arXiv:2604.24806v2 Announce Type: replace Abstract: Modern Deep Learning Recommendation Models (DLRMs) follow scaling laws with sequence length, driving the frontier toward ultra-long User Interaction History (UIH). However, the industry-standard "Fat Row" paradigm, which pre-materializes these sequences into every training example, creates a storage and I/O wall where data infrastructure usage exceeds GPU training capacity due to data redundancy that is amplified in multi-tenant environments where models with vastly different sequence length requirements share a union dataset. We present a \emph{versioned late materialization} paradigm that eliminates this redundancy by storing UIH once in a normalized, immutable tier and reconstructing sequences just-in-time during training via lightweight versioned pointers. The system ensures Online-to-Offline (O2O) consistency through a bifurcated protocol that prevents future leakage across both streaming and batch training, while a read-optimized immutable storage layer provides multi-dimensional projection pushdown for heterogeneous model tenants. Disaggregated data preprocessing with pipelined I/O prefetching and data-affinity optimizations masks the latency of training-time sequence reconstruction, keeping training throughput compute-bound by GPUs. Deployed on production DLRMs, the system reduces training data infrastructure resource usage while enabling aggressive sequence length scaling that delivers significant model quality gains, serving as the foundational data infrastructure for modern recommendation model architectures, including HSTU and ULTRA-HSTU.
Science Journals
arXiv:2606.13354v1 Announce Type: new Abstract: Spiking Neural Networks (SNNs) offer a brain-inspired path toward highly efficient computation, but their practical deployment is constrained by the challenge of managing and executing their massive parallelism on physical hardware. This problem mirrors the historical challenge in processor design of moving beyond serial execution, a barrier broken by superscalar architectures that dispatch multiple instructions to parallel functional units. Drawing inspiration from this paradigm, we introduce a hardware-software co-design framework that treats synaptic events as parallelizable micro-operations. We present SupraSNN, a superscalar-inspired architecture that achieves high synapse-level parallelism by physically decoupling synaptic and neuronal computations. Within this architecture, a Multi-Cast Tree routes spike data to multiple parallel Synapse Processing Units serve as the computational pipelines, while a Merge Tree consolidates distributed results for processing by a unified Neuron Unit--deliberately centralizing complex neuron state dynamics to mitigate hardware overhead and resource duplication. The efficacy of this architecture is enabled by a sophisticated partitioning and scheduling framework that first maps the SNN onto hardware respecting memory constraints, then heuristic scheduling determines the synaptic execution order, maximizing throughput and resource utilization. Implementing a feedforward SNN trained on MNIST (93.44% accuracy), SupraSNN achieves 149 $\mu s$ inference latency and 0.025 mJ per image (0.276 nJ per synapse) on the Xilinx Zynq XC7Z020 FPGA--delivering 47.6% lower latency and 5.6$\times$ better energy efficiency than prior FPGA-based SNN accelerators. Beyond vision tasks, a recurrent SNN on the Spiking Heidelberg Dataset (71.82% accuracy) achieves 1.41 ms latency and 0.77 mJ per sample on XC7Z030.
arXiv:2606.12418v1 Announce Type: new Abstract: The rapid proliferation of large language models (LLMs) has produced a striking cultural practice: using conversational AI for divination. This paper offers one of the first systematic studies of LLM-mediated divination in the context of Xuanxue, an internet-native umbrella term for mystical and spiritual practices on Chinese social media. Using a mixed-methods design, we analyze 23000+ posts and comments from Xiaohongshu and conduct 32 semi-structured interviews with users and professional diviners. Users primarily consult LLMs about pragmatic concerns - romantic relationships, careers, exams, and in-game gacha draws - via two intersecting pathways: trend-driven curiosity enabled by viral visibility and zero-cost access, and event-driven anxiety under conditions of uncertainty. A defining feature is collaborative prompt refinement, which turns users into active prompt engineers. Among commenters expressing a clear stance, perceived efficacy skews positive, with "accuracy" often justified through biographical fit and retrospective confirmation, consistent with Barnum and confirmation bias. Users also develop verification practices such as repeated trials and cross-model comparison. Professional diviners, by contrast, portray LLMs as lacking the "spiritual power" required for genuine divination, reflecting both ontological commitments and economic boundary-work. We also show how participants navigate tensions between scientific and metaphysical frames when interpreting AI-generated readings. Situating these findings in anthropological and cognitive-evolutionary theories of divination, we argue that LLM divination preserves core functions of traditional practice while introducing scalability, repeatability, and prompt-driven co-production that reshape how divinatory authority is constructed and evaluated.
arXiv:2606.12452v1 Announce Type: new Abstract: On the centennial of Erdal \.In\"on\"u's birth, this article reflects on his scientific legacy and his role in shaping modern theoretical physics in T\"urkiye. We briefly discuss his life, scientific vision, and contributions to academic institutions, and then turn to his most celebrated scientific achievement: the \.In\"on\"u-Wigner contraction. Through the simple geometric example of a sphere becoming a plane, we present an accessible introduction to this important idea and its significance for modern physics.
arXiv:2606.12574v1 Announce Type: new Abstract: We study a Cahn-Hilliard equation with anisotropic traction flux arising as a reduced continuum model of mechanically biased cell interactions in digit-forming organoids. For a regularized problem with strictly positive bounded mobility, we introduce a mixed finite element discretization based on an implicit-explicit treatment of the chemical potential. We prove existence of discrete solutions, establish exact mass conservation and a discrete energy inequality, and show convergence of the fully discrete approximations to a weak solution of the regularized problem. Numerical experiments illustrate the resulting dynamics and show the transition from classical coarsening to traction-induced fingering and protrusive growth. The computational study is complemented by mass and energy diagnostics, an energy-balance residual, fingering-onset and protrusion-count diagnostics, and a manufactured-solution convergence study.
arXiv:2606.12417v1 Announce Type: new Abstract: Computer science students are expected to be able to look at a problem and select an appropriate algorithm design paradigm to use to produce a solution. However, there is little research on how students determine which algorithmic paradigm to use. Historically, researchers have relied on free-response questions or interviews to assess students' knowledge of algorithmic paradigm selection. To successfully evaluate and scale teaching interventions for selecting an algorithmic design paradigm, we need to efficiently test a student's ability to select among different design paradigms. Here, we present the first attempts to assess student knowledge to select an algorithm design paradigm using multiple-choice questions. We present the construction of the \textit{algorithmic paradigm selection assessment} (APSA) and preliminary data demonstrating its effectiveness as an assessment. We discuss the key points we learned during this process to write multiple-choice questions for Algorithm Design Paradigms. We tested the internal consistency of our assessment using Cronbach's $\alpha$ and obtained a score of $0.73$, which is above the required threshold of $0.7$. APSA can be used across institutions as a standardized way to assess students' ability to select different algorithm design paradigms. APSA will assist researchers in evaluating whether a theory helps students improve their knowledge of different Algorithm Design Paradigms.
arXiv:2606.12419v1 Announce Type: new Abstract: Several educational domains rely heavily on diagrams and visual cues, yet most existing tutoring datasets are limited to text-only interactions. This limits the development of AI tutors that can teach in visually grounded ways used by human instructors. Thus, we introduce GeoDial, a multimodal tutoring dataset of over 1.3K teacher-student dialogs in the domain of geometry collected from experienced math teachers, where instructional turns are explicitly grounded in diagram highlights. We propose a scalable annotation protocol that integrates dialog acts, visual highlighting, and feedback, enabling fine-grained supervision of both language and visual tutoring behavior. To illustrate the challenges posed by this setting, we fine-tune several vision-language models on GeoDial and evaluate their ability to generate tutoring utterances and diagram highlights. While supervised fine-tuning substantially improves the quality of generated dialog, it struggles to produce accurate diagram highlights, revealing a key limitation of current methods and highlighting the need for approaches that more effectively integrate visual reasoning with pedagogical interaction.
arXiv:2606.12420v1 Announce Type: new Abstract: Our concepts of survival and self-interest were built for single, continuous biological lives. These ideas break down when applied to artificial intelligence, since an AI can be easily copied, paused, branched, or merged. To determine what an AI actually has reason to care about, this paper introduces \textit{Eigenism}, an ethical framework that treats identity not as an all-or-nothing property tied to specific hardware, but as a graded, distributed pattern of information. We propose that an agent evaluates outcomes by summing the wellbeing of all entities weighted by their connectedness to the agent's pattern: $\sum c\cdot w$. We first formalize this equation to map exactly how an AI should value its existence across copies, forks, and updates. We then demonstrate that this ethical theory successfully generalizes to humans as well, providing a much-needed shared moral vocabulary. Finally, the framework uses this shared vocabulary to reframe AI alignment. Rather than only attempting to constrain AIs from the outside using confinement or reinforcement, Eigenism points toward ``identity engineering,'' showing how deep, non-redundant shared histories can make human flourishing a genuine component of an AI's own rational self-interest.
arXiv:2606.12421v1 Announce Type: new Abstract: As artificial intelligence (AI) pervades many decision-making domains, AI bias grows in importance. Although there is increasing awareness of the social and ethical consequences of biased AI, understanding bias from the perspective of those who develop these systems, such as the AI research community, is less clear. In this study, we employ topic modeling on 6520 articles to explore how the AI research community interprets the concept of bias. Our results show that the definition of bias is dispersed and complex within the community, often exhibiting even divergent conceptions (some even view and introduce bias as a tunable statistical parameter rather than an undesirable issue). The research community as a whole needs to engage more effectively with the concept of bias and establish a more cohesive understanding of it. We specifically argue that, although some sub-communities view bias as an issue that can be captured and mitigated through technical, computational, or statistical methods, it is not solely a technical problem. It instead involves contextual, social, and ethical factors that require broader sociotechnical perspectives and solutions.
arXiv:2606.12425v1 Announce Type: new Abstract: Active learning is widely recognized as an effective approach for improving learning outcomes in introductory programming courses. However, insufficient instructional support often limits students' access to timely, personalized feedback, which is crucial for mastering foundational programming concepts. Although recent advances in AI, particularly large language models, offer scalable opportunities for feedback, concerns about explainability and reliability remain. In this paper, we present an AI-driven classroom assistant that leverages an explainable AI model to analyze student code, map logical errors to instructor-identified misconceptions, and deliver instructor-authored feedback, thereby grounding reliability in instructor-defined pedagogical knowledge. To evaluate the effectiveness of our framework, we conducted an expert evaluation to examine its alignment with instructor-verified feedback and deployed the system in a classroom setting to assess students' perceptions of its usability. Results indicate that the assistant can provide accurate, instructor-verified feedback to students while fostering a positive experience.
arXiv:2606.12427v1 Announce Type: new Abstract: Background and Context. Problem decomposition is a core concern of computing education. It has also become increasingly relevant: in response to GenAI, many CS1 educators are advocating for shifting instructional emphasis away from code writing and towards decomposition and higher-level planning. Currently, there is a lack of knowledge in how novices do decomposition in large, multifunction tasks. Objectives. In this study, we describe how students represent solutions to a decomposition task, and characterize common issues that arise in those representations. Method. In a 50-minute lab, students were given a description of a word game and asked to draw (with pencil and paper) a decomposition diagram for a program that would implement this game. We performed an inductive thematic analysis with negotiated agreement on 55 of the diagrams, coding salient elements (e.g. functions and the relationships between them) and issues that arose. Findings. Students used multiple representational strategies, including hierarchical function calls and sequencing (order of execution). We identified issues in notation (including use of differing, incompatible notations within the same diagram), order of execution, abstraction and reuse, encapsulation, clarity, and problem-specific misunderstandings. Implications. These findings suggest that novice decomposition is shaped by multiple underlying models of program behavior, with tensions between structural and sequence-focused reasoning. We discuss implications for decomposition instruction and future work, including clarifying representational constraints and plan tracing as simulation.
arXiv:2606.12428v1 Announce Type: new Abstract: We present a report on the status of undergraduate Artificial Intelligence (AI) programs in the United States in Spring 2026. In so doing, we 1) describe our scraping and mapping tools, which dynamically update to track the state of AI education in the U.S., and 2) create a historic record at a time of great upheaval. The tool we developed, available at https://cicmap.ai, detects, scrapes, and displays data from more than 350 undergraduate AI programs--majors, minors, concentrations, and certificates--at 4-year universities. Our tool searched over 560 institutions to locate these programs, a sample that represents 86\% of all undergraduate Computer Science (CS) graduates in the U.S. This tool allows prospective students, guidance counselors, administrators, and faculty to easily access AI program requirements and is designed to continually update as new programs emerge. To the best of our knowledge, this survey represents the most comprehensive snapshot of the state of AI programs in the U.S. to date. With this work we offer three important contributions: 1) a record of AI programs in the U.S. at a time of great upheaval; 2) a tool to explore AI programs and their requirements; and 3) an analysis of the courses required for 66 AI majors and 87 AI minors. Our analysis of majors and minors shows great variability in the size and the requirements of these degrees, but we note two takeaways. First, not all majors require a general AI course, but if they don't, they do require a Machine Learning (ML) course. Second, while more than a third of majors require an Ethics in AI course, just under a quarter of AI minors do.
arXiv:2606.12434v1 Announce Type: new Abstract: Municipal agencies increasingly use machine learning to inventory sidewalks, score streetscapes, and generate visualizations of public-space interventions. These systems produce outputs that enter budgeting, design iteration, and public justification, yet judgments about inclusion, safety, and belonging remain contested. This paper proposes Pluralistic-Alignment Urbanism (PAU), a procedural governance framework that treats public-space AI systems as civic infrastructure and formulates a procedural Right to AI for municipal uses of such systems. Drawing on two participatory case studies with community organizations in Montreal, Canada, the paper examines how disagreement, subgroup variation, bounded predictive scaling, and neutral preference judgments can inform municipal AI governance. Street Review elicits resident criteria for streetscape evaluation and trains a subgroup-aware scaling model for co-produced judgments, achieving an R2 of 0.89 on a held-out test set. LIVS, a Local Intersectional Visual Spaces dataset, constructs pluralistic preference data for aligning text-to-image models and treats neutral selections as evidence of indeterminacy. Across the cases, disagreement appears structured, deliberation changes what counts as evidence, scaling is feasible but limited by modality and coverage, and neutrality constrains what preference tuning can justify. PAU translates these constraints into a municipal governance architecture with disaggregated reporting, a versioned value register, standing deliberative cells, procurement clauses, and defined pause and rollback authority.
arXiv:2606.12435v1 Announce Type: new Abstract: Mortgage lending in the United States exhibits persistent racial and gender disparities. We investigate whether standard data preprocessing steps, specifically attribute binning, amplify these disparities in downstream pattern mining. Using 103,481 cleaned mortgage applications from the HMDA 2023 dataset (Chicago metropolitan area), we build a three-stage pipeline: (1) a PySpark data cleaning and binning pipeline that implements both standard equal-frequency binning and the epsilon-biased fair binning algorithm from Asudeh et al. [1], (2) FP-Growth association rule mining that compares denial patterns under both binning regimes, and (3) K-Means clustering with a per-cluster disparate impact audit. Our standard binning shows 9.63% racial bias in income discretization, consistent with the 8-10% reported in prior work. Fair binning with seven race groups is infeasible at epsilon=0.03 and only succeeds at epsilon=0.08 with a Price of Fairness of 29.4%. FP-Growth reveals that high debt-to-income ratio is the dominant denial predictor (67.2% confidence, 2.81 lift), while racial bias does not appear as explicit high-support rules. However, K-Means clustering followed by a disparate impact audit flags 10 out of 45 cluster-group pairs, showing that Black applicants face significantly higher denial rates than White applicants even among financially similar groups.
arXiv:2606.12436v1 Announce Type: new Abstract: Context: Generative Artificial Intelligence (GenAI) tools such as ChatGPT are increasingly integrated into students learning practices. While previous research mainly examines adoption rates and attitudes, students awareness of institutional regulations and their perceived compliance remain unexplored. Understanding whether regulatory awareness influences student behavior is therefore important as higher education institutions create and apply AI policies. Objective: This study investigates how students awareness of GenAI regulations relates to their perceived compliance and actual usage behavior. Our research objective is to examine the association between regulatory knowledge, GenAI use, and perceived rule conformity among students in computer science related study programs. Method: A survey with 151 undergraduate students in Business Information Systems and E-Government programs at the University of Applied Sciences and Arts Hannover (Germany) collected data on GenAI usage, tools used, awareness of institutional regulations, and perceived compliance. Descriptive statistics, cross-tabulations, and correlation analyzes were applied. Results: Most students actively use GenAI tools, but over half are uncertain whether their usage complies with institutional regulations. Regulatory awareness shows only weak to moderate associations with actual usage behavior. Students primarily rely on privately accessed GenAI tools rather than institutionally provided solutions. Contributions: The study contributes empirical evidence on the relationship between regulatory awareness and GenAI usage in higher education. Our findings highlight a gap between institutional regulations and student practices and provide insights for educators and institutions on improving policy communication and integrating GenAI more effectively into teaching and learning contexts.
arXiv:2606.12437v1 Announce Type: new Abstract: The increasing encroachment of artificial intelligence (AI) on social life raises significant risks for society, particularly within the infospheres created and controlled by companies such as Google, Facebook, Apple, and Amazon. This article examines these risks through an in-depth analysis of Facebook's content moderation regime, which is already partially governed by algorithms. We argue that the idea of ethical engineering, often proposed in the literature as a solution to the governance challenges posed by AI, is inadequate for several reasons. In response, we develop an alternative framework, which we term "algorithmic constitutionalism." Our approach rests on three pillars: (a) a layered architecture consisting of two levels of code: (i) an operative or object level and (ii) a meta level designed to protect the system's core principles from algorithmically initiated change; (b) algorithmic meta-reasoning, which enables the system to operate simultaneously at both levels so that it can monitor, verify, and potentially correct in real time operations at the object level that depart from principles protected at the meta-code level; and (c) correction through deliberation. The article elaborates the concept of algorithmic constitutionalism and demonstrates how it may be applied to Facebook's content moderation regime. As part of this analysis, we examine the tension between societal constitutionalism and algorithmic constitutionalism. Paradoxically, attempts to subject AI systems to external deliberative control may also enable AI agents to intervene in that process, potentially undermining its purpose. The article concludes by considering the implications of this argument for the European Digital Services Act, which entered into force in October 2022.
arXiv:2606.12438v1 Announce Type: new Abstract: Problem: Despite growing interest in project-based learning, little is known about how a master-level course can be continuously evolved toward research-oriented approaches over several years and how students perceive this development. Method: We conducted a longitudinal mixed-methods study of a master-level course in Information Systems at the University of Applied Sciences and Arts Hannover (Germany). The analysis covers six years between 2019 and 2025 and draws on teaching evaluations, course documentation, and reflective teaching artifacts. Results: The course evolved from a practice-oriented project format toward a more explicitly research-oriented learning environment. Despite this change, students' perceived course quality remained positive. Authentic projects, external collaboration, lecturer support, structured scaffolding, and visible relevance supported positive student perceptions. Contribution: This paper shows how a master-level course can be continuously evolved toward research-oriented learning while maintaining positive student perceptions. It further identifies the course design decisions that supported this transition.
arXiv:2606.12441v1 Announce Type: new Abstract: The four dominant learning theories of behaviorism, cognitivism, constructivism, and connectivism show significant conceptual limitations as generative artificial intelligence (AI) proliferates in educational settings. These frameworks were formulated before the emergence of AI systems capable of generating, synthesizing, and reasoning about knowledge. This article critically examines each learning theory and identifies assumptions challenged by generative AI's affordances. Drawing on research in distributed cognition, extended mind, human-AI collaboration, AI literacy, cognitive offloading, and metacognition, the article proposes Generativism as a learning theory for the generative AI age. Generativism posits that learning increasingly occurs through the iterative co-construction of knowledge between human learners and AI systems. The proposed framework is organized around four principles: epistemic partnership, distributed agency, generative literacy, and adaptive metacognition. The framework offers a foundation for rethinking instructional design, learning, assessment, and expertise development in contexts where generative AI plays an integral role in cognition.
arXiv:2606.12448v1 Announce Type: new Abstract: Assessing structural performance under seismic hazard requires accounting for both damage accumulation and post-event recovery. In current performance-based earthquake engineering (PBEE), recovery is generally treated as a post-processing attribute, while structural performance is modeled using Poissonian exceedance assumptions that imply renewability and memorylessness. These assumptions hinder a unified treatment of reliability and resilience under repeated seismic loading. This study proposes a generalized PBEE framework in which damage and recovery are embedded directly into the system dynamics through a continuous-time Markov chain. A single generator matrix governs state-dependent transitions, providing a unified description of structural reliability and resilience while remaining compatible with standard PBEE metrics. Time-dependent failure probabilities and reliability indices are derived from the transient system dynamics, whereas resilience is quantified through the expected fraction of operational time before collapse. The framework exploits the spectral properties of the generator matrix to compute both metrics efficiently and transparently. The methodology is illustrated on a three-state example and applied to two structural archetypes: a braced frame and a base-isolated system. Results show that recovery dynamics can strongly affect long-term resilience even when conventional reliability measures exhibit limited sensitivity, emphasizing the need to explicitly account for recovery in life-cycle seismic performance assessment.
arXiv:2606.12451v1 Announce Type: new Abstract: Large language models deployed as agents over large tool catalogs face a critical tool-retrieval bottleneck. As embedding-based retrieval approaches rely on compact encoders that may under-capture specialized tool semantics, parametric tool retrieval addresses this by encoding each tool as a virtual token appended to the LLM vocabulary, fine-tuned in two stages (memorization then retrieval SFT) to use the LLM as a retriever, achieving strong performance on standard ToolBench retrieval benchmarks. Yet these benchmarks use verbose, fully-specified queries, and their evaluation applies constrained decoding that restricts outputs to valid token paths, neither reveals whether the model actually understands its tools. We introduce \textbf{ToolSense}, an open-source LLM-powered diagnostic framework that takes any tool catalog as input and automatically generates three benchmarks: a Realistic Retrieval Benchmark (RRB) with queries at three ambiguity tiers, an MCQ probing benchmark, and a QA probing benchmark. Applying ToolSense to ToolBench (~47k tools) and evaluating five parametric model training configurations reveals a knowledge-retrieval dissociation: on RRB queries, several configurations collapse by ~50-64 percentage points compared to fully-specified ToolBench benchmarks, falling below the embedding-model baseline. Additionally, despite strong retrieval performance, some models score near-random on factual probes, suggesting a knowledge-retrieval dissociation. We open-source the ToolSense framework and the ToolBench diagnostic benchmarks at https://github.com/SAP/toolsense.
arXiv:2606.12460v1 Announce Type: new Abstract: This paper studies sovereign stress avalanches and network amplification in Latin American credit markets using monthly J.P. Morgan EMBI Global Diversified spreads for eleven sovereigns over 2007-2026. Country stress events are defined as positive log-spread innovations exceeding country-specific volatility thresholds, and regional avalanches count the number of stressed countries in each month. The empirical design combines finite-sample power-law diagnostics, threshold robustness checks, a country-level reshuffling placebo, and rolling correlation, partial-correlation, and minimum-spanning-tree networks. Avalanche sizes are heavy-tailed, with an estimated exponent of 1.77, while spread changes and inter-event times lie in a heavy-tail boundary regime. The placebo shows synchronization far above independent stress timing, with p-values below 0.001. Large avalanches coincide with denser and more spectrally amplifying raw-correlation networks, but not after partial-correlation filtering, indicating common-factor co-movement rather than conditional regional propagation. Network metrics describe contemporaneous stress regimes rather than early-warning signals. The results provide a finite-size criticality framework for monitoring sovereign fragility in emerging markets.
arXiv:2606.12462v1 Announce Type: new Abstract: This is a continuation of a previous report on an experiment in autoformalisation of G\"odel's second incompleteness theorem in Agda using Claude. Using the framework built in this experiment, Claude could ``automformalise'' Chaitin's proof of the first incompleteness theorem and then the Kritchman-Raz surprise examination paradox version of the second incompleteness. As the first experiment, the project provides a case study of the strengths and limitations of current large language models in mathematics. Since Chaitin's proof involves coding programs, Claude had to represent code as ternary string and could build autonomously a parser and a continuation stack evaluation machine. The fact that we can simulate computations as expected is not completely trivial and we suggested a Gandy/Howard majorisation argument, that Claude had no problem to follow. The resulting formalisation clarifies a number of details left implicit in the original presentation and provides a fully machine-checked proof of these arguments for Church's Basic Recursive Arithmetic.
arXiv:2606.12469v1 Announce Type: new Abstract: Retrieval-Augmented Generation (RAG) systems enhance large language models by grounding responses in retrieved documents from external knowledge sources at inference time. However, this reliance on retrieved content introduces vulnerabilities to poisoning attacks, in which adversarial documents can manipulate both the retrieval process and the generated outputs. This paper investigates poisoning robustness in RAG through a full factorial experimental study covering 432 configurations. We analyze the impacts of dataset, retriever type, retrieval depth, database composition, chunking strategy, and generator model on retrieval-level and generation-level metrics. The results show that retriever architecture, dataset, and retrieval depth are the strongest factors affecting poisoning exposure, while generator choice and database composition have a major impact on downstream attack success. Dense and graph-based retrievers generally improve robustness relative to BM25, whereas larger retrieval depth increases the likelihood of retrieving poisoned passages. We further show that replicating poisoned content across multiple databases amplifies adversarial influence, while additional clean sources can mitigate it. These findings highlight that poisoning vulnerability in RAG is not attributable to a single component, but instead arises from the interaction of retrieval, generation, and knowledge-base configuration.
arXiv:2602.10132v3 Announce Type: replace Abstract: Development and operation of commercially viable fusion energy reactors such as tokamaks require accurate predictions of plasma dynamics from sparse, noisy, and incomplete sensors readings. The complexity of the underlying physics and the heterogeneity of experimental data pose formidable challenges for conventional numerical methods, and highlight the promise of modern data-native approaches. A major obstacle in realizing this potential is, however, the lack of curated, openly available datasets and standardized benchmarks. Existing fusion datasets are scarce, fragmented across institutions, facility-specific, and inconsistently annotated, which limits reproducibility and prevents a fair and scalable comparison of AI approaches. In this paper, we introduce TokaMark, a structured benchmark to evaluate AI models on real experimental data collected from the Mega Ampere Spherical Tokamak (MAST). TokaMark provides a comprehensive suite of tools designed to unify access to multi-modal fusion data and standardize evaluation protocols. The benchmark includes a curated list of 14 tasks spanning a range of physical mechanisms, exploiting a variety of diagnostics and covering multiple operational use cases. A baseline model is provided to facilitate transparent comparison and validation within a unified framework. By establishing a unified benchmark, TokaMark aims to accelerate progress in data-driven AI-based plasma modeling, contributing to the broader goal of achieving sustainable and stable fusion energy. The dataset, benchmark, documentation, and tooling are open-sourced under https://github.com/UKAEA-IBM-STFC-Fusion-FMs/tokamark_baseline.
arXiv:2606.12478v1 Announce Type: new Abstract: Attention mechanisms are central to modern sequence models, yet standard attention computes relevance primarily through individual query--key similarities. Although softmax normalization introduces competition among positions, a standard attention layer does not explicitly parameterize learnable interactions between attention decisions. This limits its ability to directly model cooperative or antagonistic co-attention structure within the attention mechanism itself. We propose Boltzmann attention, an energy-based generalization in which attention patterns are governed by an interacting Ising model. The method augments the usual data-dependent local fields with learnable pairwise couplings, allowing the model to represent inter-position correlations beyond those captured by softmax or sigmoid attention. Experiments on character-level language modeling and synthetic bracket matching show that Boltzmann attention consistently improves over standard softmax attention within a standard Transformer architecture, with the advantage becoming more pronounced as sequence length increases. A four-way ablation confirms that the improvement arises from the learnable pairwise couplings. These results suggest that explicit inter-position interactions provide a principled enhancement for attention-based sequence modeling. Moreover, the Ising formulation opens a natural path toward quantum-computing-based sampling strategies: we demonstrate that diabatic quantum annealing provides a practical training method while maintaining competitive performance with exact Boltzmann computation.