arXiv preprints from January 1, 2026 through September 24, 2026 — 13:36:23 EST
Multi-agent systems based on large language models, particularly centralized architectures, have recently shown strong potential for complex and knowledge-intensive tasks. However, central agents often suffer from unstable long-horizon collaboration due to the lack of memory management, leading to context bloat, error accumulation,...
In the analysis of complex physical systems, the objective often extends beyond merely computing a numerical solution to capturing the precise crossover between different regimes and extracting parameters containing meaningful information. However, standard numerical solvers and conventional deep learning approaches, such as...
AI-driven penetration testing now executes thousands of actions per hour but still lacks the strategic intuition humans apply in competitive security. To build cybersecurity superintelligence --Cybersecurity AI exceeding best human capability-such strategic intuition must be embedded into agentic reasoning processes. We present...
Given a vertex in a $(k, \varphi, ε)$-clusterable graph, i.e. a graph whose vertex set can be partitioned into a disjoint union of $\varphi$-expanders of size $\approx n/k$ with outer conductance bounded by $ε$, can one quickly tell which cluster it belongs to? This question goes back to the expansion testing problem of Goldreich and...
Preference tuning aligns pretrained language models to human judgments of quality, helpfulness, or safety by optimizing over explicit preference signals rather than likelihood alone. Prior work has shown that preference-tuning degrades performance and reduces helpfulness when evaluated outside the training domain. However, the extent...
Access to justice remains limited for many people, leading laypersons to increasingly rely on Large Language Models (LLMs) for legal self-help. Laypeople use these tools intuitively, which may lead them to form expectations based on incomplete, incorrect, or biased outputs. This study examines whether leading LLMs exhibit gender bias...
Recent work shows that large multimodal models (LMMs) can self-improve from unlabeled data via self-play and intrinsic feedback. Yet existing self-evolving frameworks mainly reward final outcomes, leaving intermediate reasoning weakly constrained despite its importance for visually grounded decision making. We propose iReasoner, a...
Modelling a language model for a multi-lingual scenario includes several potential challenges, among which catastrophic forgetting is the major challenge. For example, small language models (SLM) built for low-resource languages by adapting large language models (LLMs) pose the challenge of catastrophic forgetting. This work proposes...
We study the joint minimization of communication and computation costs in distributed computing, where a master node coordinates $N$ workers to evaluate a function over a library of $n$ files. Assuming that the function is decomposed into an arbitrary subfunction set $\mathbf{X}$, with each subfunction depending on $d$ input files,...
Interviews are commonplace in HCI. This paper presents a documentary method of interpretation (DMI) for the analysis of the topical organisation of talk within a corpus of transcripts, topics that are endogenous to it and elaborate patterns of social reasoning about issues of relevance to research. The DMI reflexively enables...
Recent advances in Reinforcement Learning with Verifiable Rewards (RLVR) for Large Language Model (LLM) reasoning have been hindered by a persistent challenge: exploration collapse. The semantic homogeneity of random rollouts often traps models in narrow, over-optimized behaviors. While existing methods leverage policy entropy to...
Retrieval-Augmented Generation (RAG) models are critically undermined by citation hallucinations, a deceptive failure where a model cites a source that fails to support its claim. While existing work attributes hallucination to a simple over-reliance on parametric knowledge, we reframe this failure as an evolving, scale-dependent...
We introduce the first method for change-point detection on encrypted time series. Our approach employs the CKKS homomorphic encryption scheme to detect shifts in statistical properties (e.g., mean, variance, frequency) without ever decrypting the data. Unlike solutions based on differential privacy, which degrade accuracy through...
With the growth of interpreting technologies, from remote interpreting and Computer-Aided Interpreting to automated speech translation and interpreting avatars, there is now a high demand for ways to quickly and efficiently measure the quality of any interpreting delivered. A range of approaches to fulfil the need for quick and...
Community detection is a fundamental task in data analysis, and block models provide an approach for identifying a wide variety of community structures while offering high interpretability. The degree-corrected block model (DCBM) is an established model that accounts for the heterogeneity of node degrees. However, inference methods...
Large language models (LLMs) exhibit complementary strengths arising from differences in pretraining data, model architectures, and decoding behaviors. Inference-time ensembling provides a practical way to combine these capabilities without retraining. However, existing ensemble approaches suffer from fundamental limitations. Most...
Reinforcement learning (RL) has emerged as a critical technique for enhancing LLM-based deep search agents. However, existing approaches primarily rely on binary outcome rewards, which fail to capture the comprehensiveness and factuality of agents' reasoning process, and often lead to undesirable behaviors such as shortcut...
We present a generative model of human mobility in which trajectories arise as realizations of a prescribed, time-dependent Markov dynamics defined on a spatial interaction network. The model constructs a hierarchical routing structure with hubs, corridors, feeder paths, and metro links, and specifies transition matrices using...
Automated seizure detection from electroencephalography (EEG) remains difficult due to the large variability of seizure dynamics across patients, recording conditions, and clinical settings. We introduce LookAroundNet, a transformer-based seizure detector that uses a wider temporal window of EEG data to model seizure activity. The...
Direct access asks for the retrieval of query answers by their ranked position, given a query and a desired order. While the time complexity of data structures supporting such accesses has been studied in depth, and efficient algorithms for many queries and common orders are known, their practical performance has received little...
Recent advancements in Large Language Model (LLM) agents have enabled complex multi-turn agentic tasks requiring extensive tool calling, where conversations can span dozens of API calls with increasingly large context windows. However, although major LLM providers offer prompt caching to reduce cost and latency, its benefits for...
Large language models (LLMs) often fail to learn effective long chain-of-thought (Long CoT) reasoning from human or non-Long-CoT LLMs imitation. To understand this, we propose that effective and learnable Long CoT trajectories feature stable molecular-like structures in unified view, which are formed by three interaction types:...
How important is the weight of a given column in determining the ranking of tuples in a table? To address such an explanation question about a ranking function, we investigate the computation of SHAP scores for column weights, adopting a recent framework by Grohe et al.[ICDT'24]. The exact definition of this score depends on three key...
Cold-start cross-domain recommender (CDR) systems predict a user's preferences in a target domain using only their source-domain behavior, yet existing CDR models either map opaque embeddings or rely on post-hoc or LLM-generated rationales that are hard to audit. We introduce EviSnap a lightweight CDR framework whose predictions are...
In safety-critical domains, linguistic ambiguity can have severe consequences; a vague command like "Pass me the vial" in a surgical setting could lead to catastrophic errors. Yet, most embodied AI research overlooks this, assuming instructions are clear and focusing on execution rather than confirmation. To address this critical...