Self-consistency boosts inference-time performance by sampling multiple reasoning traces in parallel and voting. However, in constrained domains like math and code, this strategy is compute-inefficient because it samples with replacement, repeatedly revisiting the same high-probability prefixes and duplicate completions. We propose Distinct Leaf Enumeration (DLE), a deterministic decoding method that treats truncated sampling as traversal of a pruned decoding tree and systematically enumerates distinct leaves instead of sampling with replacement. This strategy improves inference efficiency in two ways. Algorithmically, it increases coverage of the truncated search space under a fixed budget by exploring previously unvisited high-probability branches. Systemically, it reuses shared prefixes and reduces redundant token generation. Empirically, DLE explores higher-quality reasoning traces than stochastic self-consistency, yielding better performance on math, coding, and general reasoning tasks.

Long-term conversational large language model (LLM) agents require memory systems that can recover relevant evidence from historical interactions without overwhelming the answer stage with irrelevant context. However, existing memory systems, including hierarchical ones, still often rely solely on vector similarity for retrieval. It tends to produce bloated evidence sets: adding many superficially similar dialogue turns yields little additional recall, but lowers retrieval precision, increases answer-stage context cost, and makes retrieved memories harder to inspect and manage. To address this, we propose HiGMem (Hierarchical and LLM-Guided Memory System), a two-level event-turn memory system that allows LLMs to use event summaries as semantic anchors to predict which related turns are worth reading. This allows the model to inspect high-level event summaries first and then focus on a smaller set of potentially useful turns, providing a concise and reliable evidence set through reasoning, while avoiding the retrieval overhead that would be excessively high compared to vector retrieval. On the LoCoMo10 benchmark, HiGMem achieves the best F1 on four of five question categories and improves adversarial F1 from 0.54 to 0.78 over A-Mem, while retrieving an order of magnitude fewer turns. Code is publicly available at https://github.com/ZeroLoss-Lab/HiGMem.

Recent progress in Multimodal Large Language Models (MLLMs) has demonstrated remarkable advances in perception and reasoning, suggesting their potential for embodied intelligence. While recent studies have evaluated embodied MLLMs in interactive settings, current benchmarks mainly target capabilities to perceive, understand, and interact with external objects, lacking a systematic evaluation of self-centric intelligence. To address this, we introduce MirrorBench, a simulation-based benchmark inspired by the classical Mirror Self-Recognition (MSR) test in psychology. MirrorBench extends this paradigm to embodied MLLMs through a tiered framework of progressively challenging tasks, assessing agents from basic visual perception to high-level self-representation. Experiments on leading MLLMs show that even at the lowest level, their performance remains substantially inferior to human performance, revealing fundamental limitations in self-referential understanding. Our study bridges psychological paradigms and embodied intelligence, offering a principled framework for evaluating the emergence of general intelligence in large models. Project page: https://fflahm.github.io/mirror-bench-page/.

Overparameterized neural networks often show a benign overfitting property in the sense of achieving excellent generalization behavior despite the number of parameters exceeding the number of training examples. A promising direction to explain benign overfitting is to relate generalization to the norm of distance from initialization, motivated by the empirical observations that this distance is often significantly smaller than the norm itself. However, the existing initialization-dependent complexity analyses measure the distance from initialization by the Frobenius norm, and often imply vacuous bounds in practice for overparamterized models. In this paper, we develop initialization-dependent complexity bounds for shallow neural networks with general Lipschitz activation functions. Our bounds depend on the path-norm of the distance from initialization, which are derived by introducing a new peeling technique to handle the challenge along with the initialization-dependent constraint. We also develop a lower bound tight up to a constant factor. Finally, we conduct empirical comparisons and show that our generalization analysis implies non-vacuous bounds for overparameterized networks.

Recently, 3D Gaussian Splatting (3DGS) greatly accelerated mesh extraction from posed images due to its explicit representation and fast software rasterization. While the addition of geometric losses and other priors has improved the accuracy of extracted surfaces, mesh extraction remains difficult in scenes with abundant view-dependent effects. To resolve the resulting ambiguities, prior works rely on multi-view techniques, iterative mesh extraction, or large pre-trained models, sacrificing the inherent efficiency of 3DGS. In this work, we present a simple and efficient alternative by introducing a self-supervised confidence framework to 3DGS: within this framework, learnable confidence values dynamically balance photometric and geometric supervision. Extending our confidence-driven formulation, we introduce losses which penalize per-primitive color and normal variance and demonstrate their benefits to surface extraction. Finally, we complement the above with an improved appearance model, by decoupling the individual terms of the D-SSIM loss. Our final approach delivers state-of-the-art results for unbounded meshes while remaining highly efficient.

The widespread adoption of Large Language Models (LLMs) has made the detection of AI-Generated text a pressing and complex challenge. Although many detection systems report high benchmark accuracy, their reliability in real-world settings remains uncertain, and their interpretability is often unexplored. In this work, we investigate whether contemporary detectors genuinely identify machine authorship or merely exploit dataset-specific artefacts. We propose an interpretable detection framework that integrates linguistic feature engineering, machine learning, and explainable AI techniques. When evaluated on two prominent benchmark corpora, namely PAN CLEF 2025 and COLING 2025, our model trained on 30 linguistic features achieves leaderboard-competitive performance, attaining an F1 score of 0.9734. However, systematic cross-domain and cross-generator evaluation reveals substantial generalisation failure: classifiers that excel in-domain degrade significantly under distribution shift. Using SHAP- based explanations, we show that the most influential features differ markedly between datasets, indicating that detectors often rely on dataset-specific stylistic cues rather than stable signals of machine authorship. Further investigation with in-depth error analysis exposes a fundamental tension in linguistic-feature-based AI text detection: the features that are most discriminative on in-domain data are also the features most susceptible to domain shift, formatting variation, and text-length effects. We believe that this knowledge helps build AI detectors that are robust across different settings. To support replication and practical use, we release an open-source Python package that returns both predictions and instance-level explanations for individual texts.

The growing use of Machine Learning (ML) tools comes with critical challenges, such as limited model explainability. We propose a global explainability framework that leverages Optimal Transport and Distributionally Robust Optimization to analyze how ML algorithms respond to constrained data perturbations. Our approach enforces constraints on feature-level statistics (e.g., brightness, age distribution), generating realistic perturbations that preserve semantic structure. We provide a model-agnostic diagnostic bench that applies to both tabular and image domains with solid theoretical guarantees. We validate the approach on real-world datasets providing interpretable robustness diagnostics that complement standard evaluation and fairness auditing tools.

Differential medical VQA models compare multiple images to identify clinically meaningful changes and rely on vision encoders to capture fine-grained visual differences that reflect radiologists' comparative diagnostic workflows. However, vision encoders trained using standard contrastive or classification objectives often fail to capture the subtle variations needed to distinguish true disease progression from acquisition-related variability. To address this limitation, we introduce a location-aware pretraining framework that incorporates automatic referring expressions (AREF), grounded captioning (GCAP), and conditional automatic referring expressions (CAREF). These tasks promote the learning of fine-grained, spatially grounded visual representations. When integrated with a language model, our approach achieves state-of-the-art performance on medical difference VQA by accurately identifying and reasoning about clinically relevant changes in chest X-ray images.

Spatiotemporal predictive learning (STPL) aims to forecast future frames from past observations and is essential across a wide range of applications. Compared with recurrent or hybrid architectures, pure convolutional models offer superior efficiency and full parallelism, yet their fixed receptive fields limit their ability to adaptively capture spatially varying motion patterns. Inspired by biological center-surround organization and frequency-selective signal processing, we propose PFGNet, a fully convolutional framework that dynamically modulates receptive fields through pixel-wise frequency-guided gating. The core Peripheral Frequency Gating (PFG) block extracts localized spectral cues and adaptively fuses multi-scale large-kernel peripheral responses with learnable center suppression, effectively forming spatially adaptive band-pass filters. To maintain efficiency, all large kernels are decomposed into separable 1D convolutions ($1 \times k$ followed by $k \times 1$), reducing per-channel computational cost from $O(k^2)$ to $O(2k)$. PFGNet enables structure-aware spatiotemporal modeling without recurrence or attention. Experiments on Moving MNIST, TaxiBJ, Human3.6M, and KTH show that PFGNet delivers SOTA or near-SOTA forecasting performance with substantially fewer parameters and FLOPs. Our code is available at https://github.com/fhjdqaq/PFGNet.

Large-scale verifiable prompts underpin the success of Reinforcement Learning with Verifiable Rewards (RLVR), but they contain many uninformative examples and are costly to expand further. Recent studies focus on better exploiting limited training data by prioritizing hard prompts whose rollout pass rate is 0. However, easy prompts with a pass rate of 1 also become increasingly prevalent as training progresses, thereby reducing the effective data size. To mitigate this, we propose Composition-RL, a simple yet useful approach for better utilizing limited verifiable prompts targeting pass-rate-1 prompts. More specifically, Composition-RL automatically composes multiple problems into a new verifiable question and uses these compositional prompts for RL training. Extensive experiments across model sizes from 4B to 30B show that Composition-RL consistently improves reasoning capability over RL trained on the original dataset. Performance can be further boosted with a curriculum variant of Composition-RL that gradually increases compositional depth over training. Additionally, Composition-RL enables more effective cross-domain RL by composing prompts drawn from different domains. Codes, datasets, and models are available at https://github.com/XinXU-USTC/Composition-RL.

Multilingual retrieval-augmented generation (MRAG) requires models to effectively acquire and integrate beneficial external knowledge from multilingual collections. However, most existing studies employ a unitive process where queries of equivalent semantics across different languages are processed through a single-turn retrieval and subsequent optimization. Such a ``one-size-fits-all'' strategy is often suboptimal in multilingual settings, as the models occur to knowledge bias and conflict during the interaction with the search engine. To alleviate the issues, we propose LcRL, a multilingual search-augmented reinforcement learning framework that integrates a language-coupled Group Relative Policy Optimization into the policy and reward models. We adopt the language-coupled group sampling in the rollout module to reduce knowledge bias, and regularize an auxiliary anti-consistency penalty in the reward models to mitigate the knowledge conflict. Experimental results demonstrate that LcRL not only achieves competitive performance but is also appropriate for various practical scenarios such as constrained training data and retrieval over collections encompassing a large number of languages. Our code is available at https://github.com/Cherry-qwq/LcRL-Open.

Autoregressive LLMs perform well on relational tasks that require linking entities via relational words (e.g., father/son, friend), but it is unclear whether they learn the logical semantics of such relations (e.g., symmetry and inversion logic) and, if so, whether reversal-type failures arise from missing relational semantics or left-to-right order bias. We propose a controlled Knowledge Graph-based synthetic framework that generates text from symmetric/inverse triples, train GPT-style autoregressive models from scratch, and evaluate memorization, logical inference, and in-context generalization to unseen entities to address these questions. We find a sharp phase transition in which relational semantics emerge with sufficient logic-bearing supervision, even in shallow (2-3 layer) models, and that successful generalization aligns with stable intermediate-layer signals. Finally, order-matched forward/reverse tests and a diffusion baseline indicate that reversal failures are primarily driven by autoregressive order bias rather than deficient inversion semantics.

Controlling emergent behavioral personas (e.g., sycophancy, hallucination) in Large Language Models (LLMs) is critical for AI safety, yet remains a persistent challenge. Existing solutions face a dilemma: manual prompt engineering is intuitive but unscalable and imprecise, while automatic optimization methods are effective but operate as "black boxes" with no interpretable connection to model internals. We propose a novel framework that adapts gradient ascent to LLMs, enabling targeted prompt discovery. In specific, we propose two methods, RESGA and SAEGA, that both optimize randomly initialized prompts to achieve better aligned representation with an identified persona direction. We introduce fluent gradient ascent to control the fluency of discovered persona steering prompts. We demonstrate RESGA and SAEGA's effectiveness across Llama 3.1, Qwen 2.5, and Gemma 3 for steering three different personas, sycophancy, hallucination, and myopic reward. Crucially, on sycophancy, our automatically discovered prompts achieve significant improvement (49.90% compared with 79.24%). By grounding prompt discovery in mechanistically meaningful features, our method offers a new paradigm for controllable and interpretable behavior modification.

Human Label Variation (HLV) refers to legitimate disagreement in annotation that reflects the diversity of human perspectives rather than mere error. Long treated in NLP as noise to be eliminated, HLV has only recently been reframed as a signal for improving model robustness. With the rise of large language models (LLMs) and post-training methods such as human feedback-based alignment, the role of HLV has become increasingly consequential. Yet current preference-learning datasets routinely collapse multiple annotations into a single label, flattening diverse perspectives into artificial consensus. Preserving HLV is necessary not only for pluralistic alignment but also for sociotechnical safety evaluation, where model behavior must be assessed in relation to human interaction and societal context. This position paper argues that preserving HLV as an embodiment of human pluralism must be treated as a Selbstzweck, an intrinsic value in itself. We analyze the limitations of existing preference datasets and propose actionable strategies for incorporating HLV into dataset construction to better preserve pluralistic human values.

Masked diffusion models have shown promising performance in generating high-quality samples in a wide range of domains, but accelerating their sampling process remains relatively underexplored. To investigate efficient samplers for masked diffusion, this paper theoretically analyzes the MaskGIT sampler for image modeling, revealing its implicit temperature sampling mechanism. Through this analysis, we introduce the "moment sampler," an asymptotically equivalent but more tractable and interpretable alternative to MaskGIT, which employs a "choose-then-sample" approach by selecting unmasking positions before sampling tokens. In addition, we improve the efficiency of choose-then-sample algorithms through two key innovations: a partial caching technique for transformers that approximates longer sampling trajectories without proportional computational cost, and a hybrid approach formalizing the exploration-exploitation trade-off in adaptive unmasking. Experiments in image and text domains demonstrate our theory as well as the efficiency of our proposed methods, advancing both theoretical understanding and practical implementation of masked diffusion samplers.

Predictive process monitoring supports operational decision-making by forecasting future states of ongoing business cases. A key task is case suffix prediction, which estimates the remaining sequence of activities for a case. Most existing approaches only generate activities with a single timestamp (usually the completion time). However, this is insufficient for resource capacity planning, which requires distinguishing between waiting time and processing time to accurately schedule resources and manage workloads. This paper introduces a technique to predict case suffixes that include both start and end timestamps. By predicting distinct waiting and processing intervals, the method provides a more granular view of future resource demands.

The rapid development of generative AI has transformed content creation, communication, and human development. However, this technology raises profound concerns in high-stakes domains, demanding rigorous methods to analyze and evaluate AI-generated content. While existing analytic methods often treat images as indivisible wholes, real-world AI failures generally manifest as specific visual patterns that can evade holistic detection and suit more granular and decomposed analysis. Here we introduce a content analysis tool, Language-Grounded Sparse Encoders (LanSE), which decompose images into interpretable visual patterns with natural language descriptions. Utilizing interpretability modules and large multimodal models, LanSE can automatically identify visual patterns within data modalities. Our method discovers more than 5,000 visual patterns with 93\% human agreement, provides decomposed evaluation outperforming existing methods, establishes the first systematic evaluation of physical plausibility, and extends to medical imaging settings. Our method's capability to extract language-grounded patterns can be naturally adapted to numerous fields, including biology and geography, as well as other data modalities such as protein structures and time series, thereby advancing content analysis for generative AI.

Graph Neural Networks (GNNs) have emerged as powerful tools for learning over structured data, including text-attributed graphs (TAGs), which are common in domains such as citation networks, social platforms, and knowledge graphs. GNNs are not inherently interpretable and thus, many explanation methods have been proposed. However, existing explanation methods often struggle to generate interpretable, fine-grained rationales, especially when node attributes include rich natural language. In this work, we introduce GSPELL, a lightweight, post-hoc framework that uses large language models (LLMs) to generate faithful and interpretable explanations for GNN predictions. GSPELL projects GNN node embeddings into the LLM embedding space and constructs hybrid prompts that interleave soft prompts with textual inputs from the graph structure. This enables the LLM to reason about GNN internal representations and to produce natural-language explanations, along with concise explanation subgraphs. Our experiments across real-world TAG datasets demonstrate that GSPELL achieves a favorable trade-off between fidelity and sparsity, while improving human-centric metrics such as insightfulness. GSPELL sets a new direction for LLM-based explainability in graph learning by aligning GNN internals with human reasoning.

Human intelligence scales through cumulative cultural evolution (CCE), a ratchet process in which innovations are retained against entropic drift. Large language model training, by contrast, still depends primarily on static corpora and parameter growth, leaving little room for endogenous accumulation through interaction. We present POLIS (Population Orchestrated Learning and Inference Society), a framework in which heterogeneous agents generate solutions, verify one another's outputs, retain validated artifacts in shared cultural memory, and internalize them through parameter updates. On mathematical reasoning benchmarks, populations of 1--4B-parameter models achieved average gains of 8.8--18.9 points over base models and narrowed the gap to 70B+ monoliths. Mechanistic ablations identify peer verification as the main ratchet operator and show that internalization sustains accumulation across rounds, providing computational evidence that epistemic vigilance organizes durable knowledge growth. These results position structured social interaction as a scaling lever orthogonal to parameter count.

Previous work on the learnability of transformers \textemdash\ focused primarily on examining their ability to approximate specific algorithmic patterns through training \textemdash\ has largely been data-driven, offering only probabilistic guarantees rather than deterministic solutions. Expressivity, on the contrary, has been devised to address the problems \emph{computable} by such architecture theoretically. These results proved the Turing-completeness of transformers, investigated bounds focused on circuit complexity, and formal logic. Being at the crossroad between learnability and expressivity, the question remains: \emph{can a transformer, as a computational model, simulate an arbitrary attention mechanism, or in particular, the underlying operations?} In this study, we investigate the transformer encoder's ability to simulate a vanilla attention mechanism. By constructing a universal simulator $\mathcal{U}$ composed of transformer encoders, we present algorithmic solutions to replicate attention outputs and the underlying elementary matrix and activation operations via RASP, a formal framework for transformer computation. We show the existence of an algorithmically achievable, data-agnostic solution, previously known to be approximated only by learning.

The rapid development of Artificial Intelligence Generated Content (AIGC) techniques has enabled the creation of high-quality synthetic content, but it also raises significant security concerns. Current detection methods face two major limitations: (1) the lack of multidimensional explainable datasets for generated images and videos. Existing open-source datasets (e.g., WildFake, GenVideo) rely on oversimplified binary annotations, which restrict the explainability and trustworthiness of trained detectors. (2) Prior MLLM-based forgery detectors (e.g., FakeVLM) exhibit insufficiently fine-grained interpretability in their step-by-step reasoning, which hinders reliable localization and explanation. To address these challenges, we introduce Ivy-Fake, the first large-scale multimodal benchmark for explainable AIGC detection. It consists of over 106K richly annotated training samples (images and videos) and 5,000 manually verified evaluation examples, sourced from multiple generative models and real world datasets through a carefully designed pipeline to ensure both diversity and quality. Furthermore, we propose Ivy-xDetector, a reinforcement learning model based on Group Relative Policy Optimization (GRPO), capable of producing explainable reasoning chains and achieving robust performance across multiple synthetic content detection benchmarks. Extensive experiments demonstrate the superiority of our dataset and confirm the effectiveness of our approach. Notably, our method improves performance on GenImage from 86.88% to 96.32%, surpassing prior state-of-the-art methods by a clear margin.

Consistent maps are key for most autonomous mobile robots, and they often use SLAM approaches to build such maps. Loop closures via place recognition help to maintain accurate pose estimates by mitigating global drift, and are thus key for realizing an effective SLAM system. This paper presents a robust loop closure detection pipeline for outdoor SLAM with LiDAR-equipped robots. Our method handles various LiDAR sensors with different scanning patterns, fields of view, and resolutions. It generates local maps from LiDAR scans and aligns them using a ground alignment module to handle both planar and non-planar motion of the LiDAR, ensuring applicability across platforms. The method uses density-preserving bird's-eye-view projections of these local maps and extracts ORB feature descriptors for place recognition. It stores the feature descriptors in a binary search tree for efficient retrieval, and self-similarity pruning addresses perceptual aliasing in repetitive environments. Extensive experiments on public and self-recorded datasets demonstrate accurate loop closure detection, long-term localization, and cross-platform multi-map alignment, agnostic to the LiDAR scanning patterns, fields of view, and motion profiles. We provide the code for our pipeline as open-source software at https://github.com/PRBonn/MapClosures.

Animal excretions in form of urine puddles and feces are a significant source of emissions in livestock farming. Automated detection of soiled floor in barns can contribute to improved management processes but also the derived information can be used to model emission dynamics. Previous research approaches to determine the puddle area require manual detection of the puddle in the barn. While humans can detect animal excretions on thermal images of a livestock barn, automated approaches using thresholds fail due to other objects of the same temperature, such as the animals themselves. In addition, various parameters such as the type of housing, animal species, age, sex, weather and unknown factors can influence the type and shape of excretions. Due to this heterogeneity, a method for automated detection of excretions must therefore be not only be accurate but also robust to varying conditions. These requirements can be met by using contemporary deep learning models from the field of artificial intelligence. This work is the first to investigate the suitability of different deep learning models for the detection of excretions in pigsties, thereby comparing established convolutional architectures with recent transformer-based approaches. The detection models Faster R-CNN, YOLOv8, DETR and DAB-DETR are compared and statistically assessed on two created training datasets representing two pig houses. We apply a method derived from nested cross-validation and report on the results in terms of eight common detection metrics. Our work demonstrates that all investigated deep learning models are generally suitable for reliably detecting excretions with an average precision of over 90%. The models also show robustness on out of distribution data that possesses differences from the conditions in the training data, however, with expected slight decreases in the overall detection performance.

Detecting and quantifying causality is a focal topic in the fields of science, engineering, and interdisciplinary studies. However, causal studies on non-intervention systems attract much attention but remain extremely challenging. Delay-embedding technique provides a promising approach. In this study, we propose a framework named Interventional Dynamical Causality (IntDC) in contrast to the traditional Constructive Dynamical Causality (ConDC). ConDC, including Granger causality, transfer entropy and convergence of cross-mapping, measures the causality by constructing a dynamical model without considering interventions. A computational criterion, Interventional Embedding Entropy (IEE), is proposed to measure causal strengths in an interventional manner. IEE is an intervened causal information flow but in the delay-embedding space. Further, the IEE theoretically and numerically enables the deciphering of IntDC solely from observational (non-interventional) time-series data, without requiring any knowledge of dynamical models or real interventions in the considered system. In particular, IEE can be applied to rank causal effects according to their importance and construct causal networks from data. We conducted numerical experiments to demonstrate that IEE can find causal edges accurately, eliminate effects of confounding, and quantify causal strength robustly over traditional indices. We also applied IEE to real-world tasks. IEE performed as an accurate and robust tool for causal analyses solely from the observational data. The IntDC framework and IEE algorithm provide an efficient approach to the study of causality from time series in diverse non-intervention complex systems.

Reward evaluation of episodes becomes a bottleneck in a broad range of reinforcement learning tasks. Our aim in this paper is to select a small but representative subset of a large batch of episodes, only on which we actually compute rewards for more efficient policy gradient iterations. We build a Gaussian process modeling of discounted returns or rewards to derive a positive definite kernel on the space of episodes, run an ``episodic" kernel quadrature method to compress the information of sample episodes, and pass the reduced episodes to the policy network for gradient updates. We present the theoretical background of this procedure as well as its numerical illustrations in MuJoCo tasks.

Reliable crack detection and segmentation are vital for structural health monitoring, yet the scarcity of well-annotated data constitutes a major challenge. To address this limitation, we propose a novel context-aware generative framework designed to synthesize realistic crack growth patterns for data augmentation. Unlike existing methods that primarily manipulate textures or background content, CrackForward explicitly models crack morphology by combining directional crack elongation with learned thickening and branching. Our framework integrates two key innovations: (i) a contextually guided crack expansion module, which uses local directional cues and adaptive random walk to simulate realistic propagation paths; and (ii) a two-stage U-Net-style generator that learns to reproduce spatially varying crack characteristics such as thickness, branching, and growth. Experimental results show that the generated samples preserve target-stage saturation and thickness characteristics and improve the performance of several crack segmentation architectures. These results indicate that structure-aware synthetic crack generation can provide more informative training data than conventional augmentation alone.

Attitude change - the process by which individuals revise their evaluative stances - has been explained by a set of influential but competing verbal theories. These accounts often function as mechanism sketches: rich in conceptual detail, yet lacking the technical specifications and operational constraints required to run as executable systems. We present a generative actor-based modelling workflow for "rendering" these sketches as runnable actor - environment simulations using the Concordia simulation library. In Concordia, actors operate by predictive pattern completion: an operation on natural language strings that generates a suffix which describes the actor's intended action from a prefix containing memories of their past and observations of the present. We render the theories of cognitive dissonance (Festinger 1957), self-consistency (Aronson 1969), and self-perception (Bem 1972) as distinct decision logics that populate and process the prefix through theory-specific sequences of reasoning steps. We evaluate these implementations across classic psychological experiments. Our implementations generate behavioural patterns consistent with known results from the original empirical literature. However, we find that achieving stable reproduction requires resolving the inherent underdetermination of the verbal accounts and the conflicts between modern linguistic priors and historical experimental assumptions. And, we document how this manual process of iterative model "stabilisation" surfaces specific operational and socio-ecological dependencies that were largely undocumented in the original verbal accounts. Ultimately, we argue that the manual stabilisation process itself should be regarded as a core part of the methodology functioning to clarify situational and representational commitments needed to generate characteristic effects.

The April 2026 Claude Mythos sandbox escape exposed a critical weakness in frontier AI containment: the infrastructure surrounding advanced models remains susceptible to formally characterizable arithmetic vulnerabilities. Anthropic has not publicly characterized the escape vector; some secondary accounts hypothesize a CWE-190 arithmetic vulnerability in sandbox networking code. We treat this as unverified and analyze the vulnerability class rather than the specific escape. This paper presents COBALT, a Z3 SMT-based formal verification engine for identifying CWE-190/191/195 arithmetic vulnerability patterns in C/C++ infrastructure prior to deployment. We distinguish two classes of contribution. Validated: COBALT detects arithmetic vulnerability patterns in production codebases, producing SAT verdicts with concrete witnesses and UNSAT guarantees under explicit safety bounds. We demonstrate this on four production case studies: NASA cFE, wolfSSL, Eclipse Mosquitto, and NASA F Prime, with reproducible encodings, verified solver output, and acknowledged security outcomes. Proposed: a four-layer containment framework consisting of COBALT, VERDICT, DIRECTIVE-4, and SENTINEL, mapping pre-deployment verification, pre-execution constraints, output control, and runtime monitoring to the failure modes exposed by the Mythos incident. Under explicit assumptions, we further argue that the publicly reported Mythos escape class is consistent with a Z3-expressible CWE-190 arithmetic formulation and that pre-deployment formal analysis would have been capable of surfacing the relevant pattern. The broader claim is infrastructural: frontier-model safety cannot depend on behavioral safeguards alone; the containment stack itself must be subjected to formal verification.

Machine learning-based static malware detectors remain vulnerable to adversarial evasion techniques, such as metamorphic engine mutations. To address this vulnerability, we propose a certifiably robust malware detection framework based on randomized smoothing through feature ablation and targeted noise injection. During evaluation, our system analyzes an executable by generating multiple ablated variants, classifies them by using a smoothed classifier, and identifies the final label based on the majority vote. By analyzing the top-class voting distribution and the Wilson score interval, we derive a formal certificate that guarantees robustness within a specific radius against feature-space perturbations. We evaluate our approach by comparing the performance of the base classifier and the smoothed classifier on both clean executables and ablated variants generated using PyMetaEngine. Our results demonstrate that the proposed smoothed classifier successfully provides certifiable robustness against metamorphic evasion attacks without requiring modifications to the underlying machine learning architecture.

In this paper, it is shown, for the first time, that centralized performance is achievable in decentralized learning without sharing the local datasets. Specifically, when clients adopt an empirical risk minimization with relative-entropy regularization (ERM-RER) learning framework and a forward-backward communication between clients is established, it suffices to share the locally obtained Gibbs measures to achieve the same performance as that of a centralized ERM-RER with access to all the datasets. The core idea is that the Gibbs measure produced by client~$k$ is used, as reference measure, by client~$k+1$. This effectively establishes a principled way to encode prior information through a reference measure. In particular, achieving centralized performance in the decentralized setting requires a specific scaling of the regularization factors with the local sample sizes. Overall, this result opens the door to novel decentralized learning paradigms that shift the collaboration strategy from sharing data to sharing the local inductive bias via the reference measures over the set of models.