arXiv:2607.09759v2 Announce Type: replace
Abstract: Building assistants that can continually watch the world, remember what they see, and reason over their accumulated experience is a long-standing goal, and recently multimodal agents equipped with long-term memory over video streams have attracted increasing interest. Unfortunately, existing systems either keep their memory inside the model context or in a flat feature store, and organize it around frames rather than around the persistent entities a stream is really about, which confines them to bounded videos and weakens their ability to track who and what reappears over time. In this paper, we propose ReflectWorld-MM, an entity-oriented multimodal memory system for open-ended video streams. It consists of three parts. The first is a perception front-end that turns an audiovisual stream into entity-resolved observations under a bounded short-term memory. The second is a hierarchical long-term memory, grounded in human memory theory, that couples a multi-scale episodic memory, an evolving entity-centric semantic memory, and a procedural memory. The third is a complete realization, built for real-world operation, that ingests arbitrary streams and plugs into off-the-shelf assistants. Across six long-video and lifelong-memory benchmarks, ReflectWorld-MM achieves the best accuracy on all six, outperforming strong memory agents and a frontier model.
Science Journals
arXiv:2607.10744v2 Announce Type: replace
Abstract: Benefiting from the powerful priors embedded in large-scale pre-training data and the emerging commonsense reasoning ability, large language models (LLMs) have shown unprecedented generalization capabilities in many research fields. Recently, projecting visual embeddings into the language space via vision-language models (VLMs) to achieve sim-toreal and cross-scene generalization has become a prevailing paradigm in the field of Vision-and-Language Navigation in Continuous Environments (VLN-CE). VLN requires an embodied agent to navigate through unseen environments following natural linguistic instructions. We emphasize that a VLN task can be decomposed into a sequence of sub-tasks, each corresponding to a process of 3D spatial interaction with the environments described by instructions such as "walk to the end of the sofa and turn left." However, such spatial interactions involving moving into the image along the direction of depth sensing are puzzling for VLMs as they were predominantly trained on conversations with RGB images. Rather than incorporating depth or 3D geometric information-which VLMs rarely encounter during pretrainingwe propose an alternative approach: fine-tuning VLMs to learn navigation interactions directly in 2D pixel space through autoregressive trajectory generation. Given a linguistic instruction and historical observations, our model sequentially predicts a series of pixel coordinates, drawing a trajectory from the bottom center of the current observation. While prior work has proved that pixel-goal supervision outperforms learning of discrete actions, our experiments further verify that the supervision of pixel-space trajectory significantly enhances VLN performance. Moreover, we demonstrate that our flagship model achieves state-of-the-art level performance with relatively limited computational resources and training data.
arXiv:2607.11638v2 Announce Type: replace
Abstract: City-scale outdoor navigation is currently hindered by the heavy reliance on dense maps or costly navigation supervision. In this work, we introduce a novel paradigm for leveraging directional instructions from commercial navigation tools (e.g., Google Maps). To bridge the gap between commercial instructions and executable navigation actions, while mitigating long-horizon error accumulation through robust trajectory recovery, we propose DA-Nav, a Direction-Aware vision-language Navigation framework that reformulates navigation as a discrete spatial grounding problem on the egocentric 2D image plane. To achieve trajectory recovery, DA-Nav employs a Chain-of-Thought (CoT) reasoning process encompassing deviation assessment, action prediction, and target grid selection. We further introduce ReDA, a dataset that provides direction-aware instructions and recovery trajectories to enhance spatial grounding and support CoT recovery reasoning. Extensive experiments in CARLA demonstrate that DA-Nav achieves a high success rate of 56.16% in unseen urban environments, outperforming existing State-of-The-Art (SoTA) methods while maintaining a substantially stronger recovery capability. Furthermore, without fine-tuning, DA-Nav seamlessly adapts to both quadruped and humanoid robots, enabling stable kilometer-scale closed-loop outdoor navigation in complex real world environments.
arXiv:2411.03423v2 Announce Type: replace-cross
Abstract: Quantum optics routinely uses beam splitters to generate entanglement, including in pioneering experiments conducted by Hanbury-Brown and Twiss and Hong, Ou, and Mandel. The quantum interference at beam splitters lies at the heart of what makes boson sampling hard to emulate by classical computers and is a vital component of quantum computation with light. Yet, despite overwhelming positive evidence, the conjecture that beam splitters with equal reflection and transmission probabilities generate the most entanglement for any state interfered with the vacuum has remained unproven for almost two decades [Asb\'oth et al., Phys. Rev. Lett. 94, 173602 (2005)]. We prove this conjecture for ubiquitous entanglement monotones including mixed-state generalizations of entanglement entropy and purity by uncovering monotonicity and convexity with respect to photon loss for these monotones. At the same time, we highlight an infinite class of lesser-used monotones for which the conjecture fails. Because beam splitters are so fundamental, our results yield numerous corollaries for quantum optics, including proof of a recent conjecture for the evolution of a measure of quantumness through loss and a more efficient computational strategy for optimizing entanglement generation over linear optics. These results justify the value of seeking mathematical rigour behind commonly accepted facts and the danger of trusting them unconditionally.
arXiv:2607.03581v2 Announce Type: replace
Abstract: The widespread adoption of facial masks, accelerated by COVID-19 and mandated in security-sensitive settings, has exposed limitations of conventional face recognition systems. Existing approaches relying on fixed cosine thresholds, non-adaptive CNNs, and purely data-driven features fail to generalize when facial regions are occluded, creating a gap between lab performance and real-world deployability. This paper proposes PLGSA-Transformer, a cross-modal face matching framework with three contributions. First, Periocular Landmark-Guided Spatial Attention (PLGSA) uses MediaPipe landmarks to compute Gaussian heatmaps over the eye, brow, and forehead regions, fusing them with EfficientNetB3 features via a learnable residual gate to direct attention toward discriminative visible regions. Second, a Hybrid CNN-Transformer Branch reshapes feature maps into tokens processed by a two-layer Multi-Head Self-Attention encoder, enabling cross-regional dependency modelling. Third, the Occlusion-Adaptive Cosine Threshold (OACT) is a jointly trained head that raises the matching threshold in proportion to predicted occlusion severity. The model is evaluated on 858 images from Zenodo MDMFR (60%), Kaggle CelebA-HQ masked collection (25%), and author-collected images (15%), spanning both genders, ages 21-75, with varied mask types, trained via a unified loss combining contrastive verification, identity classification, and occlusion cross-entropy. PLGSA-Transformer achieves 97.22% pair verification accuracy with ROC AUC 1.0000, surpassing VGG-16-based MUFM (Abdullah et al., 2025; 95.0%), HOG classifiers (Adnan et al., 2020; 85.0%), and Feature-based Structural Measure (Shnain et al., 2017; 86.61%). These results confirm that encoding periocular geometry into attention, with Transformer modelling and occlusion-adaptive thresholds, yields a robust, scalable solution for cross-modal masked face recognition.
arXiv:2607.10467v2 Announce Type: replace
Abstract: Healthcare organizations often cannot freely centralize patient data because medical records are sensitive, regulated, and institutionally controlled. Federated learning offers a practical alternative by allowing hospitals and clinics to train a shared model while keeping raw data local. However, federated learning is not automatically production-ready or private by default. Model updates can still leak information, and decentralized training introduces operational challenges in deployment, monitoring, rollback, debugging, and governance. This paper examines how MLOps practices and the emerging idea of Federated Learning Operations (FLOps) can make federated healthcare machine learning systems scalable, reliable, and trustworthy. It answers three research questions: how containerization and orchestration support federated deployment, how privacy-preserving mechanisms affect trade-offs among privacy, utility, scalability, and operational complexity, and which post-deployment practices are most important for long-term governance. The central argument is that federated healthcare ML requires more than privacy-preserving algorithms. It needs an integrated MLOps architecture that combines reproducible deployment, secure orchestration, model versioning, audit logging, drift monitoring, heterogeneity management, and clear governance.
arXiv:2601.10494v3 Announce Type: replace-cross
Abstract: With grid operators confronting rising uncertainty from renewable integration and a broader push toward electrification, Demand-Side Management (DSM) -- particularly Demand Response (DR) -- has attracted significant attention as a cost-effective mechanism for balancing modern electricity systems. Unprecedented volumes of consumption data from a continuing global deployment of smart meters enable consumer segmentation based on real usage behaviours, promising to inform the design of more effective DSM and DR programs. However, existing clustering-based segmentation methods insufficiently reflect the behavioural diversity of consumers, often relying on rigid temporal alignment, and faltering in the presence of anomalies, missing data, or large-scale deployments.
To address these challenges, we propose a novel two-stage clustering framework -- Clustered Representations Optimising Consumer Segmentation (CROCS). In the first stage, each consumer's daily load profiles are clustered independently to form a Representative Load Set (RLS), providing a compact summary of their typical diurnal consumption behaviours. In the second stage, consumers are clustered using the Weighted Sum of Minimum Distances (WSMD), a novel set-to-set measure that compares RLSs by accounting for both the prevalence and similarity of those behaviours. Finally, community detection on the WSMD-induced graph reveals higher-order prototypes that embody the shared diurnal behaviours defining consumer groups, enhancing the interpretability of the resulting clusters.
Extensive experiments on both synthetic and real Australian smart meter datasets demonstrate that CROCS captures intra-consumer variability, uncovers both synchronous and asynchronous behavioural similarities, and remains robust to anomalies and missing data, while scaling efficiently through natural parallelisation. These results...
arXiv:2607.10560v2 Announce Type: replace
Abstract: Mesh deformation, the process of altering the vertex positions of a 3D mesh while preserving its topological structure, is a cornerstone of computer graphics. Despite the recent emergence of numerous text-guided 3D mesh deformation methods, deforming an initial mesh into one that both adheres to text prompts and preserves its pose remains challenging. This paper proposes PoseAlign, which decomposes text-guided mesh deformation into two stages: global pose scaling and local detail sculpting. Specifically, in the first stage, we introduce the Laplacian as a differentiable mesh representation to enable more efficient yet smoother global deformation. Then, we propose a novel pose-aligned SDS loss by adapting score distillation sampling (SDS) with an attention-sharing mechanism, which sculptures fine-grained geometric details for the deformed mesh while preserving its original pose. PoseAlign significantly enhances the controllability of the overall deformation process, achieving a favorable balance between pose preservation and text alignment. Experiments demonstrate the competitive advantages of our method in text alignment and mesh quality. Code is available at: https://cousingrade6.github.io/PoseAlign
arXiv:2607.11020v2 Announce Type: replace
Abstract: Continual learning promises a language model that keeps acquiring knowledge after training, with each new fact written into its weights. Whether weight writes can support accumulation remains undecided. We follow invented facts written into Qwen3 models from creation through sequences of twenty to one hundred later writes, using held-out questions of five types, with the original model given the fact in its prompt as the reference. Across these experiments, the breadth of the training data determines the kind of knowledge created. Bare-statement training produces recitation, while diverse restatements reduce the recitation-to-use gap from 27.4 to 5.4 points without showing the model a conclusion. This difference carries into later writes: after twenty sequential writes, bare-statement facts retain 1% accuracy while facts written from broad study data retain 46%. We also find that facts can be behaviourally forgotten without being erased. Forgotten facts keep most of the log-probability added by their write, and under bare-statement training 70% of wrong answers about them contain the most recently written fact. The same writes barely degrade the model's use of facts in context, and a forgotten study fact supplied in the prompt recovers to 77-80% on its questions. These results describe knowledge that is stored but question-keyed: later writes redirect the questions that reached it. Damage to unrelated abilities tracks KL divergence from the original model, and the later writes cause interference regardless of how the earlier fact was stored. Broad data can create usable knowledge, and a frozen reference can preserve capability, but no intervention we tested, including those built on accurate local measurements of each write, keeps earlier facts reachable. When facts must be composed or survive later writes, the reliable channel is context rather than the weights.
arXiv:2607.11193v2 Announce Type: replace
Abstract: To ensure the overall quality of AI-enabled software, not only traditional software components but also AI components need to be tested and repaired. Among AI components, Transformer models are increasingly integrated into software systems, which makes their misbehaviors critical. Although prior work in the software engineering community has proposed deep neural network (DNN) repair methods, most overlook Transformer-specific structures. We propose RepTran, a search-based repair method for Transformer models. It targets their feed-forward networks (FFNs), which play a central role in the architecture. RepTran identifies suspicious weights by combining two types of scores: a variance-based neuron score and an existing bidirectional score. It then iteratively optimizes these weights using differential evolution. Our evaluation includes 18 fault benchmarks constructed from CIFAR-100 and Tiny-ImageNet. We compare RepTran against three baselines: random weight selection, Arachne (a state-of-the-art DNN repair method), and ArachneW, which enables Arachne to control the number of selected weights. RepTran achieved an average repair rate of 74.7%, statistically outperforming random selection and Arachne across all benchmarks. Effect size analysis revealed that RepTran achieved higher repair rates than ArachneW regardless of the number of selected weights. These results suggest that RepTran is effective for enhancing the reliability of AI-enabled software.
arXiv:2607.11294v2 Announce Type: replace
Abstract: Open-source EDA tools allow design-space exploration (DSE) to move beyond public knobs and into bounded source-level mechanisms inside staged optimizers. We present ReviewDSE, a protected white-box DSE framework that explores such mechanisms for a target design. ReviewDSE evaluates complete source candidates under a protected evaluator and records reusable search knowledge as reviewed mechanism-level evidence. It first constructs method evidence and source-start branches from calibration designs, then uses these fixed warm-start products to initialize target-case exploration under Teacher review and full-flow validation. We instantiate ReviewDSE on OpenROAD detailed placement as a representative staged open-source EDA optimizer. Across nine target tasks, ReviewDSE reduces final post-DPL half-perimeter wirelength (HPWL) by 1.78\% on average under a 2$\times$ runtime gate, compared with 0.38\% for public-knob black-box DSE. A runtime-aware ReviewDSE selection retains a 1.68\% reduction at 1.11$\times$ runtime, and full-flow review exposes stage-composability failures while source-mechanism exploration repairs hard cut-row legality failures.
arXiv:2607.11372v2 Announce Type: replace
Abstract: Software engineering depends on the ability of developers to understand code, yet predicting how they do so remains an open challenge despite decades of research. Existing approaches rely either on simplified proxy measures that limit accuracy or on non-trivial measurements requiring elaborate experimental setups that are difficult to scale and apply in practice. In contrast, recent work in psychology suggests an alternative perspective: Instead of modeling task-specific phenomena directly, human behavior can be captured through cognitive regularities learned from large-scale behavioral data. This idea treats complex human behavior as the observable outcome of underlying cognitive processes that manifest consistently across tasks and domains. In this paper, we explore this perspective in the context of program comprehension. We evaluate Centaur, a foundation model trained on 160 general psychological experiments, on 9 previously published program-comprehension studies. We assess how well its predicted response distributions align with human response data and compare Centaur's performance to its base model, Llama 3.1. To better understand the source of its performance, we conduct ablation studies to isolate the contribution of different sources of information, such as the code artifacts, task-related context, and prior trials and participant responses. In a nutshell, we find that Centaur more closely aligns with human response patterns than its base model, is significantly less reliant on information from prior trials and responses, and benefits more from task-related information. These findings suggest that behavioral patterns learned from general psychological data can transfer to complex software engineering tasks such as program comprehension. More broadly, they point toward foundation models of human cognition as a basis for modeling developer behavior in software engineering.
arXiv:2604.22434v2 Announce Type: replace-cross
Abstract: The non-monotonic dependence of Al dopant activation on implantation temperature in 4H-SiC has been experimentally observed, but its atomistic origin remains unclear. We present a molecular dynamics (MD) study of Al implantation at $500$,K and $900$,K over seven doses from $1\times10^{13}$ to $7.5\times10^{14}$,cm$^{-2}$, followed by up to $100$,ns of annealing at $1500$--$2500$,K. Using the Gao-Weber potential combined with a reparameterized Morse potential for Al-SiC interactions fitted to DFT migration and kick-in/out barriers, we show that implantation at both temperatures reduces Frenkel-pair formation and extended amorphous pockets compared with room-temperature implantation. Above the Al solubility limit ($>10^{20}$,cm$^{-3}$), however, annealing reveals a non-monotonic temperature dependence. Samples implanted at $900$,K form larger, kinetically stable interstitial clusters that persist throughout annealing and act as sinks and trapping centers for Al, reducing substitutional incorporation. Although the $500$,K samples initially exhibit lower crystallinity, they contain a significantly larger fraction of substitutional Al after annealing. The simulations identify two regimes: a low-dose regime dominated by isolated point defects and small complexes, and a high-dose regime characterized by defect clustering and planar-defect formation with strong implantation-temperature dependence. These results explain the experimentally observed optimal implantation window between $500$ and $900$,K and suggest that controlled nanoscale amorphization at $500$,K enhances activation through regrowth-assisted incorporation while suppressing extended defects. The simulations also identify a new basal-plane diffusion path for Al and an activation mechanism involving kick-out of a carbon antisite; both were confirmed by DFT-NEB calculations.
arXiv:2605.15862v2 Announce Type: replace
Abstract: Understanding adaptive biomechanical systems requires distinguishing observable performance, static multivariate representation, longitudinal displacement, and internal approximation of observed change. This study introduces Level 5, which examines whether the M1-M2 transformation observed in a single-subject gait dataset can be approximated within a selected PCA representation. Gait was recorded with instrumented insoles in a participant with Parkinson's disease under six occlusal observational probes during two sessions eleven weeks apart. A simplified feed-forward neural network was trained to approximate M2 PC1-PC2 coordinates from M1 coordinates, occlusal-probe descriptors, and the longitudinal-transition indicator. In the core analysis aligned with Level 4, the model preserved the Euclidean centroid-displacement hierarchy dOC3 < dONL < dOC2.5. In the extended six-probe analysis, it preserved the broad structure of the exploratory ordering. Held-out M2 and leave-condition-out analyses provided internal tests beyond the full-dataset fit, while a within-session analysis described probe positions relative to ONL. The term predictive is used only in a restricted methodological sense. The model does not provide prospective clinical prediction, patient-level forecasting, or generalization to unseen individuals. Occlusal conditions are treated as observational probes applied during measurement, not as continuous causal drivers of longitudinal evolution. The findings are exploratory, retrospective, representation dependent, and non causal. They do not establish causal occlusal effects, validated viability thresholds, therapeutic superiority, distinct physiological states, or generalizable predictive validity.
arXiv:2605.18601v2 Announce Type: replace
Abstract: Modern interactive video world models have achieved impressive visual fidelity, yet lack fine-grained multi-entity control and cross-entity, cross-world generalization. We trace this gap to the action interface: standard control protocols (e.g. animation IDs, device inputs, scene-level captions) bind action semantics to specific entities or engines at design time. We propose natural language as the interface to unlock expressiveness that no prior interface can achieve, and we present Incantation, the first interactive video world model with per-latent-frame (0.25 s) natural-language conditioning that supports simultaneous multi-entity control and concept-level cross-entity transfer beyond any fixed rendering pipeline. We pair a pretrained bidirectional video backbone with frame-local text cross-attention, and enable real-time long-horizon streaming through ODE-initialized Self-Forcing distillation with a RoPE-decoupled sliding KV-cache. We surpass the Action-Index baseline on cross-entity transfer (89% vs. 43%) and out-of-vocabulary prompts (90% vs. 0%), and our 2-step student sustains 19.7 FPS at 480p with stable FVD over 2-hour rollouts. We further apply the same architecture and training recipe to The King of Fighters, changing only the per-entity action vocabulary slots. We have released a preview subset of the Incantation dataset at https://huggingface.co/datasets/zhush/incantation-elden-ring-scenes, containing manually collected Elden Ring player-boss combat clips with structured action-oriented metadata. Larger-scale Elden Ring and KOF data will be released with the full project.
arXiv:2605.31191v2 Announce Type: replace
Abstract: We investigate how teacher-student capacity relationships modulate knowledge distillation (KD) effectiveness in ResNet-based image classification on CIFAR-10. Across four teacher-student pairs (R50->R18, R34->R18, R50->R34, and R101->R34) we compare Logit-KD and Feature-KD under a strict evaluation protocol: hyperparameters and checkpoints are selected on a held-out validation split, selected configurations are re-run with five seeds, and the test set is used exclusively for final reporting. Beyond accuracy, we measure distillation fidelity directly via teacher-student agreement and KL divergence. We report four findings. First, the student-capacity pattern survives the corrected protocol at reduced magnitude: the only statistically significant gains occur for R34 students under Feature-KD (+0.19 and +0.21 pp, p<0.05), in two pairs whose teachers differ two-fold in parameters but not in accuracy, localizing the moderating variable on the student side, while no KD gain for R18 students is distinguishable from zero. Second, Feature-KD matches or outperforms Logit-KD in all four pairs, and its students land closer to the teacher's output distribution (KL at T=1) than Logit-KD students despite never observing teacher logits. Third, top-1 teacher-student agreement is flat across all pairs, decoupling fidelity from accuracy gains. Fourth, architecture dominates KD: correcting the ResNet stem for 32x32 inputs is worth +5.5 to +7.2 pp, more than 25x the largest KD gain. We also retract an attribution made in v1: a controlled re-run shows the reported gradient-clipping bug had no measurable effect, and v1's larger gains are explained by test-set selection. Code and results: github.com/umutonuryasar/kd-capacity-gap (tag v2.0).
arXiv:2604.04089v5 Announce Type: replace
Abstract: Large language models can write scientific code, but direct paper-to-program translation remains fragile when correctness depends on tacit conventions rather than explicit equations. We frame this as a knowledge-externalization problem: index choices, gauges, fermionic signs, contraction order, validation gates, and scaling constraints must be made explicit before code generation. We evaluate a multi-stage, human-in-the-loop workflow on two quantum many-body tasks. DMRG from Schollwoeck's pedagogical review serves as calibration: specification-guided implementations pass in all 16 model pairings, compared with 6/13 direct attempts, and a prose-specification ablation shows that externalized content, not LaTeX form, is the active ingredient. Pfaffian conversion of HFB states to MPS from the five-page Letter by Jin et al. serves as the stress test: the archived runs use a closed-world NumPy/SciPy/Matplotlib protocol, with no TeNPy, TeMFpy, or external implementation code supplied to the agents, so success depends on reconstructing tacit sign, gauge, ordering, and scalability conventions within a restricted dependency setting. Here the workflow yields 11/26 audited passes, while direct prompting yields none. Cross-specification transfer is asymmetric: non-GPT specifications implemented by GPT 5.5 pass 4/4, whereas GPT 5.5 specifications implemented by the tested non-GPT models fail 4/4. The contrast supports a two-bottleneck picture. Externalization resolves the first bottleneck -- paper-to-code ambiguity -- well enough to make DMRG reproducible and Pfaffian-MPS auditable. The remaining failures expose a second bottleneck in implementation-model capability. Iterative meta-specification moves this boundary but does not eliminate it. The resulting Paper-to-Program Many-Body skill is both a reusable implementation protocol and a diagnostic instrument for AI-assisted many-body programming.
arXiv:2605.23568v3 Announce Type: replace
Abstract: Manipulating fragile deformable containers, such as disposable plastic cups filled with liquid, demands real-time grip-force adaptation within an extremely narrow force margin: insufficient force causes slip, while excessive force irreversibly deforms the thin wall. Existing approaches struggle to achieve such force-sensitive manipulation tasks. We propose a noise-statistics-based calibration-driven reflex control paradigm with vision-based tactile sensing: by analyzing the sensor's intrinsic noise characteristics (via a brief static-hold-and-unload protocol), we directly derive all controller thresholds, eliminating external force calibration, trial-and-error manual tuning, or material-specific physical models. Instantiating this paradigm, we present TactileReflex, a three-channel closed-loop controller that extracts three image-level proxies, shear intensity ($S_y$), contact intensity ($F_n$), and center of pressure ($C$), from dual visuo-tactile sensors and drives prioritized reflex channels at ~12 Hz for slip suppression, weight-adaptive release, and force protection. Each channel closes the loop directly on its proxy via noise-derived thresholds. Ablation demonstrates that only the full three-channel system is able to prevent irreversible container deformation (5/5 success vs. at most 1/5 for partial configurations). In a dynamic pouring task, fixed-effort baselines fail in all 10 attempts due to pose drift, while TactileReflex achieves 9/10 success across two water volumes. As a self-contained and interpretable controller, TactileReflex can serve as a plug-and-play safety layer beneath high-level manipulation pipelines, including haptic-free VR teleoperation and vision-language-action (VLA) policies.
arXiv:2605.07066v4 Announce Type: replace
Abstract: Autonomous systems that build structures from natural-language instructions need reliable spatial reasoning, yet large language models (LLMs) make systematic coordinate errors when generating three-dimensional block placements. We present a neuro-symbolic pipeline based on 2.5-D decomposition: the LLM plans in the two-dimensional horizontal plane while a deterministic executor computes all vertical placements from column occupancy, eliminating an entire class of errors. On the Build What I Mean benchmark (160 rounds), GPT-4o-mini with this pipeline achieves 94.6% mean structural accuracy across 12 independent runs, within 3.0 percentage points of the 97.6% ceiling imposed by architect-agent errors that no builder-side improvement can address. This outperforms both GPT-4o at 90.3% and the best competing system at 76.3%. A controlled ablation confirms that 2.5-D decomposition is the dominant contributor, accounting for 28.7 percentage points of accuracy. The pipeline transfers directly to edge hardware: Nemotron-3 120B on an NVIDIA Jetson Thor AGX achieves 96.0% mean structural accuracy with the identical pipeline, slightly exceeding the cloud result. Expanding the system prompt by four targeted examples to exceed the model's 8,320-token prefix cache page size, combined with low-effort reasoning, reduces mean per-request latency by 3X to 19.7 seconds at 95.6% accuracy. The underlying principle, removing deterministic dimensions from the LLM's output space, applies to any autonomous construction or assembly task where gravity or other physical constraints fix one or more degrees of freedom. A transfer experiment on 500 IGLU collaborative building tasks confirms the effect generalizes beyond the primary benchmark.
arXiv:2606.31651v2 Announce Type: replace
Abstract: Recent automated research systems show that language-model agents can generate hypotheses, run experiments, and write complete manuscripts, but most evidence still comes from selected examples, human-framed topics, or a few pre-defined research tasks. We present FARS (Fully Automated Research System), a fully automated AI-for-AI research system designed to operate across research topics at scale. FARS autonomously generates and advances projects through ideation, planning, experimentation, and writing, using stage-specific agents coordinated through a shared workspace that records proposals, code, logs, results, and manuscripts. In its first public deployment, FARS produced 166 complete research papers spanning 67 fine-grained AI/ML topics while preserving intermediate artifacts as an auditable corpus rather than a curated set of successes. We evaluate this corpus with 282 structured reviews from volunteer reviewers covering 140 papers, including overall ratings, sub-scores, integrity checks, and LLM-use disclosure. The reviews indicate that FARS can produce review-worthy and occasionally strong AI/ML research artifacts in a large-scale public deployment, while also exposing recurring failure modes in narrow experimental scope, methodological limitations, and integrity issues.
arXiv:2607.11517v1 Announce Type: new
Abstract: Adversary emulation plans describe multi-step attacker procedures using MITRE ATT&CK techniques, privilege requirements, and observable telemetry. Translating them across operating systems supports cross-platform defender evaluation, and large language models (LLMs) can automate this task. However, a translation may only rename tools while retaining source-platform logic, giving defenders little target-platform coverage. Binary scoring can overestimate fidelity because it measures countable features rather than structural, observable, and rule-level equivalence. Graph-Based Structural Evaluation (GBSE) models each procedure as a directed attributed graph and calculates normalized Graph Edit Distance (GED) across four layers: technique, tactic, telemetry class, and Sigma logsource. GBSE was applied to a 29-step ALPHV/BlackCat Windows-to-Linux plan, comparing a reconstructed Windows control with the unmodified LLM-generated Linux version. Technique and tactic structure were fully preserved (GED=0, similarity=1.000). Telemetry similarity fell to 0.897 (GED=3) because three steps contained unmapped or drifting observables, while Sigma logsource similarity was 1.000. Every state was classified as Medium Fidelity, with a best composite score of 0.674. The 0.80 deployment threshold was not reached because technical realism scored 0.43 against the required 0.990. The framework includes bipartite GED, a telemetry-intent parser that converts free text into observable classes, and 49 validated Sigma rules: 19 for Linux and 30 for Windows. The rules provide complete ATT&CK technique coverage and pass validation with zero findings. Additional analysis reveals technique-level divergence, including RDP-based external access mapped to unencrypted exfiltration and credential-store access mapped to remote-system discovery. Results were reproduced and verified against recorded outputs.
arXiv:2607.09979v1 Announce Type: new
Abstract: Static analysis is widely used for finding security weaknesses in source code before deployment, but it often produces far more alerts than analysts can review. We study how well large language models (LLMs) can adjudicate (classify as a real bug or a false alarm) static-analysis alerts. We use two mistake-mitigation methods: (1) a consistency check (CC) that runs the LLM multiple times and checks that the verdicts are consistent with each other, and (2) an LLM reasoning evaluation (LRE) step that runs the LLM multiple times and then asks the LLM to choose a verdict after evaluating the reasoning provided by each run.
We evaluated several LLMs on three test suites: Juliet, FormAI, and SV-COMP. Across all three suites, the mid-tier reasoning LLMs that we tested (o4-mini, gpt-oss-120b, gpt-oss-20b) reach high recall (percent of real bugs that the tool correctly flags as needing repair / manual attention) and specificity (percent of actually false alerts that the tool correctly dismisses as false alarms). With mistake mitigation, they reach at least 98% recall and at least 94.8% specificity on every suite (with CC alone on Juliet and SV-COMP, and with LRE+CC on FormAI).
We probe Juliet memorization and show that o4-mini can often reconstruct sanitized test cases' original identities, so we base our generalization claims primarily on FormAI, scored against our own unpublished manual adjudications. We also report results of using the LLM to synthesize a program that dynamically triggers the flaw as independent evidence; a validity check rejected every trigger driver aimed at a false alarm, so a valid trigger proved to be strong evidence of a real flaw.
arXiv:2607.11607v1 Announce Type: new
Abstract: Distributional reinforcement learning agents learn full return distributions that are increasingly read at face value: for interpretability, risk-sensitive control, and safety monitoring. We ask a question theory anticipates but that has not been measured directly: are the risk claims of a trained distributional agent true? Our audit combines a decision-relevant screening metric (the excess Wasserstein gap between the top two actions, which equals the mass by which first-order stochastic dominance is violated), ground truth from snapshot-restart Monte Carlo, and a statistical harness (permutation nulls, bootstrap refutation, FDR control) without which the audit itself manufactures false conclusions. Across QR-DQN, C51, and IQN on MinAtar (33 runs), 40-95% of the strongest claimed risk trade-offs are refuted at 95% confidence, the placement of the strongest claims is statistically indistinguishable from truth-blind, and essentially no claim is confirmable: for these agents, the learned "risk" reflects a training artifact rather than environment stochasticity. The artifact is structural (fully formed early in training, uncorrelated with final score, idiosyncratic to each seed) and appears unchanged at full-Atari scale, with every top Breakout claim of a pretrained near-state-of-the-art QR-DQN refuted. Positive controls of known magnitude confirm 96-100% of real claims (correlation 0.89-0.92): the reading measures the agents, not the audit. Acting on the heads' CVaR advice at their most-flagged states ranges from beneficial to significantly worse than chance. Neither training for risk nor ensembling removes the artifact, and recalibration passes the audit only by nullifying the claims: the head is uninformative, not merely miscalibrated. We release the toolkit and document two silent pitfalls that produced convincing but wrong audits of our own.
arXiv:2607.11880v1 Announce Type: cross
Abstract: Open questions in collisionless plasma dissipation can be addressed using space-based observations in different astrophysical environments, with implications for both astrophysical and laboratory plasma systems. We study a low-$\beta$, highly imbalanced, sub-Alfv\'enic stream observed by Parker Solar Probe (PSP) to identify and distinguish between signatures of stochastic heating (SH) and resonant heating (RH) by parallel ion cyclotron waves (ICWs). Prior work studying this stream (Bowen et al., 2025) showed that the SH rate, accounting for intermittency, matched the amplitude of the local energy transfer (LET) rate while the RH rate did not. This comparison relied on a number of assumptions regarding the nature of the diffusive process, and the calculation of the LET rate. We introduce a novel technique of inverting the proton guiding center equation to empirically measure velocity-space diffusion coefficients using three-dimensional proton velocity distribution functions (VDFs), from the ion electrostatic analyzer (SPANi) on PSP. Measured diffusion coefficients are used to determine phase-space heating rates, leading to a calculation of a fully kinetic heating rate independent of assumptions made in prior work. We show that scale-dependent analytic expressions for SH via non-coherent fluctuations match the empirical measurements from PSP data, provided that we account for intermittency in the heating calculation. In contrast, the derived heating rates for SH that accounts for the effects of the helicity barrier, and heating rates for RH via $\parallel$-ICWs do not peak in the same region of velocity-space as the empirical measurements, nor reach the required magnitude. Our approach provides novel methodology to uniquely identify and constrain heating processes in collisionless plasmas, and shows evidence of a Fokker-Planck like diffusive process in the near-Sun solar wind.
arXiv:2606.29550v3 Announce Type: replace
Abstract: Forks share git history, so a commit surfaces in many repositories and any spread- or popularity-based measure over raw repositories is inflated by orders of magnitude. We release a curated deforking map for the World of Code (WoC) version V2604: p2PFull, which collapses every raw repository p into the deforked project P to which it belongs, built from the global shared-commit relation (51.79M shared-commit groups) via a hub-node star encoding and parallel Louvain clustering, plus capped variants (cap250/cap500) that bound mega-cluster size. The naive shared-history union over-merges: the project graph welds unrelated software into giant clusters (largest uncapped cluster 443,010 repositories, bridged by shared-commit groups as large as 267,200), for the same structural reason author-identity graphs do. A cheap size cap removes the boilerplate-hub bridges; a structural-bridge diagnostic, the cut that dissolved the analogous author mega-cluster, run here but deliberately not applied, shows the post-cap residual is genuine vendored history, robust to the cut, so we leave it intact. We validate the map against GitHub's declared fork graph reconstructed from GHArchive ForkEvents, finding 99.01% edge agreement conditional on both repositories being in WoC. Disagreements fall into two classes: a completeness byproduct (edges GitHub asserts but WoC has not ingested) and the central contribution, WoC-only fork families that GitHub's platform graph cannot represent, including 5.53% multi-forge families and 1.54% whose fork root is not on GitHub. We additionally release a refreshed fork-exclusion list (134.1M children, 3.4x the GHTorrent-era 39.5M) and a detached-fork inventory (455,550 hard-detached edges; 240,441 genuine independent origins). All artifacts are a self-contained, independently hosted replication package keyed to the WoC V2604 collection.