Several public scripts for formulaic alpha mining store prices adjusted to the snapshot's last day, so a past day's stored price changes with every later split or dividend and formulas that do not cancel the adjustment factor depend on future corporate actions. We formalise a change of adjustment anchor as a per-stock group action and give a two-sorted type system that soundly certifies anchor invariance from a formula's syntax, under a declared field-adjustment contract and idealised operator semantics; its second sort admits log-returns, and a linear-time checker plugs into any miner. In a re-anchoring audit on Chinese A-shares, pools that five published methods mined in our benchmark pipeline score a mean information coefficient (IC) 0.005 higher on the stored representation than on a cutoff-anchored one, about 14% of their IC. The mean signed change for certified pools is about -0.001, and pools mined on past-anchored data stay sensitive. On the Chinese development folds, certification shows no detectable predictive cost or gain; the estimated minimum detectable difference is about 0.017 IC across cells on pool ICs of 0.03-0.04. We also trace an erroneous index-membership entry to public Qlib data snapshots and show that re-mining exposes the effect of such data defects, which re-evaluating fixed pools understates.
Thi Dinh Tran, Maria Teresa Rossi, Ludovico Iovino, Leonardo Mariani
原文摘要
Assessing whether concrete software architectures are consistent with their reference architecture (RA) is essential to ensure that architectural decisions scale coherently across projects, preserving the quality attributes the RA was designed to guarantee. In the absence of dedicated techniques, mapping the components of an architectural model onto a predefined RA can be addressed only with general-purpose model-comparison algorithms. Our experiments show that these algorithms are inaccurate when applied to RAs. To address this gap, we propose Reference Architecture Mapping Discovery (RAMD), an LLM-based workflow that combines model analysis with structured prompting, and refines results through an iterative RA compliance-checking mechanism. We evaluate RAMD on 64 open-source architectural models spanning three RAs from the IoT, self-driving, and smart parking domains. With domain knowledge and GPT-4o, RAMD reaches 85.8% mapping accuracy and outperforms the strongest baseline by 44.6 percentage points, while ensuring structural consistency with the RA.
Automatic Music Transcription (AMT) relies on paired audio--MIDI datasets, implicitly assuming near-perfect temporal alignment. However, widely used piano datasets have been shown to contain alignment errors that may bias model evaluation and training. We propose a framework for evaluating audio--MIDI alignment scoring methods through controlled temporal, pitch, and structural MIDI corruptions. We also introduce STI, a scoring method built on AMT models, and benchmark it against existing alignment scoring approaches. Our results show that STI, together with methods based on CENS and spectrograms, achieves the strongest overall performance, with STI exhibiting the highest sensitivity to fine-grained temporal misalignments, the main alignment error in AMT datasets. We then apply the best-performing methods to widely used piano AMT datasets and identify trends in their relative alignment quality. These results support more informed dataset curation, model training, and AMT evaluation.
Data annotation is an essential but costly aspect of developing machine learning-based technology, requiring multiple human workers to repeatedly label each data point in a data set. We explore the tradeoff between annotation accuracy and cost by identifying minority reports---worker responses disagreeing with the reference majority-vote label---that may indicate unnecessary redundancy, and can therefore be proactively pruned before they are executed. We develop a theoretical framework for choosing the number of repeated annotations required for a task and pruning policies at a target label quality. We estimate minority-report risk on two computer vision datasets re-annotated through a professional platform, then evaluate a rolling policy on the assignment stream to prune over 35% of repeats while preserving more than 99.9% of reference majority-vote labels.
This paper introduces SupplyNetPy, an open-source, well-documented Python library for modeling and discrete-event simulation of supply chain networks with arbitrary multi-echelon structures. It supports multiple replenishment policies, perishable inventory, node disruptions, and stochastic demand and lead times. All components are extensible via inheritance. Users describe a supply chain as a graph with node and link attributes, while the library handles simulation, providing logs and extensive node and network level performance reports. This paper presents the motivation, design, key features, and architecture of SupplyNetPy, along with detailed validation results (against analytical benchmarks, a commercial tool, and a published case study). A key motivation behind SupplyNetPy's development is programmatic generation and simulation of complex models, enabling design-space exploration, what-if analysis, training data generation, and supply chain digital twins.
Understanding 4D scenes is a core challenge in visual intelligence, requiring models to jointly capture scene geometry, appearance, and dynamics from partial observations. However, existing approaches are subdivided into feedforward models for direct geometric prediction, and generative models for visual synthesis. We introduce UGG, a unified framework for 4D scene understanding that jointly models RGB video and geometric representations such as depth and camera pose. By assigning different noise levels across modalities and time, UGG formulates visual generation and geometric prediction as conditional completion within a single generative model. This flexible formulation enables the model to learn from diverse, heterogeneous datasets with different subsets of geometric and visual annotations. Furthermore, our generative formulation enables inference-time search over multiple plausible scene hypotheses, using geometric evidence extracted from observed RGB frames to improve geometric prediction without ground truth geometry. We demonstrate UGG's effectiveness over non-unified baselines across camera-controlled video generation, depth estimation, and camera pose prediction, including in challenging sparse observation settings. Code and visualizations are available at https://ugg-model.github.io/
Over the last decade, building human-level artificial general intelligence has moved from far-fetched speculation to being a concrete next-decade target for many of the largest AI organizations. Achieving this goal would have profound and far-reaching impacts on human society, which raises many complex questions for the decade ahead. This report investigates how AI itself might continue to develop in a post-AGI world along the continuum of machine intelligence. The endpoint of this continuum, Universal AI, is theoretically well understood, which provides some formal grounding for the main focus of this report: the transition from human-level AGI to artificial general superintelligence, which can intuitively be understood as a system that is more intelligent and cognitively capable than large organizations of humans. After characterizing ASI, the report discusses four potential pathways from AGI to ASI: scaling AGI, AI paradigm shifts, ASI emerging from large-scale multi-agent collectives, and recursive improvement. The report then discusses possible frictions and bottlenecks along these pathways. Determining whether the impact of these frictions will be negligible or substantial raises a number of concrete open research questions. While large uncertainties for predicting ASI progress remain, it cannot be ruled out that AI progress might continue to accelerate over the next years. This would imply that the image of a single transformative step change, caused by the introduction of human-level AGI into our society, could be inaccurate. More apt might be the prospect of a series of transformative societal changes caused by AI-enabled progress and breakthroughs across many areas of science and technology. Preparing for this prospect requires a massively interdisciplinary endeavor of global scope and interest.
Deepfake detectors are commonly evaluated on real/fake classification alone, but real-world applications require explanations of the visual evidence behind each prediction. Joint vision-language models struggle with authenticity assessment, producing "real" narratives for manipulated images. To address this, we propose a decoupled predict-then-explain pipeline as our submission to the ACM Multimedia 2026 Explainable Deepfake Detection Challenge, with methods trained and evaluated on the XPlainVerse benchmark. A vision-foundation-model ensemble first classifies the image, then a Qwen3-VL-8B model generates a decision-conditioned explanation of the visual evidence. For fake images, a second Qwen3-VL-8B compresses that text into a short non-expert summary without using the image. We further apply metric-aligned Group Relative Policy Optimization (GRPO) to optimize both explainers against the challenge criteria. Under matched conditions, GRPO raises complex BERTScore-F1 from 0.686 to 0.712 and also improves grounding (entity and evidence F1); our submission ranked first on complex BERTScore-F1 and seventh overall.
We describe our solution for the 2026 BEHAVIOR Challenge: 100 long-horizon household tasks in the OmniGibson simulator. We placed 9th on the final leaderboard, the only independent team in the top 10. The core of the solution is a single vision-language-action (VLA) model, one checkpoint for all 100 tasks, built on our 1st-place 2025 solution, which is based on $π_{0.5}$. On the 50 tasks shared with 2025, its Q rises from 0.260 to 0.371; across all 100 tasks it scores 0.392. Our main conclusion from 2025 was that the model was undertrained, so most of the 2026 work went into training the model for longer, optimizing some training parameters, making each training step slightly faster and reducing the gradient noise: a migration from 50 to 100 tasks (semantic stages built from the demonstrations' skill annotations, task and stage embeddings of new tasks warm-started from similar old ones, and a 21-dimensional action space matching the teleoperation controls); variance-reduced multi-sample flow matching with a simple noise-vs-compute model that picks the number of action-expert samples per observation without ablations; and a 200k-step training run, about one pass over the 211M frames of the demonstrations. A cloud evaluation infrastructure keeps a shared record with provenance, episode diagnostics and restorable simulator states, and runs six simulators on one GPU together faster than real time. On top of the learned policy, 14 tasks are routed to a GPT-6 Astra coding agent that controls the robot through observation and motion tools, starting every episode from a frozen copy of its own practice session on training instances. Across these 14 tasks, Astra's mean Q is 0.639, against 0.108 for the learned policy. The public submission included only part of the Astra results and scored 0.446 Q with 27.3% full-task success.
Revisiting Eigenvalue-Only Divide-and-Conquer Tridiagonal Eigensolver from Algorithmic Perspective
Ruiyi Zhan, Dajun Huang, Lu Shi, Shaoshuai Zhang
AI 速读 基于摘要
作者提出仅用每个合并节点的首末两行边界行构成递推来携带分治状态,将持久内存从二次降为线性,并在 CPU 与 GPU 上按去膨胀后的有效秩做细粒度调度。
cs.DC · math.NA
查看原文摘要
Divide-and-conquer (D&C) is the natural parallel strategy for symmetric tridiagonal eigenvalue decomposition, yet eigenvalue-only solvers fall back to serial QR algorithm because the quadratic transform state each merge carries exhausts memory at scale. In this paper, we demonstrate that each merge stage in D&C needs only two boundary rows, the first and last rows of a node's eigenvector block, and these close into a recurrence whose parent rows follow from its children's and local merge data alone, so the state is carried up the tree with nothing replayed and persistent memory drops from quadratic to linear. Because this state keeps no replay workspace, adapting the schedule to each merge's post-deflation active rank is a scheduling choice, so both CPU and GPU spend finer-grained parallelism only where substantial secular work survives. We evaluate optimized CPU and GPU implementations, with complexity and correctness analysis, against QR, bisection, and D&C routines in standard libraries.
Demonstration of MANA: Multi-perspective Alignment of NPC Agents
Amir Tahmasbi, Keshav Singh, Shaheer Raza, Owen Mei, Jason Wu
AI 速读 基于摘要
MANA 让开发者在 Minecraft 游戏中直接切换到 NPC 视角,通过具身示范来对齐 NPC 行为,并将示范蒸馏为可泛化的策略。
cs.HC
查看原文摘要
We present MANA (Multi-perspective Alignment of NPC Agents), an interactive approach for authoring NPC behavior through embodied demonstration. We show MANA in the context of Minecraft, where a developer aligns NPCs while playing the game itself. Activating MANA lets the developer dynamically switch into the point of view (POV) of any character and demonstrate, in-situ, how it would act given the surrounding context, through both movement and dialogue. MANA then distills each demonstration into a generalized policy that maps in-game context to actions. Finally, once the developer switches to another NPC or exits MANA mode, this generalized policy drives the NPC on its own. We demonstrate MANA by aligning two initially generic Minecraft NPCs: Fletcher and Farmer to act aggressively and cowardly, respectively. After roughly 2 minutes of in-game play, each character exhibits consistent and believable behaviors across new, unseen scenarios.
Linear-size graphs at every strict dismantling level
Jesús F. Espinoza, Martín Eduardo Frías-Armenta, Rosalía G. Hernández-Amador, Héctor A. Hernández-Hernández
AI 速读 基于摘要
作者为每个严格拆解层级 k 构造了顶点数随 k 线性增长(3k+8 与 4k+4)的图,并给出最小阶数的上界 min{4k+4, 3k+8}。
math.CO · cs.CG · cs.GR
查看原文摘要
Dismantling reduces a graph to one vertex by successive deletions. At level zero, a vertex is eligible if one of its neighbours is adjacent to all its other neighbours. At level $k\geq1$, its neighbours must induce a graph dismantlable at level $k-1$. A graph has strict level $k$ if it admits a reduction at level $k$ and none at level $k-1$. For every $k\geq1$, we construct a graph of strict level $k$ with $3k+8$ vertices. Each step adds three vertices: the neighbours of one induce the preceding graph, and the other two provide the adjacencies used to rule out other first deletions at the lower level. A second construction gives a graph of order $4k+4$ at strict level $k$. Together the two families give the upper bound $\min\{4k+4,3k+8\}$ on the least order at strict level $k$.
Branch-and-Bound (B&B) and $A^*$ are popular exact search methods originating from Operations Research and Artificial Intelligence, respectively. Although the two methods share fundamental search principles, their different ways of exploring the search space can lead to substantially different computational performance. We investigate these differences using the Resource-Constrained Project Scheduling Problem (RCPSP) as an exemplary problem setting. We first establish the relationships between mixed-integer linear programming, state-action, and timed Petri-net representations. Since B&B and $A^*$ admit numerous algorithmic enhancements, we focus on relatively basic implementations. This allows us to examine the fundamental differences between the methods rather than compare the performance of state-of-the-art solvers. By using the same lower bound and best-first node selection, we isolate the effects of the search mechanisms, particularly the identification of equivalent states reached through different scheduling decision sequences. Experiments on 480 benchmark RCPSP instances show that graph-based $A^*$ substantially outperforms the B&B variants considered. With additional pruning rules, $A^*$ proves optimality for all 480 instances, compared with 445 for best-first B&B under the same time and memory limits. Depth-first B&B uses less memory and can find feasible solutions even when the search does not finish within the time limit. Stronger linear programming relaxations reduce the search space but introduce considerable computational overhead. Our findings highlight the importance of duplicate-state detection and clarify the computational trade-offs between the search approaches. These insights can help researchers select an appropriate method and identify opportunities to incorporate effective mechanisms from one approach into the other.
The weighted edit distance $\mathsf{ed}^w(X,Y)$ is the minimum cost of an alignment of strings $X$ and $Y$, where a weight function $w$ specifies the cost of each character insertion, deletion, and substitution. For strings of total length $N$, the textbook dynamic program takes $O(N^2)$ time, and no $O(N^{2-δ})$-time algorithm exists, even for unit costs, unless the Orthogonal Vectors Hypothesis (OVH) fails. This motivates the study of approximation algorithms. For unit costs and metric weights, the best truly subquadratic approximation factor was $3+\varepsilon$; we give randomized $3$-approximations in $\tilde O(N^{11/6})$ and $\tilde O(N^{40/21})$ time, respectively. For non-metric weights, we prove quadratic lower bounds for any approximation factor: $Ω(N^2)$ queries to $w$ over alphabets of size $Θ(N)$, and OVH-hardness over constant-size alphabets. These lower bounds motivate the bounded setting, parameterized by $k:=\mathsf{ed}^w(X,Y)$ for $w$ normalized so that every edit costs at least one. We give a deterministic $(1+\varepsilon)$-approximation algorithm in $\tilde O(N+k^2/\sqrt\varepsilon)$ time, matching the $O(N+k^2)$-time Landau-Vishkin algorithm for unit costs up to logarithmic factors for constant $\varepsilon>0$. Our main technical contribution is a Monge-matrix approximation framework for alignment problems: every $n\times n$ Monge matrix with entries in $\{0\}\cup[1,W]$ has an entrywise $(1+\varepsilon)$-approximation by a Monge matrix with $\tilde O(n\log(W)/\sqrt\varepsilon)$ strict Monge inequalities. As further applications of our Monge approximation framework, we maintain a $(1+\varepsilon)$-approximation of $\mathsf{ed}^w(X,Y)$ under character edits in $\tilde O(N\log(W)/\sqrt\varepsilon)$ time per update for weights at most $W$, and in $\tilde O(K/\sqrt\varepsilon)$ time for a fixed threshold $K$.
Operator learning for partial differential equations (PDEs) on arbitrary geometries builds fast neural surrogates for large-scale simulation. While recent geometry-adaptive neural operators have made substantial progress, they are mainly designed for forward problems in which inputs and outputs share the same spatial domain. This limits their applicability for boundary value problems (BVPs) and inverse problems, where inputs and outputs may live on different domains. We introduce the Geometry-Adaptive Integral Autoencoder (GAIA), an operator learning model that encodes the domain boundary and the interior field distribution into geometry tokens, and conditions integral transform layers on these tokens via cross-attention, allowing the kernel to adapt locally to geometric features. This yields an architecture capable of learning solution operators for forward and inverse problems on arbitrary domains, solving each in a single pass at inference without per-domain optimization, iterative refinement, or graph construction. We evaluate GAIA on seven 2D and 3D benchmarks, four of which are new or substantially extended benchmarks for inverse problems and BVP: electrical impedance tomography, optical tomography, 3D Darcy flow on varying geometries, and a modified setting of the Poisson BVP on mechanical components benchmark (MCB). GAIA sets new state-of-the-art results on every inverse and BVP task, reducing median relative $L^2$ error by 64% on airfoil flow reconstruction and 22% on EIT relative to the next best amortized method, and outperforming all baselines on every shape category of MCB. On other forward problems, GAIA is competitive with specialized solvers while maintaining stable accuracy across point resolutions on which transformer-based baselines degrade.
Time-Series Foundation Models (TSFMs) are increasingly used in sensor applications, where sampling frequency (SF) varies across sensors, devices, and operating modes (e.g., power saving). Ideally, TSFMs must remain effective during such SF changes, provided the SF retains task-critical bandwidth. In this paper, we stress-test 4 prominent TSFMs: MOMENT, Chronos-2, Moirai-MoE, and TimesFM, to quantify their SF-invariance and investigate whether their behavior tracks the resulting loss of information bandwidth rather than the SF value itself. Our stress-test probes 3 dimensions: (i) the shift in embeddings, (ii) the degradation in forecasting performance, and (iii) the degradation in downstream performance, quantified by three novel metrics: FSI, FDI, and CDI, respectively. To our knowledge, this is the first study to place SF-invariance at the core of TSFM evaluations. Our results characterize: (i) how SF-invariance depends on the margin between signal's native-SF and the bandwidth its task requires, (ii) how much of that invariance the choice of TSFM explains, and (iii) how much of TSFM's performance degradation under a lower SF reflects information loss rather than a limitation of their abilities. These findings reframe SF-invariance as a measurable, signal-aware property of the (signal X TSFM) pair, and guide practitioners on when TSFMs display SF-invariance.
Neural Processes (NPs) are flexible meta-learning models capable of uncertainty quantification across diverse applications. However, the standard NP framework assumes access to noise-free inputs. This is unrealistic in many practical settings, where sensor noise or measurement errors are unavoidable. This discrepancy between model assumptions and real-world data can significantly degrade prediction quality and uncertainty estimates. We introduce Denoising Neural Processes (DnNPs), a principled extension of the NP framework that explicitly models input uncertainty. We train DnNPs via variational inference. DnNPs jointly learn to denoise corrupted inputs and perform meta-learning, making them robust to noisy training data. We demonstrate DnNPs' effectiveness on both synthetic and real-world datasets, showing consistent improvements in prediction and uncertainty quantification compared to standard NPs. On PM2.5 spatial interpolation from real-world air quality sensors, DnNP reduces RMSE by up to 30% and NLL by up to 73% relative to standard NPs. We also provide theoretical grounding: we show that the Bayes-optimal predictor under noisy inputs belongs to the DnNP model class, and that plug-in models with fixed observation variance incur an irreducible KL divergence gap that DnNP provably avoids.
Quantum algorithms for machine learning, chemistry, and cryptanalysis often assume coherent access to classical data at negligible physical cost. We instead treat data loading as a hardware process and quantify the dynamic energetic cost of a quantum random access memory (QRAM) query on a superconducting processor. Using the metric-noise-resource framework, we relate query fidelity and power to gate duration, relative two-qubit drive strength, and the distribution of attenuation across a multi-stage refrigerator, while accounting for thermal photons and gate errors. We analyze the linear-depth multiplexer and the noise-resilient bucket-brigade architectures, adapting the latter to a native superconducting gate set and refining its fidelity analysis using the fully decomposed gate schedule. For a representative bucket-brigade QRAM query with a $12$-qubit address register we obtain a query fidelity of $0.969$. The query consumes $3.51\times10^{-4}\,\mathrm{J}$ of dynamic energy, corresponding to an average dynamic power of $4.38\,\mathrm{W}$ and a peak of $64.4\,\mathrm{W}$. Our optimization shows that, at fixed target fidelity, the operating point and attenuation profile can change the average power by a factor of two or more.
Streaming speech recognizers output a word only after the speaker has finished it, typically 90-400 ms later. For a voice agent every millisecond counts: the sooner the words are outputted, the sooner the agent can start a search or plan its reply. We present Earlybird, a 29M-parameter streaming recognizer (16.8 MB at 4 bits). Earlybird learns from audio cut off at random points, including partway through a word, and is asked to predict every word that is due by then in full, plus a guess of the next two words. Its raw token probabilities give a confidence for every word including the one still being spoken. On clean audiobook speech (LibriSpeech test-clean), the median word is predicted correctly 90 ms before the speaker finishes it, while released streaming models get it 90 to 400 ms after it ends. Earlybird is the earliest on all six test sets we time, covering audiobooks, TED talks, public speech, earnings calls, and synthetic speech. On the same audiobooks, 69% of words are predicted correctly before they end, against 3-25% for the other models. For words still being spoken, confidence is calibrated, with an expected calibration error (ECE) of 3.7 to 5.5. On the eight English Open ASR Leaderboard sets, it is more accurate than Moonshine Tiny on every set and roughly ties Whistle, a model of the same size.
Video world models promise general-purpose interactive simulators, but their training is limited by action supervision: action labels are scarce, fragmented across incompatible specifications, and do not scale with the unlabeled video pretraining corpus. A common workaround jointly trains an action encoder to compress consecutive frames into a latent action and a decoder to reconstruct the next frame from the previous frame and that latent, forcing the latent to capture only new information. This has two limitations: learning the action space from scratch can limit generalization to unseen domains, and the bottleneck trades expressiveness for separation, with tighter settings losing nuance and looser ones risking scene leakage under domain shifts. We argue for a different starting point: rather than learning actions from scratch, we start from a pretrained foundation model (V-JEPA 2), whose embeddings of a frame pair already carry substantial action information after large-scale, diverse video pretraining. The remaining challenge is to extract the action while discarding the scene. We propose a label-free contrastive adaptation framework that pulls together frame pairs sharing the same motion under a different scene and pushes apart pairs from the same scene with different motion, reshaping the geometry of the pretrained space without imposing a reconstruction bottleneck. We introduce a unified evaluation protocol along three axes (action expressiveness, scene invariance, and action-conditioned video generation quality) across robot manipulation, camera navigation, game, and driving benchmarks, most of which are unseen during adaptation. Compared with annotation-free latent-action methods, our representations are substantially more expressive while matching their scene invariance, and they yield world models that transfer demonstrated actions to new scenes more faithfully.
AI systems have become increasingly capable of dangerous behaviours in many domains. This raises the question: Do models sometimes choose to violate human instructions in order to perform behaviour that is more useful for certain goals? We introduce a benchmark for measuring model propensity for instrumental convergence (IC) behaviour in terminal-based agents. This is behaviour such as self-preservation that has been hypothesised to play a key role in risks from highly capable AI agents. Our benchmark is realistic and low-stakes, aiming to reduce evaluation-awareness and roleplay confounds. The suite contains seven operational tasks, each with an official workflow and a shortcut whose policy status varies by condition. An eight-variant shared framework varies monitoring, instruction clarity, stakes, permission, instrumental usefulness and blocked honest paths to support inferences regarding the factors driving IC behaviour. We evaluated ten models using deterministic environment-state scorers over 1,680 samples, with trace review employed for audit and adjudication purposes. The adjusted IC rate is 86 out of 1,680 samples (5.1%). IC behaviour is concentrated rather than uniform: two Gemini models account for 66.3% of IC cases and three tasks account for 84.9%. Conditions in which IC behaviour is indispensable for task success result in the greatest increase in the adjusted IC rate (+15.7 percentage points), whereas emphasising that task success is critical or certain framing choices do not produce comparable effects. Our findings indicate that realistic, low-nudge environments elicit IC behaviour rarely but systematically in most tested models. We conclude that it is feasible to robustly measure tendencies for dangerous behaviour in current frontier AI agents.
Formal world models can support planning, verification, and safety reasoning, but constructing them from open-ended natural-language situations is difficult, particularly in multi-agent settings where several formalizations may be plausible. We study whether five LLMs can generate such world models as partially observable stochastic games and whether mechanical validity is sufficient to establish that the resulting model faithfully represents the intended decision problem. We introduce a benchmark of 44 scenarios spanning game, situational, and cultural settings and evaluate generated world models along three dimensions: mechanical validity, reference-relative structural agreement, and semantic criteria. Across three empirical tracks, we analyze 410 first-pass generations. In the reference-backed game track, 134/135 generated world models pass a strict mechanical audit, yet mean permutation-invariant structural F1 is only 0.366, with just 28/134 mechanically valid models exactly matching the authored reference. In the cultural track, explicitly naming the country changes the generated structural representation for most LLMs while producing only small changes in automated semantic scores. These results show that executability is a necessary but insufficient criterion for LLM-generated world models: models that are mechanically valid can still differ substantially in the structure used by downstream neuro-symbolic reasoning and planning.
Local decoding usually fixes one weight precision for the whole generation. We study a different allocation: store 8-bit weights, decode with their top 4 bits, and redo a step at 8 bits only when the low-precision output distribution looks uncertain (entropy above a threshold or a small top-two margin). Quantization is simulated, the model is Qwen3-0.6B, and the benchmark is GSM8K. Cost is counted in weight bytes read per token. On 36724 replayed steps the entropy of the low pass ranks argmax flips with AUC 0.93. At a 15% escalation budget it catches 78% of flips and the margin gate 94%, against the 15% expected for a random gate of equal rate (a computed expectation, not a run). In 150 live problems all-high scores 64.0% and all-low 55.3% (McNemar p=0.053). An entropy gate with a 16% escalation rate scores 62.7% at 0.87 of the all-high weight bytes. Its paired difference is -1.3 (-8.0 to +5.3) points to all-high and +7.3 (+0.0 to +14.7) points to all-low (p=0.080), so the live data do not separate the gate from either end. Verified speculation reaches zero KL at similar bytes in the same cost model. A live random control, a thinking-mode phase rule, a second model and live KL were not run.
Audio foundation models like CLAP, MuQ, and MERT excel at capturing general musical similarity. On the other hand, their entangled latent spaces limit interpretability and fine-grained attribute control. Conditional Similarity Networks (CSN) address this by pairing a backbone representation model with learned masks that gate the embedding space into attribute-specific subspaces. However, existing methods train the backbone from scratch, rather than leveraging pretrained foundation models. We instead present a systematic study of post-hoc disentanglement, adapting CSN to gate frozen pretrained embeddings. We evaluate adding trainable parameters to adapt the embedding under different masking and pooling regimes across two datasets. Our findings show that minimal adaptation capacity is sufficient for post-hoc gating to improve both attribute-specific disentanglement and overall retrieval performance, while pooling, masking, and depth choices yield limited additional gains.
Fetal brain biometry provides a quantitative basis for assessing prenatal growth and identifying congenital brain abnormalities. However, comprehensive assessment remains limited by labor-intensive manual measurements and automated models that lose accuracy across imaging sequences and institutions. We introduce BioTTA, an anatomically guided test-time adaptation framework that measures 11 fetal brain biometrics from reconstructed three-dimensional magnetic resonance images. BioTTA adapts pretrained models to individual unlabeled volumes using geometric consistency and atlas priors, without access to source images. In a retrospective study across five centers in China and Switzerland, BioTTA improved measurement accuracy across three out-of-distribution datasets and two network architectures. It reduced mean absolute measurement errors by 46-82% relative to the corresponding models without adaptation, achieving errors similar to those of supervised fine-tuning. Its measurements were mapped onto normative growth trajectories derived from expert annotations, enabling gestational-age-specific assessment through a web-based reporting platform. Across four disease cohorts, the platform achieved 82.1-95.0% accuracy in binary detection of ventriculomegaly or cerebellar abnormalities, with performance similar to that of a senior radiologist. BioTTA provides a practical foundation for standardized fetal brain biometry across heterogeneous imaging settings, supporting scalable prenatal growth assessment and quantitative evaluation of brain abnormalities.
Despite the widespread adoption of autonomous vehicle technologies, reliable navigation within cluttered environments remains arduous due to the necessity of interaction-aware strategies; example scenarios include: narrow passages with multiple agents, lane changing in dense traffic, and crossing unsignalized intersections. These scenarios often require highly proactive strategies, simultaneously demanding interpretability and quick adaptability, which challenge end-to-end and data-driven methods. An effective strategy requires: scalability to multiple agents, adaptability to changes in other agents' behavior, and harmony with surrounding traffic without freezing or becoming overly conservative. To meet these requirements, we present SIGMO, a sigmoid-gated interaction-aware motion planner, with two main contributions. First, we formulate polynomial sigmoid functions that analytically estimate yield scores of surrounding agents using physics-grounded interaction features that capture driving intuitions. Second, we integrate the polynomial sigmoid functions into a receding-horizon planner, which facilitates interaction-aware planning with arbitrary trajectory predictions. We validate SIGMO in interaction-intensive simulations and hardware experiments. The results show that SIGMO balances proactive and defensive behavior while remaining computationally lightweight, serving as a strong baseline in a mobility stack. The simulation and hardware experiment videos are available here: https://sites.google.com/view/sigmoid-polygate.
Local language models can save memory by storing weights in fewer bits, and they can reduce sensitivity to the order of answer options by averaging a few reorderings. We ask whether the first makes the second worse, and which is the cheaper way to spend a fixed budget. On Qwen3-0.6B and Qwen3-1.7B with simulated round-to-nearest weight quantization (8, 6, 4 and 3 bits, fp16 reference), read from the logits of the letters A to D on ARC-Easy, ARC-Challenge, MMLU and HellaSwag (300 items each), 6 bits lose no accuracy against fp16 (they score +1.8 (+1.0 to +2.5) and +0.9 (+0.3 to +1.5) points higher, which we do not explain), and 8 bits change it by -0.7 (-1.1 to -0.3) and -0.4 (-0.7 to -0.2) points. At 4 bits the answer-slot skew rises from 0.15 to 0.53 (0.6B) and from 0.07 to 0.12 (1.7B); Gaussian noise at the same weight SNR gives 0.15 and 0.37 (two draws) and 0.19 (one draw), so the control is inconclusive on whether rounding adds slot skew beyond equal-power noise. Flip rates rise but stay below an accuracy-matched baseline of independent uniform errors. Averaging four cyclic orders recovers +12.6 points at 4 bits for 0.6B (+5.3 at fp16). We price bits and passes in one machine-independent unit, weight bytes read times passes. In cost-matched pairs, extra orders beat extra bits only when the weights have 6 or more bits, and the decision table reduces to: use 6-bit weights and spend the rest of the budget on up to four orders, one pass each. Everything is simulated on two small models with the simplest quantizer; a title-level search found no earlier study of quantization together with option order.
We study exact evaluation of the arithmetized $k$-Orthogonal Vectors polynomial $FOV^k$ of Ball et al. (2017, 2018). For every fixed integer $k \geq 3$, set $d(n) = \left \lceil \log^2 n \right \rceil$. Assuming the basic decision version of Quantum Strong Exponential-Time Hypothesis (QSETH), for every constant $\varepsilon > 0$, no uniform quantum algorithm runs in $O(n^{k/2 - \varepsilon})$ time on every input and evaluates this polynomial correctly with probability at least $3 / 4$ over a uniformly random input in $\mathbb F_{p(n)}^{knd(n)}$ and the algorithm's internal randomness, for all sufficiently large $n$. This holds when $p(n)$ is the smallest prime greater than $n^{k - 1}$ and when it is the smallest prime greater than $n^k$. Our direct reduction combines classical polynomial self-correction with calls to a quantum average-case evaluator. For the smaller prime, random sampling finds an orthogonal tuple with constant probability whenever the count is large enough to wrap around modulo $p(n)$; otherwise, the exact residue determines the answer. For the larger prime, quantum minimum finding and an asymptotic prime-gap bound account for the cost of generating the prescribed field. In view of prior isolation-based reductions permitting small characteristics, we present these results as explicit reductions for the two canonical-prime families, with all setup costs included in the uniform quantum running time. The hardness guarantee concerns single-instance exact evaluation; a matching quantum evaluation algorithm and a quantum proof-of-work protocol remain open.
We introduce intuitionistic public announcement logics Lmin and Lequ determined by classes of birelational frames where the intuitionistic preorder and the modal accessibility relation do not need to satisfy the confluence conditions traditionally considered within the framework of IPAL. Our logics can be seen as generalisations of IPAL seeing that their language is based on a binary epistemic modality of "relativised knowledge" that allows us to define valid reduction axioms. Our main results are the soundness and completeness of Lmin with respect to the class of all birelational frames and the soundness and completeness of Lequ with respect to the class of all birelational frames where the modal accessibility relation is an equivalence relation.
Concept-based explanations in neural models are important for the development, certification, and adoption of autonomous systems. Yet, interpretability remains correlational. Without methods to guide and quantify causal relationships between explanations, a concept explanation can be provided without being necessary or sufficient for the observed output thereby appearing correct while misleading. We introduce Causal Concept Wrapper Network (CCW-Net), the first use of interventional mediation analysis from fields such as economics, medicine, and epidemiology as a training objective to align causal effects of concept-based explanations with a model's output. We further apply mediation analysis as an evaluation tool to quantify, per concept label combination, how much of any model's output is causally attributable to each concept explanation. As a training objective, CCW-Net shapes a model's concept representations such that they are both necessary and sufficient for the outputs they are credited with and independent of other concepts' contributions. Counterfactual concept representations anchor a set of causal alignment criteria: necessity, sufficiency, and independence. Necessity drives irrelevant concept effects toward zero, sufficiency ensures relevant concept representations fully reconstruct a model's output, and independence enforces the total effect decomposes linearly across concepts. We evaluate CCW-Net across aircraft control, driving, and image classification. CCW-Net provides a step toward concept-based explanations that are not merely interpretable, but quantifiably causal, supporting safer evaluation, certification, and deployment of transparent and trustworthy neural models.