Linear probes can decode safety‑relevant concepts such as truthfulness from language‑model activations, but probe accuracy may reflect only decodability, not causal influence on model behavior. The authors show that probe weight geometry alone cannot identify the features the model actually uses, because geometrically aligned features need not be causally relevant. They introduce a sparse‑autoencoder (SAE) decomposition that ranks features by probe alignment and gradient sensitivity, and demonstrate that ablating shared, probe‑only, and random feature sets reveals a sharp dissociation: shared features drive model output changes far more than probe‑only or random features, confirming that causal relevance requires intervention beyond weight geometry.
By Devesh Tiwari, Camille Davis, Shivank Sinha, Talia Weaver, Aditya Shah, Maheep Chaudhary
ContrAgent is a contract‑based framework that provides symbolic temporal supervision for large language model agents. It records an agent’s tool‑call sequence as a trace of checkable predicates and formalizes desired behaviors with assume‑guarantee contracts expressed in linear temporal logic over finite traces (LTLf). Each contract is compiled into a deterministic finite automaton that both gates actions online and evaluates recorded traces offline, enabling deterministic, reproducible verdicts and significantly lower per‑call latency compared to existing LLM‑judge and rule‑based guardrail baselines.
By Yifeng Xiao, Pierluigi Nuzzo
Re2A is a new framework for situated conversational recommendation that models user interactions within shared physical environments. It introduces rubric-based preference reasoning to explicitly capture user preferences from dialogue history and scene context, and a preference-conditioned optimization to align generated responses with both user satisfaction and situational consistency. Experiments on two SCR datasets show that Re2A outperforms existing methods, providing more precise and context-aware recommendations.
By Dongding Lin, Jian Wang, Xiaoyan Zhao, Wenjie Li
The article discusses the necessity of building trust in AI for railway applications, highlighting that AI is currently limited to non‑safety critical uses due to stringent industry standards. It proposes focusing on three key areas—robustness, Operational Design Domain (ODD), and explainability—to meet compliance and safety requirements. By integrating these domains within a safe MLOps environment, the authors argue that regulatory acceptance and public confidence can be achieved, enabling broader AI adoption in mission‑critical railway systems.
By Lefebvre Renard Cl\'ement, L\'eb\'e Vincent, Da Silva Ribeiro Pereira Ricardo, Sundell Johan, Jaoul Arnaud Saiah Kenza, Mijatov\'ic Nenad
The paper introduces "cunning questions"—non‑safety prompts that contain misleading premises or subtle inconsistencies—to train large language models (LLMs) to scrutinize underlying intent and assumptions. Experiments show that incorporating these questions improves robustness against out‑of‑distribution jailbreak attacks and enhances subsequent safety fine‑tuning, achieving a new state‑of‑the‑art reduction in mean ASR from 17.40% to 15.05% across nine backbone–benchmark combinations. The authors argue that this training fosters vigilance, enabling models to prioritize safety judgments before engaging in harmful planning.
By Youjia Wang, Lin Xu, Yang Sun, Yuxiao Lu, Chengfang Fang, Jie Shi
The Flag Game is a toy model designed to study how AI agents form collective beliefs. In the game, each agent sees only a private crop of a hidden country flag and can share beliefs with peers, leading to complex phenomena such as non‑monotonic performance scaling, accuracy gains from social awareness, and polarization that degrades performance at large population sizes. The authors introduce social circuit attribution to identify key agents and views, and develop a statistical mechanical theory to explain collective belief collapse and polarization in larger populations.
By Elizabeth Pavlova, Hidenori Tanaka
The paper introduces BLADE, a reliability‑aware method for selecting the boundary between spiking and artificial neural network components in hybrid event‑based object detectors. It jointly optimizes boundary placement and early‑exit settings for reliability, accuracy, execution time, and energy, using fault injection to guide design. Experiments show a 0.691 mAP@0.5 with 15.82 mJ energy when the ANN exits early, and that protecting a single floating‑point exponent bit eliminates catastrophic failures while a fully SNN configuration retains 96.5% reliability under severe faults.
By Mahdi Taheri, Alwin Paul
REVERSAL-BENCH is a benchmark that introduces a continuous reversibility parameter ρ∈[0,1] and a reset oracle to evaluate how well reinforcement learning agents can recover from irreversible states across eight manipulation tasks in five physics engines. Experiments show a sharp reversibility cliff: reset‑free agents become trapped in irrecoverable states as ρ increases, while episodic agents continue learning steadily. The benchmark also provides a large multi‑simulator dataset and demonstrates that safety shields can predict recoverability but only succeed when the agent can avoid the trap.
By Riyaaz Shaik, Chandru Venkataraman
The paper proposes a split‑control architecture for adaptive security at the network edge, where an untrusted planner emits typed security intents that are vetted by a deterministic governor before being enacted. The governor checks each intent against safety, resource, temporal‑stability, and proportionality invariants, issuing signed receipts for admitted actions that are compiled into eBPF map updates. Experiments on a Raspberry Pi 5 connected to a university 5G test network show the governor can admit, reject, and bound intents at microsecond cost without disrupting protected‑flow regularity.
By Ijaz Ahmad, Ijaz Ahmad, Flavio Esposito, Erkki Harjula
The paper introduces AWED, a word‑level emission‑delay metric, and demonstrates that post‑training a delayed‑streaming model with a joint reward (GRPO) improves both accuracy and latency. Using a single operating point (τ=6 frames), the method outperforms both its supervised baseline and the Voxtral Realtime backbone across all lookahead budgets, reducing WER by up to 30.8% at 80 ms delay and lowering median AWED from 1.17 s to 1.04 s.
By Xiuwen Zheng
The paper presents a physics-informed neural network (PINN) that accelerates multilayer spectral inversion (MLSI) of solar chromospheric lines Hα 6562.8 Å and Ca II 8542.1 Å. The PINN predicts MLSI parameters from observed line profiles and uses a differentiable forward model to synthesize spectra, trained in two stages—first with spectral reconstruction loss, then fine‑tuned with conventional MLSI results on a single reference image. Applied to Fast Imaging Solar Spectrograph data, the method reproduces key spatial structures and achieves a 12–60× speedup, processing a raster in 5–15 s versus 3–5 min for traditional MLSI.
By Ziyang Zhang, Qin Li, Vasyl B. Yurchyshyn, Kangwoo Yi, Haimin Wang, Wenda Cao, Bo Shen
The paper reports the first systematic audit of open‑weight large language models (LLMs) in hiring contexts, examining how job‑posting language influences recruiter and job‑seeker simulations across six models. It finds that agentic language lowers recruiter scores for female candidates while communal language mitigates this effect, and that coded‑exclusion language sharply reduces recruiter scores for non‑White candidates and discourages non‑White personas from applying. The study also identifies the explicit demographic label as the main causal factor and proposes a concrete pre‑deployment audit protocol aligned with EU and U.S. regulatory requirements.
By Kosuke Kitahara, Nobuhiro Yamaguchi
The paper surveys Generative Physical Artificial Intelligence (GPAI), a field where large foundation models are integrated with physical robots. It introduces a taxonomy of five approaches—Robot Foundation Models, Vision‑Language Action models, Large Behavior Models, Diffusion Policy Models, and World Foundation Models—and discusses how they complement each other across domains such as autonomous vehicles, industrial automation, healthcare robotics, and humanoid systems. The review highlights performance gains, data‑efficient learning, sim‑to‑real transfer, edge‑compatible architectures, and safety frameworks as key research directions.
By Satyam Gaba, Krutiksinh Rana, Siva Sai, Vinay Chamola, Dusit Niyato
The paper introduces a conditional variational autoencoder (CVAE) to generate synthetic electric vehicle (EV) charging sessions from real transaction-level data. It trains on engineered features such as plug‑in duration, charging duration, delivered energy, charging delay, and cyclical time‑of‑week, conditioning on day of week and managed charging status. Evaluation shows the synthetic data preserves key statistical properties and supports predictive modelling tasks via a Train‑on‑Synthetic‑Test‑on‑Real protocol.
By Graeme Kelly, Emilio J. Palacios-Garcia, Barry P. Hayes
F-DACE is a decision layer for causal machine‑learning‑based decision‑support systems that fuses multiple causal estimators using fuzzy memberships for precision, overlap, stability, and agreement, and abstains when diagnostics fail or evidence is weak. In simulations it made decisions in 67.2 % of runs while limiting false recommendations to 17.2 %, outperforming individual estimators. In a Walmart retail panel application, F‑DACE abstained on all five markdown indicators and, when integrated with a LangGraph conversational agent, achieved perfect tool‑routing accuracy and high groundedness on live queries.
By Sourish Dey
The paper investigates how a minority of biased agents in a multi‑agent system of large language models (LLMs) can amplify bias through textual interactions. Even a small percentage of persistently extreme agents causes significant opinion shifts among the non‑biased agents, with the effect occurring faster in the Llama 3.2 model than in a classical Friedkin‑Johnsen model. Semantic analysis shows that rhetorical consistency rises with biased exposure and that non‑biased agents adopt the biased vocabulary even when their numerical opinions change only modestly.
By Omran Berjawi, Giuseppe Fenza, Rida Khatoun
The paper introduces COMPASS-ABS, a scheduling framework for shared GPU clusters that reduces resource fragmentation for deep learning training jobs. It defines a new metric, Scheduler-Induced Fragmentation (SIF), which does not rely on historical workload data, and presents the COMPASS algorithm that confines cluster states within an Anchor-Based Space (ABS) to keep fragmentation low. Experiments on both a physical and a simulated cluster show that COMPASS-ABS improves resource utilization and shortens job completion times by mitigating fragmentation.
By Yukai Zhou, Hongfan Wu
The paper introduces Contrastive Noise Alignment (CNA), a training-time method for generative flow models that dynamically aligns Gaussian noise with data samples using a cross-modal InfoNCE objective. By modeling noise as an interacting particle system and regularizing with angular entropy and radial norm penalties, CNA reduces arbitrary data-noise couplings and flow curvature. Empirical results show that CNA improves generation quality, lowering FID by over 50% for few-step pixel-space generation compared to standard rectified flow and outperforming optimal transport baselines by at least 24%.
By Lennart Wittke, Vinicius Azevedo
The paper examines whether standard image‑fidelity metrics (SSIM, PSNR, MSE) accurately reflect the performance of GAN‑generated synthetic sonar data in robotic perception tasks. Using a Pix2Pix GAN with four discriminator configurations (PixelGAN, PatchGAN‑16, PatchGAN‑70, ImageGAN), the authors train object detectors (YOLOX‑S, YOLOX‑L, Faster R‑CNN) solely on real sonar images and evaluate them on the synthetic outputs. Results show a mismatch: the discriminator that yields the best pixel‑level scores does not always produce the best detection performance, with PatchGAN models achieving strong downstream results despite lower SSIM/PSNR/MSE values.
By Hannan Ejaz Keen, Muhammad Moazam Fraz, Karsten Berns
The paper introduces FairInference, a system that guarantees token-level latency isolation for well-behaved clients in multi-tenant LLM serving. It provides a δ-token fairness guarantee, ensuring that a token generated in isolation within time d will be produced within d + δ in a shared environment. The approach enforces per-token deadlines, bounds GPU compute sharing delays, and accounts for shared KV cache overhead, leading to reduced latency spikes and higher overall throughput compared to existing LLM serving systems.
By Dev Bali, Soujanya Ponnapalli, Yichuan Wang, Natacha Crooks, Scott Shenker, Matei Zaharia