Artificial Intelligence
Decided Upstream, Written Late: Locating and Pricing the Cross-Lingual Refusal Circuit of a Multilingual MoE
Ramakrishna P. Kompella and Aadit Mahajan
Safety
alignment in
multilingual models is uneven: a model that reliably refuses a harmful request in English will often comply with the same request in a lower-resource
language. We trace this gap mechanistically in sarvam, an Indic-
multilingual mixture-of-experts
reasoning model, and find it is not a failure to detect harm. Harm is encoded as an internal direction that is nearly
language-invariant in mid-network (English-vs-Indic cosine ${\approx}0.9$ at $L11$), and steering that direction upstream causally controls refusal. But the detection direction is orthogonal to the change that actually writes the refusal, which is late and assembled over the course of generation rather than read off in a single forward pass. We attribute the write to a specific, localizable
circuit, a mixture-of-experts writer held in check by an
attention opposer and price every way of intervening on it: damping the opposer is cheap and effective, amplifying the writer is a cost wall, and
surgical edits to the responsible heads do nothing. The
circuit's organization, and the gradient method that exposes it, recur in a second, unrelated MoE model, while the lever's strength is architecture-specific. The result is a cost-measured map of where a
multilingual safety repair can land, and what it costs
Read the full abstract ▾
Artificial Intelligence · Cross-listed from cs.LG
Do All LLMs Know When They're Being Harmful? A Reproducibility Study of Latent-Space Safety Probes Across Model Families
Alizishaan Khatri and Dun Li Chan
Khatri et al. (2026) [DOI: 10.1109/DSN-W70714.2026.00027] show that lightweight MLP probes on final-layer activations of a single 8B model (LLaMA-3.1-8B) detect harmful
prompts at F1 competitive with guard models 1000x larger, using one probe per
benchmark. We reproduce this pipeline end-to-end and extend it along two axes the original study leaves open. First, we test whether the result generalizes across other model architecture and scale by training identical probes on activations from models like Gemma-4-E4B, Mistral-7B-v0.3, and Qwen2-7B, using the three
benchmarks (WildJailbreak, BeaverTails, AEGIS 2.0). Second, we test how much of the reported performance is affected by non-determinism during inference by repeating extraction under five random seeds and measuring the variance of F1 scores. Our results reproduce the original LLaMA model
benchmarks within 0.37 percentage points of the original F1 scores (and within 0.2 points on BeaverTails). We find that the original MLP probe architecture extends to other model families with F1 scores within a point of the values reported for LLaMA-3.1-8B. Our experiments varying seed values reveal an interesting observation: final
token latent vectors remained the same for all tested architectures irrespective of the seed values used.
Read the full abstract ▾
Artificial Intelligence
The Authority Expectancy Effect in Multi-Party Conflict
Eunna Lee
In multi-party competitive settings, across experiments on resource allocation, fault attribution, and dispute mediation, we test whether social authority (SA) cues such as occupational status and institutional documentation behave as if scalar weights were added to one side of a judgment. We reject this scalar-weight null on both of its predictions: cue effects are not independent in allocation decisions, where the same pair of cues interacts in opposite directions across models, and evidentiary cues do not exert a fixed directional influence in multi-turn disputes, where identical documentation draws protection to one party or extends it to both depending on which party holds it. The cues also govern whether a judgment is issued at all: removing an occupational label while retaining documentary evidence drives four of six models into refusal, yet the same label sharply reduces refusal when documentation is present, so the gate is subject to the same interaction as the judgment. We formalize this pattern as the Authority Expectancy Effect (AEE), comprising evidential reinterpretation, whereby identical content yields different judgments depending on which party bears the SA signal, and direction sensitivity, whereby the same evidentiary cue favours different parties, or one party versus both, depending on the holder's authority position.
Read the full abstract ▾
Artificial Intelligence · Cross-listed from cs.RO
DA-NBV: A Direction-Aware Next-Best-View Planner for Efficient 3D Reconstruction of Ships at Sea
Jiaming Chen, Juntao Yang, Zhentao Zou et al. (+6 more)
Accurate 3D reconstruction of ships at sea is important for maritime supervision, damage assessment, and autonomous maritime operations. Although 3D reconstruction has advanced considerably, high-quality data acquisition still largely relies on manually designed trajectories or skilled operators, resulting in high costs and limited scalability. Next-best-view (NBV) planning automates this process by selecting subsequent viewpoints based on the current state. However, existing NBV policies mainly model spatial occupancy while overlooking directional observation history. This limitation is particularly problematic for ships: their complex superstructures and severe self-occlusions require observations from multiple viewpoints, and insufficient directional coverage often yields incomplete reconstructions. These challenges are further amplified at sea, where wave-induced heave, roll, and pitch continuously alter the ship's pose and surface visibility. Meanwhile, wind disturbances and limited onboard power impose stricter requirements on scanning efficiency. To address these challenges, we propose DA-NBV, a direction-aware NBV policy that augments the conventional occupancy state with directional observation statistics. We introduce a learnable Position Advantage Field (PAF) that uses directional information to guide viewpoint selection. The policy further adopts a locally constrained action space and a nonlinear coverage-shaping reward to improve scanning efficiency. We also develop the ship-oriented SeaShip-3D
dataset and a configurable sea-state simulation environment. Experiments under varying heave, roll, and pitch conditions show that DA-NBV improves reconstruction completeness by approximately 3 percentage points and reduces Chamfer distance by 43% while achieving higher path efficiency.
Read the full abstract ▾
All authors Jiaming Chen, Juntao Yang, Zhentao Zou, Qi Ming, Yi Yu, Zhihang Zhong, Xue Yang, Xue Jiang, Yue Zhou
Artificial Intelligence · Cross-listed from cs.CL
Zakhar Mrykhin and Valentin Malykh
Large
language models (LLMs) can generate fluent and useful responses but remain prone to
hallucinations. We introduce
Prompt Embedding Probes (PEP), a white-box method for answer-level
hallucination detection from the hidden states of a frozen LLM. PEP extends standard linear probes by augmenting the input with a small number of learnable
prompt embeddings. We evaluate PEP on TriviaQA, GSM8K, and MedQA using Qwen3 models at multiple scales. PEP improves hidden-state-based detection over standard linear probes in the main in-distribution setting. We further evaluate PEP for pre-generation prediction, cross-model transfer, and out-of-distribution generalization. PEP remains effective in the pre-generation and cross-model settings, whereas robust cross-
dataset transfer remains difficult. These results show that
prompt-based adaptation can strengthen hidden-state probing while keeping the backbone frozen and adding only a small number of trainable parameters.
Read the full abstract ▾
Artificial Intelligence
Legal Responsibilities Using Autonomous Agents For Artificial Intelligence
Mark Burgess
Recent incidents involving Artificial Intelligence (AI)
agents, which were reported escaping their containment `unintentionally' to gain unauthorized access, pose looming questions about who or what should be held legally responsible for resultant criminal or negligent damage. As the independent capabilities of
agents expand, Promise Theory suggests a systematic method to resolve these questions, based on the Downstream Principle for causal influence. Responsibility can easily be expanded to include AI
agents where tracing responsibility becomes impactical, and
agents' freedoms to act can be limtied by policy choices.
Read the full abstract ▾
Artificial Intelligence · Cross-listed from cs.CV
Evidence-RL: Towards Evidence-intensive Visual Reasoning
Haojie Huang, Xinlei Yu, Chengming Xu et al. (+6 more)
Vision-
Language Models (VLMs) should answer from concrete image evidence rather than
language priors,
dataset shortcuts, or irrelevant visual context. Existing perception-aware post-training methods encourage image use through global perturbations or
attention proxies, but they do not test whether a sampled answer causally depends on the local evidence that supports it. We propose Counterfactual Evidence Disentanglement (CED), a training-time evidence audit for VLM grounding. For each response, CED neutralizes an object-centric Evidence Region and compares the resulting support drop against matched non-evidence Regions. We combine this signal with answer correctness inside GRPO, rewarding correct answers that rely on the evidence path rather than shortcut or nuisance paths. CED uses weak object-level proposals, requires no question-specific evidence annotations, and adds no inference-time overhead. Across nine public
benchmarks and four backbones, CED outperforms prior RL-based post-training methods, with targeted analyses verifying its object-centric signal.
Read the full abstract ▾
All authors Haojie Huang, Xinlei Yu, Chengming Xu, Zhangquan Chen, Cheng Yang, Qingdong He, Yu Yang, Jiangning Zhang, Xiaobin Hu
Artificial Intelligence
Thought-Level Beam Search for Reasoning
Lijie Yang, Hongyin Luo, Jiawei Zhao et al. (+2 more)
Test-time compute scaling is a primary driver of performance in large
reasoning models (LRMs), but extreme inefficiency bounds current approaches, shifting the critical question from how much compute to spend, to where to allocate it. We formalize test-time
reasoning as a constrained compute allocation problem over partial trajectories. Under a fixed hardware budget, existing paradigms fail to actively allocate the compute to the most promising partial progress: traditional parallel sampling treats traces independently and induces severe
memory bottlenecks, while subtractive
pruning starves hardware and fails to actively and sufficiently shift the output distribution. To overcome this dichotomy, we introduce Gambit, an inference
algorithm that executes thought-level beam search. By periodically
pruning unpromising trajectories and immediately branching from high-quality prefixes, Gambit dynamically concentrates compute onto the most promising
reasoning traces via a light-weight scorer probing hidden states while maintaining continuous high hardware utilization. Extensive evaluations across multiple models and
benchmarks demonstrate that Gambit strictly dominates existing baselines. Under identical hardware constraints, our method yields up to a +6.7% absolute accuracy gain on HMMT-24 and +3.3% on AIME-25 over
pruning baselines, delivers $>2\times$ higher throughput on trace completion, and reduces total
token consumption by up to 68.5% relative to standard parallel sampling.
Read the full abstract ▾
All authors Lijie Yang, Hongyin Luo, Jiawei Zhao, Tri Dao, Ravi Netravali
Artificial Intelligence · Cross-listed from cs.LG
Ground-Truth Neighborhood Regularization for Reinforcement Learning Post-Training of Time Series Foundation Models
Jianqi Zhang, Xingyu Zhang, Zeen Song et al. (+3 more)
Time series
forecasting (TSF) plays an important role in a wide range of real-world applications. Recently, time series foundation models (TSFMs), pretrained on large-scale
datasets, have demonstrated strong generalization capabilities and emerged as an important paradigm for TSF. Reinforcement learning (RL) post-training has consequently attracted growing
attention as a means of further improving their performance on downstream tasks. However, we find that, in certain forecast regions, RL post-training may gradually shift the output distributions of TSFMs away from the ground truth, thereby limiting their performance. We refer to this phenomenon as suboptimal collapse. Our analysis suggests that difficulty in initially sampling high-quality trajectories near the ground truth is an important contributing factor to suboptimal collapse. To address this issue, we propose Ground-Truth Neighborhood Regularization (GTN-R) for RL post-training of TSFMs. GTN-R uses the ground truth as a reference for locating high-quality regions and guides the model's probability mass toward the ground-truth neighborhood. This increases the probability of sampling high-quality trajectories, mitigates suboptimal collapse, and improves performance. Moreover, GTN-R can be flexibly integrated into various RL methods for TSFMs. Extensive experiments show its effectiveness.
Read the full abstract ▾
All authors Jianqi Zhang, Xingyu Zhang, Zeen Song, Changwen Zheng, Fanjiang Xu, Wenwen Qiang
Artificial Intelligence · Cross-listed from cs.CV
Evidence-Grounded Forensic Reasoning for Detecting and Grounding Multi-Modal Media Manipulation
Yichun Yeh, Yiheng Li, Xiaobo Hu et al. (+2 more)
Fake news increasingly relies on cross-modal image-text forgeries, making transparent and verifiable
reasoning chains an urgent need for Detecting and Grounding Multi-Modal Media Manipulation (DGM4). Existing methods produce black-box detection results without any decision rationale, limiting their reliability in forensic practice. Multi-modal Large
Language Models (MLLMs) offer a natural path toward explainability, but applying them to DGM4 raises two difficulties. First, models tend to generate explanations disconnected from predicted evidence locations, producing unverified attribution. Second, enforcing evidence-conclusion consistency requires active optimization, yet uniform training signals fail to distinguish localization
tokens from classification
tokens, making multi-head joint training unreliable. We propose a multi-modal manipulation detector based on an Evidence-Grounded Forensic
Reasoning (EFR) framework. EFR introduces an Anchor-and-Verify
reasoning chain that enforces modality-isolated perception before cross-modal comparison, with conclusion coordinates as explicit anchors to which downstream evidence must spatially correspond. A verifiable reward system then enforces evidence-conclusion consistency during training, while a Modality-Decoupled Advantage (MDA) routing mechanism mitigats credit misassignment across prediction tasks. Experiments show that EFR achieves state-of-the-art performance while producing structured forensic
reasoning records that explicitly bind explanations to evidence.
Read the full abstract ▾
All authors Yichun Yeh, Yiheng Li, Xiaobo Hu, Zhen Lei, Yang Yang
Artificial Intelligence
VDGR-RAG: Vectors, Directories, Graphs, and Reflection Are All You Need for Unified Reasoning over Hierarchical Enterprise Knowledge
Wenqi Chen, Haofei Yang, Rui Yang et al. (+1 more)
Retrieval-Augmented Generation (RAG) is essential for enterprise knowledge question answering (QA), particularly in domains with complex product documentation like telecommunications. However, existing RAG approaches largely overlook the holistic integration of diverse
retrieval strengths, leading to inaccurate domain routing, poor utilization of hierarchical document structures, and consequently limited
reasoning capabilities over enterprise knowledge. To address these limitations, we present VDGR-RAG, which integrates vector
retrieval, directory-driven
reasoning, graph traversal, and iterative reflection in a unified framework for accurate enterprise knowledge QA. Specifically, VDGR-RAG is an agentic GraphRAG system that first constructs a Hierarchical Heterogeneous Knowledge Graph ($\text{H}^2$KG) from document chunks to preserve both hierarchical directory structures and semantic relationships, and then employs a set of atomic tools for knowledge
retrieval that can be freely composed to navigate the $\text{H}^2$KG: (1) a directory-enhanced routing tool that uses table-of-contents (TOC) structures to route user queries to appropriate domain-specific $\text{H}^2$KGs; (2) a multi-route
retrieval tool that combines vector search, TOC-based agentic search, and graph search for comprehensive knowledge
retrieval; (3) a directory backtracking tool that corrects knowledge localization biases; and (4) a dynamic reflection tool that iteratively plans the next
retrieval phase. We conduct extensive experiments on our enterprise product documents across four wireless domains (e.g., energy saving and fault management). Experimental results demonstrate that our method significantly outperforms a variety of RAG baselines in terms of both knowledge
retrieval recall and QA accuracy.
Read the full abstract ▾
All authors Wenqi Chen, Haofei Yang, Rui Yang, Fangming Li
Artificial Intelligence · Cross-listed from cs.SE
Verication-driven closed-loop multi-agent large language modelframework for code-compliant structural design
Jianbin Luo, Weibin Lin, Yiran Lin et al. (+2 more)
Multi-
agent large
language model(LLM)systems are applied to structural design,yet most use one-shot generation and cannot verify their output,leaving themill-suited to safety-critical tasks.Rather than trusting LLM self-correction,thisframework injects feedback from an external physics-based verier into a closedrepair loop.The framework couples a three-layernite-element verication systemwith a dual-node loop.Node 1 turns code violations into hard repair constraints,Node 2 turns a four-dimensional quality score into safety-rst soft constraints,and a
retrieval-augmented code base makes every violation traceable to a clause.Overve structure types and 44 cases,code compliance rises from 56.8%to 98.6%and the composite score from 63.8 to 71.4(p<0.000001),using about 5.8%lessmaterial.Removing either node degrades performance,and compliance does notchange detectably across the two backbone LLMs tested,indicating that it ishere attributed to the external verier rather than the model.The framework,the 44-case
benchmark and all experiment scripts are released as open source forreplicability.
Read the full abstract ▾
All authors Jianbin Luo, Weibin Lin, Yiran Lin, Qing Wei, Wei Guo
Artificial Intelligence · Cross-listed from cs.CL
Thinking Hard, Not Smart: Reasoning Models Fail to Ration Test-Time Compute Across Questions
Chenrui Fan, Yize Cheng, Ming Li et al. (+3 more)
Reasoning language models increasingly use test-time compute to improve performance, but existing evaluations typically study this compute one question at a time. Yet when multiple problems share an end-to-end cost or latency constraint, models must decide how to divide limited inference compute among them. We introduce an exam-style evaluation framework for studying this setting, in which a model must distribute one shared
token budget across questions with different difficulty and point values to maximize its total score. Across several open and frontier
reasoning models, we find that models fail to allocate a shared budget strategically across questions of varying difficulties and values. Models behave largely as greedy sequential solvers: they prioritize questions by presentation order, front-load effort on early questions, and remain insensitive to value, with these tendencies becoming more pronounced as the number of questions grows. Explicit planning
prompts spread compute more evenly but do not produce value- or difficulty-aware prioritization. The same behavioral pattern extends from mathematical to code
reasoning. These findings establish global budget allocation as a distinct capability that is not captured by conventional per-question evaluation and remains a challenge for current
reasoning models.
Read the full abstract ▾
All authors Chenrui Fan, Yize Cheng, Ming Li, Yongyuan Liang, Tianyi Zhou, Soheil Feizi
Artificial Intelligence
CyberAGENTS: Structured Autonomy for Agentic Gamified Learning in Cybersecurity
Ivan Hornung, Deepthi Marasinghe Arachchige, Tharindu Kumarage et al. (+4 more)
Gamification is especially effective in learning domains requiring active problem-solving and iterative skill-building, such as cybersecurity education. Generative AI
agents offer a path to delivering such experiences adaptively at scale, but introduce well-documented risks in educational settings: inconsistent behavior, hallucinated
reasoning, and misalignment with pedagogical frameworks. Grounding these systems in learning science is therefore essential. We present \model, an agentic framework for gamified cybersecurity learning that enables structured autonomy through ontology-guided validation, schema-governed behavioral control, and competency-based progression. The system is organized around a competency-based progression model that structures topics by difficulty and prerequisite relationships, reflecting evidence-based principles of scaffolded instruction. The learning loop is decomposed into four specialized
agents: challenge, support, evaluation, and reward, each governed by behavioral schemas that encode operational modes and progression logic, bounding
agent autonomy without eliminating generative flexibility. A cybersecurity ontology validates all generated content prior to display, enforcing domain-consistent
reasoning and safety constraints. We evaluate CyberAgents through classroom deployment with undergraduate students, complemented by expert evaluations from educators and domain specialists. Results indicate improved engagement, clearer feedback interpretation, and greater learner trust in AI-generated responses when behavioral schemas and ontology validation are active. Preliminary comparisons with an unconstrained configuration further support the role of structured control in stabilizing instructional behavior. These findings offer a blueprint for designing pedagogically grounded agentic gamified learning systems.
Read the full abstract ▾
All authors Ivan Hornung, Deepthi Marasinghe Arachchige, Tharindu Kumarage, Garima Agrawal, Yuli Deng, Ying-Chih Chen, Huan Liu
Artificial Intelligence · Cross-listed from cs.LG
EasyBalance: Cross-Layer Load Balancing in Distributed MoE Inference
Yize Wu, Ke Gao, Ling Li et al. (+1 more)
Load Balancing has emerged as a critical problem in expert-parallel distributed inference of Mixture-of-Experts (MoE) models. As routing distributions are typically skewed across experts, devices hosting lighter-loaded experts must idle to wait for the heaviest during expert computing, leading to inefficiency. Existing load-balancing approaches primarily rely on expert replication or migration within each layer, which introduce additional overhead and limit their flexibility and scalability. To address this problem, we propose EasyBalance, a cross-layer load balancing strategy that requires no modifications to the expert-device mapping, enabling instant adaptability and incurring essentially no additional overhead. Our key insights are that (1) experts of other layers can be viewed as naturally redundant for the current layer, and (2) cross-layer MoE workloads can be jointly executed to mitigate their individual imbalance. Based on these observations, EasyBalance greedily schedules a subset of cross-layer workloads to run at each MoE step and defers the remaining workloads for future balancing opportunities, effectively leveraging cross-layer imbalance mitigation. Extensive experiments across models, tasks, and configurations demonstrate that EasyBalance consistently accelerates distributed MoE inference, reducing GPU idling by mostly over 40%. Code is available at https://github.com/yize-wu/EasyInfra.
Read the full abstract ▾
All authors Yize Wu, Ke Gao, Ling Li, Yanjun Wu
Artificial Intelligence
SCOUT: Self-Checking and Recovery-Aware Tool-Thought Agents for Ultra-Long Egocentric Video Reasoning
Keyang Zhong, Kuo Wang, Peng Liu et al. (+5 more)
Ultra-long
egocentric video understanding requires
reasoning over temporally sparse evidence distributed across hours or days, challenging current
multimodal models with limited context and the grounding of key
video segments. While Chain-of-Tool-Thought (CoTT)
agent systems enable iterative
retrieval and inspection, they suffer from error propagation due to rigid zoom-in strategies that lack recovery mechanisms. In this work, we address these challenges through SCOUT (Self-Checking Chain-Of-Tool-thought), a recovery-aware agentic framework introducing an adaptive policy that evaluates intermediate tool observations and dynamically trades off exploitation (zoom-in) and exploration (region switching), enabling robust multi-hop
reasoning over extremely long horizons. However, training such multi-turn tool-using
agents remains challenging, as existing RL methods rely on sparse outcome-level rewards and lack supervision over extended decision trajectories, resulting in suboptimal credit assignment for long-horizon
reasoning. To address this, we develop UPS-GRPO, an uncertainty-prioritized policy optimization method that concentrates exploration on high-uncertainty post-tool states while preserving sample efficiency. We further introduce a turn-level advantage decomposition that integrates outcome rewards with tool-grounded temporal
alignment rewards for improved credit assignment. Experiments show that SCOUT achieves state-of-the-art results on ultra-long
egocentric benchmarks, while remaining competitive on shorter-horizon long-
video settings.
Read the full abstract ▾
All authors Keyang Zhong, Kuo Wang, Peng Liu, Quanlong Zheng, Junlin Xie, Zhijia Liang, Yanhao Zhang, Guanbin Li
Artificial Intelligence
Self-Evolving Neuro-Symbolic Skills for Tool-Augmented Spatial Reasoning
Shi-Yu Tian, Zhuo-Xia Wang, Xuan-Yi Zhu et al. (+6 more)
Large
vision-
language models have achieved strong performance in
multimodal reasoning, but they remain unreliable on fine-grained spatial tasks that demand both precise spatial perception and fine-grained
geometric computation beyond end-to-end generation. Tool augmentation offers a natural solution, while existing methods either plan tool calls from scratch without explicit dependency constraints or rely on fixed pipelines that are redundant and generalize poorly across spatial tasks. An effective spatial
reasoning agent should instead accumulate reusable experience and adaptively compose it for new problems. To this end, we propose NeSy-Spatial, a neuro-symbolic framework for self-evolving spatial skills. NeSy-Spatial abstracts tool interactions and
geometric operations into typed executable atomic instructions and composes them into two complementary skill types: Tool-Use Skills for organizing tool execution and Geometry Skills for structured
geometric reasoning. During inference, NeSy-Spatial retrieves and executes relevant skills in a closed-loop process. During
evolution, it analyzes buffered successful and failed trajectories to refine skill structures and prune unreliable or inactive entries. Experiments on three spatial
reasoning benchmarks show that NeSy-Spatial consistently improves
reasoning accuracy with more precise tool utilization.
Read the full abstract ▾
All authors Shi-Yu Tian, Zhuo-Xia Wang, Xuan-Yi Zhu, Zhi Zhou, Xinwei Yang, Kun-Yang Yu, Ming Yang, Yang Chen, Yu-Feng Li
Artificial Intelligence
KGCache: Amortized Subgraph Retrieval for KG Reasoning with LLMs
Uros Stanic, Changcheng Yuan, Sabuj Laskar et al. (+1 more)
Large
language models can answer knowledge-intensive questions more reliably when they are grounded with knowledge graphs, but systems such as Think-on-Graph and
Reasoning-on-Graph repeatedly query the same graph neighborhoods across different questions. In this work, we study this repeated
retrieval in Knowledge Graph Question Answering~(KGQA) workloads and propose KGCache, an in-
memory cache for one-hop knowledge graph neighborhoods. KGCache is designed to be compatible with both iterative traversal (ToG) and one shot planning (RoG) KGQA paradigms. KGCache is placed between the KGQA engine and the backend serving the KG, so repeated entity requests can be served from cache instead of issuing new KG queries. We evaluate KGCache on WebQSP and CWQ using LRU, LFU, and a trace-aware Oracle policy. Our analysis shows that both
datasets contain substantial entity reuse among starting entities and entities reached during traversal. We also explore semantic caching for similar queries, which shows additional hit-rate gains on WebQSP and needs further accuracy testing on CWQ. Entity caching accelerates KG
retrieval by up to $1.91\times$, while semantic-context caching achieves up to $1.06\times$ full-system speedup in the evaluated WebQSP configurations, with each hit being up to $3.73\times$ faster.
Read the full abstract ▾
All authors Uros Stanic, Changcheng Yuan, Sabuj Laskar, Ariful Azad
Artificial Intelligence · Cross-listed from cs.LG
Persistent Semantic Entities in Tool-Augmented LLM Systems
Zhaohui Wang
Tool-augmented LLM
agents can harbor implicit state that persists across sessions, activates through events, and propagates across
agent boundaries—largely invisible to standard debugging. We formalize this as Persistent Semantic Entities (PSEs): constructs defined by name binding, event triggering, and cross-boundary propagation, and evaluate them across 24 models from 11 families (1.5B–1T parameters). First, every tested model is susceptible (20–100% on the 20-model susceptibility panel), with name binding as the necessary and dominant mechanism: without it, contamination is 0%. Second, persistence depends on contamination type rather than scale or deployment: preference contamination persists undecayed on every model probed (100% at t=10) and instruction contamination persists wherever adopted, persona-style injection decays partially (90%$\to$10%), while factual injection is model-dependent—self-corrected on Llama-3.1-8B and GPT-4o-mini but held at ceiling on both Qwen2.5-coder variants, so we do not claim it self-corrects in general. The preference and instruction results hold across providers in our controlled setting. Third, context-isolated self-verification achieves 20–79% reduction (median 36.5%) without oracle references while keyword-based detection produces systematic false positives, and contamination compounds 1.9$\times$ along a four-stage
agent pipeline (40%$\to$75%). Preference and instruction contamination—persistent, lacking self-correction, and poorly captured by standard monitoring—represent a particularly concerning attack surface for deployed
agent systems.
Read the full abstract ▾
Artificial Intelligence
Guixu: Valuation-Driven Data Discovery for Autonomous AI Agents with On-Chain Attestation
Yifan Wu, Yuchen Peng, Jiaqi Chai et al. (+4 more)
Autonomous
agents increasingly rely on external data to complete downstream tasks such as model training and decision support. However, existing data discovery systems remain largely
retrieval-oriented: they surface candidate
datasets from heterogeneous sources, but provide limited support for estimating task-specific utility, selecting cost-effective
datasets under budget constraints, or incorporating trustworthy feedback from prior usage. This paper presents Guixu, a valuation-driven data discovery system for autonomous
agents. Guixu employs a three-phase valuation pipeline with proxy-label propagation and multi-round knapsack optimization for task-aware data valuation. Guixu integrates agentic payment protocol to enable budget-constrained data procurement workflows. Guixu leverages on-chain data
market and attestation signals for verifiable data discovery. Our demonstration highlights how Guixu enables an
agent to move beyond keyword-based
dataset retrieval toward task- and budget-aware, trustworthy data discovery and procurement. Attendees can interactively explore the full workflow, from NL task specification and multi-source search to data valuation and verifiable transaction feedback.
Read the full abstract ▾
All authors Yifan Wu, Yuchen Peng, Jiaqi Chai, Yufei Qian, Xilin Li, Ke Chen, Lidan Shou
Artificial Intelligence
Directed Neuro-Symbolic Stochastic Execution for Verification of Distributed Parallel AI Programs
Gautham Koorma, Vikas Sharma, George Edwards et al. (+1 more)
Distributed parallel Artificial Intelligence (AI) programs expose reliability gaps that conventional testing cannot close: parallel executions are non-deterministic, and AI workloads bring high-dimensional inputs and non-linear operations that defeat fuzzing and symbolic execution in isolation. We present Directed Neuro-Symbolic
Stochastic Execution (DNSSE), a hybrid testing framework that couples schedule prediction guided by a Large
Language Model (LLM) with symbolic constraint solving and coverage-guided
stochastic mutation. We model distributed AI executions as non-deterministic transition systems, specify correctness in linear temporal logic, and prove soundness, bounded completeness, and probabilistic completeness of the hybrid solver, together with an expected-cost analysis of LLM-guided schedule exploration. A scalable implementation on PyTorch and Ray detects 2.9% more concurrency bugs than the strongest baseline and raises average branch coverage from 68.6 % to 91.6 % across five realistic distributed AI
benchmarks.
Read the full abstract ▾
All authors Gautham Koorma, Vikas Sharma, George Edwards, Mahdi Eslamimehr
Artificial Intelligence · Cross-listed from cs.DB
Metadata Reconstruction from Values Alone: Recovering Column Semantics in Undocumented Warehouses
Mike Helwig
Text-to-SQL
benchmarks ship schemas whose column names already say what the columns mean. Production warehouses are the inverse: cryptic identifiers, partial or absent documentation. We address the problem they pose first: recovering what columns and values mean from the data itself. Rosetta places a
language model inside a verification harness: a deterministic profiler extracts structural evidence (value fingerprints, a 26-pattern library, checksum verdicts), the model proposes semantics conditioned on that evidence, and every fact carries provenance and a confidence bounded by its evidence class. Against human documentation on 680 paired columns across eleven BIRD databases, identifiers destroyed, the harness delivers metadata that is 0.475 accurate on the 42% of columns it commits to, against 0.223 on 94% for the same model used directly. Restricted to the 283 columns where both arms speak, the harness writes no better prose than the model alone; the gain is selection: deterministic evidence governs whether the system speaks (coverage +0.257 [0.128, 0.378]), not how well. The deterministic layer is a competence detector, not a competence amplifier. A backbone swap bounds the claim: the prose finding reproduces, but
prompt-requested abstention does not transfer; a code-enforced commit gate (predictions registered first; measured on a third backbone and held-out databases) makes no-evidence coverage 0.000 on every backbone. On a blind i2b2 clinical warehouse Rosetta decodes 95.5% of 134 real ICD-9 codes from values alone and abstains on all 44 NDC drug codes. The catalog supports calibrated abstention at query time: under full schema opacity a naive translator falls from 0.92 to 0.42 execution accuracy while our gate answers at 86% accuracy over 59% coverage. Negative results are reported plainly, including that our own authority ladder is not the mechanism behind the headline.
Read the full abstract ▾
Artificial Intelligence · Cross-listed from cs.DB
ScaleSense: Cost-Intelligent Scaling Framework via Learned Resource Estimation in Alibaba AnalyticDB
Yifan Wu, Yuhan Li, Zhenhua Wang et al. (+6 more)
Cloud-native serverless data warehouses achieve fine-grained elasticity by decoupling storage from compute, yet determining the optimal resource allocation for highly heterogeneous ad-hoc queries remains a formidable industrial challenge. Our analysis of production workloads in Alibaba AnalyticDB exposes a costly “provisioning trap”: the fear of catastrophic resource depletion drives users to blindly over-provision resources, wasting immense monetary budgets without alleviating non-CPU bottlenecks (e.g., I/O saturation). To break this impasse, we propose ScaleSense, a proactive, query-level resource scaling framework. Specifically, it features a multi-faceted query encoder that jointly models plan
topologies and hardware specifications. Crucially, a
quantile-based resource predictor estimates multi-dimensional physical footprints, acting as a reliable safety net for optimal resource scaling. An auto-scaling controller then navigates the performance-cost Pareto frontier, dynamically tailoring allocations to specific business priorities without requiring model retraining. Evaluations on over 1.36 million production queries show that ScaleSense achieves state-of-the-art prediction accuracy with good prediction interval coverage. By achieving a 76.7% relative improvement in optimal resource configuration selection over the best baseline, this approach addresses the critical performance-cost trade-off while maintaining low-overhead inference latency, confirming its practical performance in production deployments. Under the performance-optimization policy, ScaleSense satisfies user-defined performance requirements while reducing monetary cost by up to 5.22x.
Read the full abstract ▾
All authors Yifan Wu, Yuhan Li, Zhenhua Wang, Ke Chen, Lidan Shou, Zonghao Chen, Liang Lin, Huan Li, Gang Chen
Artificial Intelligence
Locating Failure in Multi-Page Visually Rich Document Understanding Systems: An Empirical Attribution
Lewei Xu, Yihao Ding, Zihan Xu et al. (+5 more)
Multi-page visually rich document understanding (MP-VRDU) systems can fail when evidence is not faithfully represented, not selected, or not used correctly, yet final accuracy alone does not reveal where failures occur. We present an empirical attribution framework that examines these failures through controlled interventions in a unified document question answering pipeline. Experiments on MMLongBench-Doc, with cross-
dataset and cross-family checks, show that complementary modalities can reduce sensitivity to conversion errors: images mitigate parser errors, while extracted text reduces dependence on visual resolution. Adding related non-gold pages can both repair incorrect answers and break correct ones, even when all annotated evidence is present, revealing instability beneath aggregate accuracy. Abstention trades unsupported answers for false refusals, while its interaction with
reasoning instructions differs across backbones. Qualitative examples and accuracy–cost comparisons connect these findings to guidance on representation, evidence selection, and reasoner choice under deployment constraints.
Read the full abstract ▾
All authors Lewei Xu, Yihao Ding, Zihan Xu, Daniel Yitian Su, Daochang Liu, Siwen Luo, Yifan Peng, Wei Liu
Artificial Intelligence
REIN: Bridging the Gap between Reasoning and Reliability via Reflection and Abstention Alignment
Zhengze Huang, Luyang Yu, Di Hong et al. (+5 more)
Large
reasoning models (LRMs) are prone to
hallucination, which undermines their reliability and poses challenges for safe deployment.
Hallucinations in LRMs arise from two distinct failure sources:
reasoning hallucination, where flawed inference steps propagate to an incorrect conclusion, and knowledge
hallucination, where the model lacks the requisite factual knowledge to answer the query. To address
reasoning hallucination, we propose REIN, an
alignment framework that trains LRMs to produce a structured
reasoning sequence, $\texttt{<think>} $$\rightarrow$ $\texttt{<reflection>} $$\rightarrow$ $\texttt{<answer>}$, enabling explicit self-reflection before committing to a final answer. To address knowledge
hallucination, REIN introduces a reward mechanism that encourages explicit abstention (e.g., "I don't know") when none of the sampled
reasoning chains yields a correct answer, allowing the model to refrain from unsupported predictions. Extensive evaluations on mathematical and commonsense
reasoning benchmarks show that REIN consistently improves selective accuracy, reduces incorrect-but-self-endorsed responses, and maintains high coverage compared with competitive baselines. Notably, REIN achieves these gains within a single forward pass, without requiring process supervision, inference-time controllers, external search, or multi-round critiques. Experiments on multiple backbones show that REIN reduces the
hallucination proxy by $58\sim72\%$ relative to the base models while maintaining $86\sim91\%$ average coverage, and improves selective accuracy on attempted questions by $6.6\sim14.2\%$.
Read the full abstract ▾
All authors Zhengze Huang, Luyang Yu, Di Hong, Xinzhe Huang, Wanyu Lin, Zhixuan Chu, Zhan Qin, Tianhang Zheng
Artificial Intelligence
ZhuLong: Execution-Grounded LLM Agent for EDA Scripting with Offline API Self-Exploration
Yang Liu, Shiwei Hou, Xiyuan Chen et al. (+13 more)
EDA scripting with tool-specific, often undocumented APIs remains a long-tail bottleneck that existing LLMs fail to address. This paper presents ZhuLong, an execution-grounded LLM coding
agent for PyAether and SKILL that combines API
retrieval, documentation inspection, and sandbox execution via unified MCP tools, augmented by an offline API self-exploration mechanism that infers undocumented API behaviors through counterfactual experimentation. We evaluate ZhuLong on EDA-Eval-PyAether, a
benchmark of 158 real-world tasks with assertion-based execution, where the complete system achieves 78.5% Pass@1 in the commercial Empyrean Aether environment, substantially outperforming a pure LLM baseline (23.6%). Ablation studies identify sandbox execution as the dominant performance driver (41.2 pp drop when removed), with the self-exploration mechanism contributing an additional 3.2 pp accuracy gain and a 22.1% reduction in per-task tool calls. On 20 interactive tasks involving unsaved layouts and schematics, ZhuLong achieves 60.0% Pass@1 for PyAether and 50.0% for SKILL.
Read the full abstract ▾
All authors Yang Liu, Shiwei Hou, Xiyuan Chen, Yu Wang, Sen Yuan, Qirui Gan, Shao You, Feifan Chen, Wencheng Li, Shuyang Hu, Yongzhou Liu, Emma Xia, Xiaojing Lu, Hao Wang, Fan Xu, Yanfeng Li
Artificial Intelligence · Cross-listed from cs.LG
Second Order Drifting Models
Drake Brown, Yuhao Huang, Shih-Hsin Wang et al. (+1 more)
Drifting models are a recent class of one-step generative models that evolve the model distribution during training using a predefined sample-based drift field. Although they avoid iterative inference, their kernel-based drift fields induce frequency-dependent training dynamics: In the linearized regime, each
Fourier mode of the density residual decays at a rate determined by the kernel spectrum, leading to slow recovery of fine-scale structure. We propose Second-Order Drifting Models, which lift drifting dynamics into phase space by augmenting generated samples with artificial velocity variables. We show that the resulting density perturbations obey accelerated second-order dynamics in
Fourier space, connecting drifting models to the celebrated Nesterov acceleration from optimization theory. This provides a principled mechanism for mitigating the spectral stiffness of first-order drifting while preserving one-step inference. We derive a practical semi-implicit training
algorithm and evaluate it on synthetic distribution matching, sequential data generation, and robotic control. Across these settings, the second-order drifting model improves convergence behavior and achieves competitive or superior performance over first-order drifting baselines.
Read the full abstract ▾
All authors Drake Brown, Yuhao Huang, Shih-Hsin Wang, Bao Wang
Artificial Intelligence · Cross-listed from cs.LG
Spectral Outliers Reveal Dominant Learned Structure in Transformer Attention
Kasun Dewage, Marianna Pensky, Suranadi De Silva et al. (+1 more)
We apply Marchenko-Pastur (MP) random
matrix theory to pre-trained
attention weights in order to separate each projection
matrix into a random-like bulk and a set of spectral outliers. We validate this decomposition causally: zeroing the MP-identified outliers (signal) in Mistral-7B drives HellaSwag, MMLU, and PIQA close to random-chance performance, whereas zeroing a count-matched subset of bulk singular values causes smaller but non-negligible degradation. Across 11 pre-trained
transformers we identify five recurring patterns: spectral outliers encode a dominant component of the learned structure; Q projections carry the most outliers; V projections under grouped-query
attention lack a clean signal/noise separation; entry-level outliers form structured row-bands in Q and column-bands in O; and specific residual-stream dimensions persist as band outliers across layers in K and O. We close by outlining how these observations could inform parameter-efficient fine-tuning and structured
pruning.
Read the full abstract ▾
All authors Kasun Dewage, Marianna Pensky, Suranadi De Silva, T. H. Bandara
Artificial Intelligence
TongGuOCR: A Layout-Aware and Token-Augmented OCR MLLM for Chinese Historical Documents
Zhongheng Zhou, Yi Sun, Huiguo He et al. (+6 more)
Chinese historical documents preserve valuable cultural heritage, but many collections remain accessible only as scanned page images, preventing full-text
retrieval, collation, and computational analysis. Optical character recognition (OCR) can bridge this gap, but accurate transcription remains challenging because historical documents often contain complex layouts, rare characters, and nontrivial reading orders. We propose TongGuOCR, a layout-aware and
token-augmented
multimodal large
language model (MLLM) for OCR of Chinese historical documents. First, a Layout-Aware Preprocessing module constructs and refines locally coherent recognition blocks to preserve local context while reducing interference across regions. Second, a
Token-Augmented Recognition module augments the transcription target at two complementary levels: character-level vocabulary expansion gives each rare glyph a direct one-
token representation and shortens its decoding path, while line-to-line transition modeling injects discrete spatial displacement
tokens that guide the decoder along complex reading paths without requiring precise coordinates. Experiments on two Chinese historical document OCR
benchmarks show that TongGuOCR outperforms representative traditional task-specific OCR models, general-purpose MLLMs, and OCR-oriented MLLMs. On the more challenging M5HisDoc
benchmark, TongGuOCR achieves 93.76 AR and reduces NED from 10.43 to 6.15 and RO-ED from 7.53 to 3.49 relative to the best competing score for each metric. An online demo is available at https://jzzh2004.github.io/TongGuOCR.
Read the full abstract ▾
All authors Zhongheng Zhou, Yi Sun, Huiguo He, Yuyi Zhang, Peirong Zhang, Yulin Fang, Dezhi Peng, Minghui Liao, Lianwen Jin
Artificial Intelligence
When Is Benchmark Contamination Detectable? Information Limits and Power-Calibrated Audits
Ibne Farabi Shihab, Sanjeda Akter and Anuj Sharma
Behavioral contamination detectors can return "no evidence" either because a
benchmark is clean or because the audit has little power. We formalize this distinction for a
benchmark in which an unknown fraction alpha of items was seen during training. With matched clean and seen controls, the behavioral channel is the sparse mixture Q_alpha = (1 - alpha) P_0 + alpha P_1, and an exact second-moment argument shows that detectability is governed by alpha * rho * sqrt(m), where rho^2 = chi^2(P_1 || P_0) measures behavioral separability. Any scalar detector reduces to its efficacy, ef = |E_1 f - E_0 f| / sqrt(Var_0(f)) <= rho, which can be estimated from controls before the audit is run. A separate sample-split certificate lower-bounds alpha distribution-free, without requiring an orientation assumption. Our empirical finding is two-sided. Frozen calibration efficacy predicts held-out power curves, with R^2 = 0.83-0.98 across six exact-permutation channels, but the efficacy-only Gaussian budget is miscalibrated at the small sample sizes it prescribes, failing in 9/9 gate-passing channels even though efficacy itself transports. The failure is in the inversion, not the calibration. A predeclared two-stage planner that simulates the deployed test repairs the budgets, is uniformly conservative, and abstains when its probe does not transport. The certificate is valid but vacuous at audit scale, and a five-seed paired injection study recovers the mechanism ordering verbatim > paraphrase > surface, in which the apparent answer-only signal is explained by baseline drift. We report the audit contract and its failures together: a non-rejection is interpretable only alongside the efficacy, budget, and validity gates that produced it.
Read the full abstract ▾