The rise of LLM-based agents has opened new frontiers in AI applications, yet evaluating these agents remains a complex and underdeveloped area.This survey provides an in-depth overview of the emerging field of LLM agent evaluation, introducing a twodimensional taxonomy that organizes existing work along (1) evaluation objectives-what to evaluate, such as agent behavior, capabilities, reliability, and safety-and (2) evaluation process-how to evaluate, including interaction modes, datasets and benchmarks, metric computation methods, and tooling.In addition to taxonomy, we highlight enterprise-specific challenges, such as role-based access to data, the need for reliability guarantees, dynamic and longhorizon interactions, and compliance, which are often overlooked in current research.We also identify the future research directions, including holistic, more realistic, and scalable evaluation.This work aims to bring clarity to the fragmented landscape of agent evaluation and provide a framework for systematic assessment, enabling researchers and practitioners to evaluate LLM agents for real-world deployment.

Abstract In this paper, an improved method to arrange the light emitting diodes (LED) on the ceiling for an indoor visible light communication (VLC) system is proposed. More precisely, a LiFi‐based connection is considered where several LEDs are placed on the ceiling of an office and communicate with several receiver (user) located randomly within the room area. It is proposed to find the optimal location of the LEDs by minimizing the outage probability at the user's location. Both cases of static and mobile users are addressed. First, the closed‐form expression for the outage probability at the user's location is derived. Then, by minimizing the average outage probability, the optimal location of the LEDs is found. The numerical results show that the proposed LED arrangement outperforms the classically used uniform LED arrangement in terms of average outage probability at the receiver and reduces the necessary signal‐to‐noise ratio (SNR) to achieve a target average outage probability by about 3 and 2.2 dB for mobile and static users, respectively. They also show that the performance superiority of the proposed method over classical arrangement remains valid even under large estimation errors.

Privacy is an important concern for our society where sharing data with partners or releasing data to the public is a frequent occurrence. Some of the techniques that are being used to achieve privacy are to remove identifiers, alter quasi-identifiers, and perturb values. Unfortunately, these approaches suffer from two limitations. First, it has been shown that private information can still be leaked if attackers possess some background knowledge or other information sources. Second, they do not take into account the adverse impact these methods will have on the utility of the released data. In this paper, we propose a method that meets both requirements. Our method, called table-GAN , uses generative adversarial networks (GANs) to synthesize fake tables that are statistically similar to the original table yet do not incur information leakage. We show that the machine learning models trained using our synthetic tables exhibit performance that is similar to that of models trained using the original table for unknown testing cases. We call this property model compatibility . We believe that anonymization/perturbation/synthesis methods without model compatibility are of little value. We used four real-world datasets from four different domains for our experiments and conducted indepth comparisons with state-of-the-art anonymization, perturbation, and generation techniques. Throughout our experiments, only our method consistently shows balance between privacy level and model compatibility.

The best practice to prevent Cross Site Scripting (XSS) attacks is to apply encoders to sanitize untrusted data. To balance security and functionality, encoders should be applied to match the web page context, such as HTML body, JavaScript, and style sheets. A common programming error is the use of a wrong encoder to sanitize untrusted data, leaving the application vulnerable. We present a security unit testing approach to detect XSS vulnerabilities caused by improper encoding of untrusted data. Unit tests for the XSS vulnerability are automatically constructed out of each web page and then evaluated by a unit test execution framework. A grammar-based attack generator is used to automatically generate test inputs. We evaluate our approach on a large open source medical records application, demonstrating that we can detect many 0-day XSS vulnerabilities with very low false positives, and that the grammar-based attack generator has better test coverage than industry best practices.

Input sanitization mechanisms are widely used to mitigate vulnerabilities to injection attacks such as cross-site scripting. Static analysis tools and techniques commonly used to ensure that applications utilize sanitization functions. Dynamic analysis must be to evaluate the correctness of sanitization functions. The proposed approach is based on unit testing to bring the advantages of both static and dynamic techniques to the development time. Our approach introduces a technique to automatically extract the sanitization functions and then evaluate their effectiveness against attacks using automatically generated attack vectors. The empirical results show that the proposed technique can detect security flaws cannot find by the static analysis tools.