arXiv subjects

Ali ArjomandBigdeli

Publications and source records attributed to Ali ArjomandBigdeli.

3 recordsLinked to original sources

arXiv2026

Large Language Models as Falsifiers for Cyber-Physical Systems

Falsification searches for counterexamples to formal specifications in cyber-physical systems (CPS). With specifications written in Signal Temporal Logic (STL), falsification can be formulated as a robustness optimization problem, traditionally tackled with black-box search algorithms. In parallel, large language models (LLMs) have recently emerged as surprisingly effective optimizers when coupled with iterative prompting. In this work, we connect these ideas and introduce LLM-Falsifier, an LLM-based approach that falsifies specifications by minimizing the STL robustness degree. Beyond generic prompt-based optimization, our key idea is to expose the LLM to semantic information that is natural for language models but absent from standard numerical optimizers, including natural-language input and output names, output trajectories, and critical-time witnesses for the minimum robustness value. These additions enable smarter and more sample-efficient robustness search. On the ARCH-COMP falsification benchmarks, LLM-Falsifier is shown to outperform existing falsification tools based on a range of optimization paradigms, from surrogate-based and Bayesian optimization to search-based testing, on 14 of 21 specifications when measured by the average number of simulations required to find a counterexample.

eess.SY↗

arXiv2025

Activation-Informed Merging of Large Language Models

Model merging, a method that combines the parameters and embeddings of multiple fine-tuned large language models (LLMs), offers a promising approach to enhance model performance across various tasks while maintaining computational efficiency. This paper introduces Activation-Informed Merging (AIM), a technique that integrates the information from the activation space of LLMs into the merging process to improve performance and robustness. AIM is designed as a flexible, complementary solution that is applicable to any existing merging method. It aims to preserve critical weights from the base model, drawing on principles from continual learning (CL) and model compression. Utilizing a task-agnostic calibration set, AIM selectively prioritizes essential weights during merging. We empirically demonstrate that AIM significantly enhances the performance of merged models across multiple benchmarks. Our findings suggest that considering the activation-space information can provide substantial advancements in the model merging strategies for LLMs, with up to a 40% increase in benchmark performance.

cs.CL↗

arXiv2024

Verification of Neural Network Control Systems in Continuous Time

Neural network controllers are currently being proposed for use in many safety-critical tasks. Most analysis methods for neural network control systems assume a fixed control period. In control theory, higher frequency usually improves performance. However, for current analysis methods, increasing the frequency complicates verification. In the limit, when actuation is performed continuously, no existing neural network control systems verification methods are able to analyze the system. In this work, we develop the first verification method for continuously-actuated neural network control systems. We accomplish this by adding a level of abstraction to model the neural network controller. The abstraction is a piecewise linear model with added noise to account for local linearization error. The soundness of the abstraction can be checked using open-loop neural network verification tools, although we demonstrate bottlenecks in existing tools when handling the required specifications. We demonstrate the approach's efficacy by applying it to a vision-based autonomous airplane taxiing system and compare with a fixed frequency analysis baseline.

eess.SY↗