Abstract

Large Language Models (LLMs) have demonstrated strong capabilities as autonomous agents through tool use, planning, and decision-making abilities, leading to their widespread adoption across diverse tasks. As task complexity grows, multi-agent LLM systems are increasingly used to collaboratively solve problems. However, safety and security of these multi-agent systems remains largely unexplored. Existing benchmarks and datasets predominantly focus on single-agent settings, failing to capture the unique vulnerabilities of multi-agent dynamics and co-ordination. To address this gap, we introduce textbf{T}hreats and textbf{A}ttacks in textbf{M}ulti-textbf{A}gent textbf{S}ystems (textbf{TAMAS}), a dataset designed to evaluate the robustness and security of multi-agent LLM systems. TAMAS includes five distinct scenarios comprising 250 adversarial instances across five attack types and 163 different normal and attack tools, along with 100 harmless tasks. We assess system performance across 5 backbone LLMs and 3 agent interaction configurations from Autogen framework, highlighting critical challenges and failure modes in current multi-agent deployments. Our findings show that multi-agent systems are highly vulnerable to adversarial attacks, with Impersonation reaching a 73% success rate and other attacks ranging from 27% to 67%, underscoring the need for stronger defenses.

Abstract

Generation Alpha (born 2010-2024) is the first generation fully raised within the digital ecosystem. They exhibit unique linguistic behaviours influenced by rampant online communication and platform-specific cultures. This study examines the rapid evolution of Gen Alpha slang through a comparative analysis of Millennial and Gen Z vernacular. We identify three core linguistic patterns: extreme lexical compression, digital culture-driven semantic shifts and part-of-speech conversion. We construct a comprehensive slang corpus sourced from online platforms and evaluate the performance of four AI translation systems (viz. Google Translate, ChatGPT 4, Gemini 1.0, DeepSeek v3) on over 100 slang terms. Our results reveal significant translation challenges rooted in culturally-bound terms from gaming, meme culture, and mental health discourse. Most errors are the result of inadequate cultural contextualization, with literal translations dominating the error patterns. Our findings highlight the critical limitations in current language models and emphasize the need for adaptive, culturally sensitive and context-aware frameworks that can handle the dynamic lexicon of evolving youth vernacular.

Abstract

Identification of hallucination spans in black-box language model generated text is essential for applications in the real world. A recent attempt at this direction is SemEval-2025 Task 3, Mu-SHROOM-a Multilingual Shared Task on Hallucinations and Related Observable Over-generation Errors. In this work, we present our solution to this problem, which capitalizes on the variability of stochastically-sampled responses in order to identify hallucinated spans. Our hypothesis is that if a language model is certain of a fact, its sampled responses will be uniform, while hallucinated facts will yield different and conflicting results. We measure this divergence through entropy-based analysis, allowing for accurate identification of hallucinated segments. Our method is not dependent on additional training and hence is cost-effective and adaptable. In addition, we conduct extensive hyperparameter tuning and perform error analysis, giving us crucial insights into model behavior.

Abstract

Algorithms are a fundamental part of computer science education, with expressing and tracing forming key aspects of their study. Both pseudocode and code are used to express algorithms. Pseudocode helps abstract away specifics of the programming language when specifying an algorithm, but it can introduce ambiguities due to its informal nature and does not support a structured way to trace an algorithm. On the other hand, tracing a program provides a precise description but can conflate algorithmic logic with language-specific details, obscuring the inherent structure of an algorithm. In this experience report, we propose employing the Mapcode framework for expressing and tracing algorithms. Mapcode represents an algorithm as a set of transformations between well-defined spaces, naturally producing execution traces that are independent of programming constructs. Unlike traditional notional machines, which model the execution of programs, Mapcode machines model the high-level execution of algorithms. Mapcode traces reflect underlying logic and are independent of programming-language-specific data and control structures like while loops. We report preliminary findings from two classroom activities which explored the application of Mapcode to express and trace algorithms. In the first activity, third-year undergraduate students drew Mapcode Execution Diagrams to trace simple algorithms already specified in the Mapcode framework. We found that a majority of students were successful in constructing the diagrams. In the second activity, students in a Principles of Programming Languages course successfully defined algorithms from various paradigms (greedy, dynamic programming, sorting, etc.) as Mapcode machines and implemented them in Python and Racket.

Abstract

Effective management and optimization of urban infrastructures necessitates scalable and accessible simulation frameworks. Modern BIM-based solutions present a promising avenue for novel construction projects; however, these solutions are often inapplicable to the extensive array of legacy infrastructures developed prior to the establishment of BIM as a construction standard. Commensurate with this, we present a new modeldriven simulation approach for urban infrastructures that (i) utilizes a novel domain-specific modeling language (DSML) to represent both structural and behavioral characteristics of these infrastructures and (ii) employs the Discrete Event System Specification (DEVS) formalism for simulation purposes. Reframing urban infrastructures as IoT-based, event-driven systems facilitates efficient, hierarchical simulations of complex dynamic environments, including resource management and water networks. Simulation artifacts produced from the DSML through model-to-text generation are executed within the DEVS simulation framework. We validate our approach through a case study conducted at IIIT Hyderabad's Smart City Living Lab, illustrating its capacity to identify optimization opportunities within urban infrastructures.

Abstract

The ability of students to read, understand and explain code is tightly interlinked to their ability to trace. Multiple studies have shown that tracing helps students improve their comprehension of code. Analysis of the traces also provides insight into student misconceptions about program execution. We are interested in exploring pen-and-paper activities related to tracing control-flow. While the literature presents several tracing methodologies, existing pen-and-paper activities prioritise tracking the values of variables as they change throughout the program; control-flow is not adequately addressed. To fill this gap, we propose a comprehensive suite of class activities around control-flow of programs and report our experience deploying it in a first-year university programming course. In these activities, students construct six hand-drawn artefacts representing the static and dynamic aspects of a program's control-flow. We use a simplified subset of Python that includes sequential, conditional, iterative and (first-order) procedural control. We chose this subset to match the syllabus of the introduction to computer programming course of our university. We report the results of a preliminary study conducted to explore the feasibility of constructing the artefacts. Most students correctly understood the structure of the artefacts. We also report what these artefacts revealed about students' misconceptions related to control-flow in programs. Students had difficulty identifying the destination of a procedure call and its return. A significant number of their misconceptions were related to transfer of control when encountering errors.

Abstract

In the circuit model of quantum computing, amplitude amplification techniques can be used to find solutions to NP-hard problems defined on n-bits in time poly(n)2^{n/2}. In this work, we investigate whether such general statements can be made for adiabatic quantum optimization, as provable results regarding its performance are mostly unknown. Although a lower bound of Ω(2^{n/2})has existed in such a setting for over a decade, a purely adiabatic algorithm with this running time has been absent. We show that adiabatic quantum optimization using an unstructured search approach results in a running time that matches this lower bound (up to a polylogarithmic factor) for a broad class of classical local spin Hamiltonians. For this, it is necessary to bound the spectral gap throughout the adiabatic evolution and compute beforehand the position of the avoided crossing with sufficient precision so as to adapt the adiabatic schedule accordingly. However, we show that the position of the avoided crossing is approximately given by a quantity that depends on the degeneracies and inverse gaps of the problem Hamiltonian and is NP-hard to compute even within a low additive precision. Furthermore, computing it exactly (or nearly exactly) is #P-hard. Our work indicates a possible limitation of adiabatic quantum optimization algorithms, leaving open the question of whether provable Grover-like speed-ups can be obtained for any optimization problem using this approach.

Abstract

Data science and artificial intelligence (DSAI) based methods can process massive amounts of data to extract useful knowledge and can be employed to build decision support systems in various domains. Like other domains, the legal systems in several countries are being digitized and there is a scope to build DSAI-based frameworks to improve the performance of legal systems. In the literature, research efforts are being made to investigate DSAI-based methods to improve justice delivery. The Indian legal system is currently experiencing a major problem with a substantial backlog of cases. This paper provides an overview of DSAI-based efforts in the legal domain related to India. We also listed potential research issues to be explored. We hope the issues will encourage further research to improve justice delivery performance in India and other countries.

Abstract

Dysarthric speech poses significant challenges in developing as- sistive technologies, primarily due to the limited availability of data. Recent advances in neural speech synthesis, especially zero-shot voice cloning, facilitate synthetic speech generation for data augmentation; however, they may introduce biases to- wards dysarthric speech. In this paper, we investigate the effec- tiveness of state-of-the-art F5-TTS in cloning dysarthric speech using TORGO dataset, focusing on intelligibility, speaker sim- ilarity, and prosody preservation. We also analyze potential biases using fairness metrics like Disparate Impact and Parity Difference to assess disparities across dysarthric severity lev- els. Results show that F5-TTS exhibits a strong bias toward speech intelligibility over speaker and prosody preservation in dysarthric speech synthesis. Insights from this study can help integrate fairness-aware dysarthric speech synthesis, fostering the advancement of more inclusive speech technologies.

Abstract

This paper presents a novel Transformer (XFMR) design intended for integrated millimeter-wave (mmWave) applications. The XFMR consists of an 8-shaped primary inductor and a dumbbell-shaped secondary inductor. This structure addresses key challenges in on-chip XFMR design by improving electromagnetic interference (EMI) suppression, reducing unwanted signal coupling, and optimizing magnetic field symmetry. Additionally, the geometry enables better layout efficiency and isolation, contributing to improved performance in tightly integrated environments. To demonstrate the practical advantages of the proposed XFMR, it is implemented in a low-voltage, low-phase-noise (PN) LC VCO designed for 23.8-27.2 GHz operation. Simulation results in TSMC 65~nm CMOS technology show a tuning range of 3.4~GHz, a PN lesser than −110 dBc/Hz at 1 MHz offset, and a figure-of-merit (FoM) greater than 186 dBc/Hz, at 0.6 V supply. These results validate the effectiveness of the XFMR in enhancing VCO performance, highlighting its potential for use in high-performance mmWave design.