential equations, and the asymptotic behavior of physical systems. We focus on fractional-calculus modifications of the Boltzmann, Fokker–Planck, and Langevin equations and distinguish time-fractional (memory) from space-fractional (spatial nonlocality) generalizations and their main stochastic-process mechanisms. The goal is to identify situations where mathematics and physics combine in models that not only provide better fits to data, but also animate the mathematics, allowing it to be tailored to new applications.
Inspired by the rapid development of Large Language Models (LLMs), LLM agents have evolved to perform complex tasks. LLM agents are now extensively applied across various domains, handling vast amounts of data to interact with humans and execute tasks. The widespread applications of LLM agents demonstrate their significant commercial value; however, they also expose security and privacy vulnerabilities. At the current stage, comprehensive research on the security and privacy of LLM agents is highly needed. This survey aims to provide a comprehensive overview of the newly emerged privacy and security issues faced by LLM agents. We begin by introducing the fundamental knowledge of LLM agents, followed by a categorization and analysis of the threats. We then discuss the impacts of these threats on humans, environment, and other agents. Subsequently, we review existing defensive strategies, and finally explore future trends. Additionally, the survey incorporates diverse case studies to facilitate a more accessible understanding. By highlighting these critical security and privacy issues, the survey seeks to stimulate future research towards enhancing the security and privacy of LLM agents, thereby increasing their reliability and trustworthiness in future applications.
While large language models (LLMs) like ChatGPT have shown impressive capabilities in Natural Language Processing (NLP) tasks, a systematic investigation of their potential in this field remains largely unexplored. This study aims to address this gap by exploring the following questions: (1) How are LLMs currently applied to NLP tasks in the literature? (2) Have traditional NLP tasks already been solved with LLMs? (3) What is the future of the LLMs for NLP? To answer these questions, we take the first step to provide a comprehensive overview of LLMs in NLP. Specifically, we first introduce a unified taxonomy including (1) parameter-frozen paradigm and (2) parameter-tuning paradigm to offer a unified perspective for understanding the current progress of LLMs in NLP. Furthermore, we summarize the new frontiers and the corresponding challenges, aiming to inspire further groundbreaking advancements. We hope this work offers valuable insights into the potential and limitations of LLMs, while also serving as a practical guide for building effective LLMs in NLP.
Most Large Language Model (LLM) agent memory systems rely on a small set of static, hand-designed operations for extracting memory. These fixed procedures hard-code human priors about what to store and how to revise memory, making them rigid under diverse interaction patterns and inefficient on long histories. To this end, we present MemSkill, which reframes these operations as learnable and evolvable memory skills, structured and reusable routines for extracting, consolidating, and pruning information from interaction traces. Inspired by the design philosophy of agent skills, MemSkill employs a controller that learns to select a small set of relevant skills, paired with an LLM-based executor that produces skill-guided memories. Beyond learning skill selection, MemSkill introduces a designer that periodically reviews hard cases where selected skills yield incorrect or incomplete memories, and evolves the skill set by proposing refinements and new skills. Together, MemSkill forms a closed-loop procedure that improves both the skill-selection policy and the skill set itself. Experiments on LoCoMo, LongMemEval, HotpotQA, and ALFWorld demonstrate that MemSkill improves task performance over strong baselines and generalizes well across settings. Further analyses shed light on how skills evolve, offering insights toward more adaptive, self-evolving memory management for LLM agents.
