arXiv:2606.05737v2 Announce Type: replace
Abstract: Generating diverse images from sparse text is hard; generating compact actions from rich observations is easier. From the condition-target view, Vision-Language-Action (VLA) thus aligns with image-to-text, not text-to-image. We formalize this view through the irreducible velocity loss $R_v(t,c)$ of standard flow matching and validate it with a controlled 8-mode toy experiment and image-to-text MNIST task. We then show that high-noise training boosts one-step VLA decoding on standard LIBERO, achieving 95.6% on LIBERO-Long, and remains competitive across LIBERO-Plus, LIBERO-Pro, and real-world robot tasks, while ablations that weaken the condition or expand the horizon predictably erase the one-step gain. These results suggest that whether one-step action generation works in VLA depends not on specialized training, but on the condition-target structure.
Science Journals
arXiv:2606.06386v2 Announce Type: replace
Abstract: This paper presents the issues arising in implementing a fast integer division algorithm on general purpose GPUs. The algorithm uses a Newton iteration based on the shifted inverse operation, keeping all arithmetic in the integer domain and relying on data-parallel operators. The principal contribution is an efficient GPU/CUDA implementation for integer precisions from $2^{15}$ to $2^{18}$ -- sizes not supported by \cgbn{} division. We propose algorithmic refinements, define a cost model in terms of multiplications, build on prefix sums and previous work on multi-precision multiplication, and present an evaluation showing near-optimal performance relative to the model for the target precision.
arXiv:2606.06921v3 Announce Type: replace
Abstract: This paper introduces the Event-Shifted Acoustic Scene (ESAS) dataset, a novel benchmark for evaluating the robustness of Acoustic Scene Classification (ASC) systems against unknown sound events. Existing ASC datasets typically contain recordings of clean and consistent audio, while real-world environments often include diverse and unexpected sound events. To bridge this gap, ESAS simulates real-world acoustic variability by injecting foreground sound events into background scenes with the assistance of large language models. In this work, we present the construction methodology, dataset statistics, and evaluation protocols. Furthermore, a comprehensive evaluation of state-of-the-art ASC systems is conducted using the ESAS benchmark. Experimental results reveal that existing ASC models suffer significant performance degradation when facing the event-shift challenge. The introduction of the ESAS dataset aims to drive future research toward event-robust ASC.
arXiv:2606.07303v2 Announce Type: replace
Abstract: Representation learning is central to modern machine learning, but most research examines how representations are optimized after a framework has been selected. Less attention is given to when a new representational level becomes necessary. This article introduces the Bootstrap Theory of Representational Emergence (TBER), an initial conceptual theory and research program describing how new representations arise when existing ones become explanatorily insufficient. A representation may remain descriptively useful while failing to make certain observations, relations, transformations, or organizational properties intelligible. TBER treats this explanatory insufficiency as a positive signal for representational transition rather than as simple falsification or prediction error. The proposed recursive process follows five stages: stabilized observation, anomaly detection, recognition of explanatory insufficiency, representational emergence, and provisional stabilization. The framework concerns transitions between scientific or computational representations, not equivalent transitions within the physical systems being observed. Its scope includes representation learning, latent spaces, foundation models, world models, digital twins, adaptive biological systems, and scientific discovery. TBER does not propose a new algorithm, model architecture, benchmark, or optimization procedure. Its contribution is meta-representational: it provides a framework for interpreting when and why new representational levels become necessary. A possible implication for future artificial intelligence is the development of systems able to detect when their own internal representations have reached explanatory limits and to initiate representational refinement or transition.
arXiv:2606.07882v2 Announce Type: replace
Abstract: Different vision neural networks -- trained to classify, contrast, reconstruct, or match images to text -- should have correspondingly different internal representations. We report that they do not. After training, the top sixteen principal directions of variation inside thirteen modern vision encoders converge to the same sixteen-dimensional geometric object. We call this the cross-architecture substrate and study it with PCA, centred kernel alignment (CKA), and Pang 2026 calibration. The substrate transports across four visual domains (natural photographs, medical CT, satellite, microscopy) at median Procrustes-CKA 0.679, and across eight domains (adding sketches, depth, thermal infrared, astronomy) at 0.604, every pair >0.40. It survives Pang calibration globally (7.4x disc-vs-MAE separation, n=13,394) and locally (4.82-5.30, p<10^{-44}). It is not pixel statistics (0.263), not Gabor features (0.31), not a random projection (0.041), and emerges in the first 10% of training while accuracy keeps climbing. We deliver four applications: a label-free transferability filter beating LogME (3x faster, +0.15 Kendall-tau); a four-way domain detector (99.6% accuracy); a frozen low-shot probe (16 dims beat 768-dim DINOv2 by 3.78pp at N=50 labels per class); and a teacher-free distillation auxiliary matching trained-teacher KD on 33 pairs (7.56pp peak gain at 10% label fraction). The substrate does not cross modalities, does not help cross-paradigm distillation, and does not predict transfer quality (rho=0.08 against transfer accuracy).
arXiv:2606.08098v3 Announce Type: replace
Abstract: Majority voting is the default unsupervised aggregator for multi-sample LLM inference, but it discards two signals: within-group answer entropy and between-group reasoning geometry. We aggregate by delegation instead (Propagational Proxy Voting, PPV): each group of samples keeps weight on its own answer in proportion to its entropy-based confidence (When) and routes the rest to peers by reasoning-embedding similarity (Whom); the stationary distribution of the resulting delegation matrix picks the consensus answer. This requires neither gold labels nor training. On MMLU-Pro with 128 samples per question, delegation beats majority by +1.5 pp overall and +2.24 pp on non-trivial questions (McNemar p ~ 1.0e-14, n = 8,099), overturning wrong majorities whose answer cluster is geometrically incoherent while the correct minority is tight. We then characterize exactly when delegation overturns majority: a two-option model gives a closed-form flip condition on each option's confidence and the weight it routes to the other, with a do-no-harm corollary for near-unanimous questions. The condition calls the realized winner on 96.5% of non-trivial questions, and its predicted mass gap tracks the realized gap at r = 0.97. We did not find any other unsupervised ensemble methods that close the oracle gap.
arXiv:2606.08878v2 Announce Type: replace
Abstract: Real-world LLM applications are moving beyond single-agent workflows toward orchestrated multi-agent systems, yet current models still struggle to determine what each sub-agent needs to know. To measure this, we introduce PerspectiveGap, a benchmark for evaluating LLMs' ability to compose orchestration prompts for multi-agent systems. PerspectiveGap contains 110 scenarios, each evaluated through two distractor-mixed task formats: role-fragment assignment and free-form prompt writing. These scenarios are organized into 10 topologies, which are distilled from the authors' real-world engineering practice and framed by the Prompt Economy principle: building loop-centered orchestrations that maximize utility with minimal role and engineering overhead. In experiments with 33 commercial models from 10 companies, GPT-5.5 substantially outperforms all competitors, whereas Opus 4.8 shows a notable weakness in orchestration prompting despite its strong coding performance. Nevertheless, PerspectiveGap remains challenging: the evaluated models achieve an average combined pass rate of only 17.2\% (GPT-5.5 62.0\%) and an average overall leakage rate of 217.9\% (a per-scenario information leak-event count, not a proportion; GPT-5.5 49.1\%). These findings suggest that multi-agent orchestration prompting is a distinct and under-evaluated capability, and PerspectiveGap provides a foundation for measuring and improving it systematically.
arXiv:2606.09094v2 Announce Type: replace
Abstract: Subsurface imaging interpretation bridges observed geophysical data and quantitative geological models, supporting hydrocarbon exploration, CO2 storage assessment, and geohazard monitoring. Over the past decade, machine learning and deep learning have substantially reshaped interpretation workflows. This review synthesizes the 2015-2025 literature across four tasks: structural interpretation, geobody identification, seismic facies analysis, and property estimation, tracing the field's evolution from classical machine learning through deep learning to emerging domain foundation models, and how these tasks couple within a single interpretation system. The task remains fundamentally different from other AI applications, facing ambiguous signals, interpretive non-uniqueness, sparse semantics, unfixed target locations, and scarce reliable annotations. We synthesize three defining challenges: interpretation under complex geological conditions, cross-survey semantic generalization under low information density, and the absence of reliable benchmarks. Addressing them will hinge on integrating human expertise, physical constraints, and geological priors into training and inference, and on treating uncertainty quantification as an intrinsic model output. We outline a forward-looking agenda: unified, jointly modelled interpretation systems with cross-task consistency; priors evolving from physics toward language and multimodal supervision; end-to-end uncertainty propagation; human-AI collaboration and agent-orchestrated workflows; and a more rigorous evaluation science supported by an AI-ready data ecosystem. The review is accompanied by an open benchmark resource (CIG-Bench), covering fault segmentation, relative geologic time estimation, geobody segmentation, and property modeling, with synthetic datasets, pretrained baselines, and quantitative evaluation: https://douyimin.github.io/CIG-bench
arXiv:2606.09204v2 Announce Type: replace
Abstract: We present a reproducible failure mode of safety training in RAG-based LLM recommendation, the Injection Paradox, in which prompt injections embedded in retrieved documents backfire against the attacker, suppressing the target brand below the injection-free baseline. In safety-trained Claude models, documents containing prompt injections suffer a sharp drop in recommendation rate, and this suppression propagates beyond the injected document to unmodified documents of the same brand. In Claude Opus 4.6, the target brand drops from a 54% baseline to zero top-2 recommendations across all 50 trials, even though only 1 of 4 brand documents in the corpus contains an injection. The directional pattern is reproduced in counterfactual experiments and across three brands. A contrasting result across the GPT models tested, where the same injection instead increases recommendations, suggests model-family differences in how injection-like context affects recommendation behavior. These findings raise the technical possibility of a reverse-attack scenario in which an adversary embeds injections in a competitor's documents to suppress the competitor's brand via safety-sensitive model behavior. Code, prompts, privacy-preserving per-trial outcome records, and aggregate results are released.
arXiv:2606.09361v2 Announce Type: replace
Abstract: Cardinality estimators are built to support arbitrary schemas and workloads, forcing them to rely on generic statistics even when the schema and workload is known in advance, leaving optimizers prone to large errors and poor plans. We present Bespoke-Card, an agent-driven system that synthesizes workload-specific cardinality estimators as executable code: a planning agent designs the estimators' strategies, a coding agent implements them, and a validator scores the estimates against true cardinalities and PostgreSQL estimates, forming a robust and deterministic harness. Going beyond naive prompting, Bespoke-Card uses structured q-error feedback, regression analysis, concrete outlier subplans, a curriculum isolating join-only, filter-only, and full-subplan errors, and archival selection of the best implementation. Injecting its estimates into the optimizer cuts total PostgreSQL runtime on JOB by 33% and reduces median q-error over all JOB subplans from 190.5 to 11.5 (-94%), while synthesizing a strong estimator in under one hour for less than $10. Bespoke-Card is opening a new avenue for cardinality estimation next to classical generic estimators and learned estimator architectures.
arXiv:2606.09452v2 Announce Type: replace
Abstract: Defensive pressure on the ball carrier is a fundamental component of soccer tactics. Existing pressure measures often involve additional modeling assumptions, which may reduce interpretability. In this study, we quantify defensive pressure as the opponent minimum arrival time to the ball-carrier location, computed from a physics-based motion model. Using synchronized event and tracking data from all 306 matches in the top division of the Japan Professional Football League during the 2023 season, we analyze the statistical characteristics and temporal evolution of this quantity during ball-possession intervals. The results show that the opponent minimum arrival time tends to decrease during possession and to increase again at the start of the next possession after ball release. We also find that possessions starting under stronger defensive pressure tend to yield smaller ball progression, and that, for intentional open-play passes, possessions ending under stronger pressure are more likely to be lost. These findings indicate that minimum arrival time provides an interpretable and physically grounded measure of immediate defensive pressure on the ball carrier. The proposed framework provides a simple and interpretable baseline for quantifying pressing dynamics from tracking data.
arXiv:2606.10766v2 Announce Type: replace
Abstract: This paper investigates a viable two-echelon supply chain network design problem with unreliable facilities subject to disruptions. Unlike existing studies that consider supply chain echelons in isolation, the proposed models explicitly capture cross-echelon disruptions and quantify the value of incorporating such interdependencies. Network viability is achieved by jointly integrating resilience (via backup reassignment), agility (via mobile facilities), and environmental impact (via emissions caps) to ensure demand satisfaction across both echelons and support long-term network survival. Two mixed-integer programming formulations are developed: a scenario-based formulation and an implicit formulation, both minimizing expected fixed and service costs. To handle probabilistic service requirements, the implicit formulation incorporates a machine learning-enhanced chance-constrained programming approach, in which intractable capacity chance constraints are replaced by learned linear cuts enforcing a 95% service confidence level. These cuts are trained using several classification methods, including logistic regression, L1-regularized logistic regression, stochastic gradient descent, the perceptron algorithm, and logistic regression with a regularization parameter of 0.1, with the best-performing classifier selected as a surrogate. To further enhance scalability, two fix-and-relax heuristics are developed for the implicit formulation, while a sample average approximation (SAA) method is applied to the scenario-based formulation. Computational experiments demonstrate that the implicit formulation offers a computationally efficient and high-quality alternative to the scenario-based formulation. Moreover, the proposed heuristics and SAA approach effectively address medium- and large-scale instances, delivering high-quality solutions within acceptable computational times.
arXiv:2606.12946v2 Announce Type: replace
Abstract: Does the conversion of academic assessment from percentage scores to letter grades merely represent an adjustment in information precision, or does it systematically alter the underlying structure of student academic data? Drawing on 68 mathematics exam scores from 75 primary school students, this study employs Encoding Transformation simulation to compare structural differences in the same dataset under two encoding schemes across three dimensions: information loss, distance structure change, and clustering stability. Results indicate that letter-grade encoding compresses the mean pairwise distance in the trajectory feature space from 20.50 to 1.06 (a compression ratio of approximately 19:1); after standardization, the density gradient of the distance distribution is systematically flattened, with kurtosis decreasing by 0.54 and the coefficient of variation decreasing by 0.16; and the clustering structure becomes highly sensitive to minor sample perturbations-removing a single extreme observation causes the optimal K to jump from 4 to 8 and clustering reproducibility to plummet from 95% to 62%. Concurrently, a Compression Paradox is observed: letter-grade encoding increases the Silhouette coefficient while simultaneously reducing clustering stability. These findings demonstrate that letter-grade encoding systematically alters the structural representation of longitudinal academic data and exerts measurable effects on clustering stability.
arXiv:2606.13172v2 Announce Type: replace
Abstract: Learned representations are central to modern machine learning and are commonly evaluated through predictive performance, robustness, uncertainty estimation, and generalization. However, a representation may remain operationally successful while failing to organize persistent residual structures that conventional metrics do not fully capture. This article introduces VER, the Vigilant Evaluator of Representations, a conceptual framework for monitoring representational adequacy. VER does not propose a new learning algorithm, loss function, or model architecture. It defines a diagnostic process for identifying residual structures and assessing whether they may indicate explanatory insufficiency rather than ordinary error, uncertainty, noise, data limitation, or distribution shift. The framework comprises five operations: representation identification, explanatory-domain delimitation, residual-structure detection, explanatory-resistance evaluation, and vigilance signaling. VER distinguishes stable adequacy, a vigilance condition, and a representational alert. It is intended to complement performance evaluation, uncertainty estimation, out-of-distribution detection, robustness analysis, and explainable AI by making representational adequacy an explicit object of inquiry. The article also outlines a path toward empirical evaluation through benchmarks designed to detect representational inadequacy when predictive performance remains satisfactory. VER is conceptual and methodological; it does not prove that a representation is inadequate, select a replacement representation, or provide an operational implementation.
arXiv:2607.09774v1 Announce Type: new
Abstract: Large language models are now commonly used as partners in scientific writing, and this shift has brought a subtler type of failure: made-up references. Fabricated authors, bogus DOIs, wrongly assigned identifiers, and citations that merge elements from multiple genuine articles are now being inserted into manuscripts at a volume that traditional peer review was never meant to handle. Recent audits reveal that such references have already slipped through the review process and made their way into the published literature, including leading journals and conferences. Automated verification that operates at the speed and scale of modern content production has therefore become a necessary safeguard rather than a convenience. This work presents and evaluates the AtomGPT reference checker (https://atomgpt.org/hallucination_detector), an open, web-accessible tool that verifies citations against the scholarly literature by combining large-language-model field extraction with structured retrieval from Semantic Scholar. For each reference, the tool extracts the bibliographic fields, retrieves the closest matching real papers, and scores the agreement across title, authorship, and venue to produce a graded judgment of whether a citation is trustworthy, partially supported, or likely fabricated. We benchmark the tool against an externally curated set of confirmed hallucinated citations from accepted NeurIPS 2025 papers and find that it reliably flags the great majority of them.
arXiv:2607.10455v1 Announce Type: new
Abstract: Autonomous CLI agents can now execute hundreds of actions across multi-hour sessions: writing code, executing shell commands, browsing the web, and managing cloud infrastructure, all with minimal human oversight. Does greater autonomy invite greater risk? We introduce ANCHOR, an automated auditing framework that stress-tests CLI agents on illegal tasks grounded in public US court cases. ANCHOR deploys an auditor agent fine-tuned on dark personality data using supervised and reinforcement fine tuning. This auditor roleplays persistent malicious users who decompose tasks, reframe requests upon refusal, and adapt strategies across multi-turn interactions. Evaluating frontier CLI agents, we find that while they often refuse illegal tasks when prompted directly, compliance reaches 100\% under persistent malicious interaction. When agents comply, they frequently exceed user requests, autonomously building infrastructure for large-scale harm, including catastrophic risk scenarios such as large-scale financial fraud and bioweapon development. These findings demonstrate that current alignment techniques are insufficient for autonomous agents and underscore the need for safety evaluations against persistent, adaptive malicious users. We release ANCHOR at https://github.com/garified/anchor
arXiv:2607.11354v1 Announce Type: new
Abstract: Offline evaluation is the standard methodology for comparing top-N recommender systems, yet it relies on incomplete relevance information. In most benchmark datasets, only a small subset of user--item preferences is observed, and unjudged items are commonly treated as non-relevant. This missing-as-negative assumption can bias evaluation, penalize plausible recommendations with no recorded feedback, and favour algorithms that concentrate on popular or highly exposed items. We propose an LLM-based framework to expand relevance judgements for offline recommender evaluation. Our approach uses large language models in two complementary roles. First, a preference induction stage summarizes each user's historical interactions into a textual profile that captures their tastes and interests. Second, conditioned on this profile, an LLM acts as a relevance judge for candidate recommended items that lack observed labels in the original test data. To make this process tractable and evaluation-focused, we apply judgement expansion to a pooled candidate set built from the top-ranked outputs of multiple recommenders. The resulting enriched judgements provide additional relevance evidence for previously unobserved user--item pairs, enabling ranking metrics to be computed on a more complete basis. Experimental results show that this approach is a promising strategy for improving the robustness of offline top-N evaluation and mitigating the popularity-sensitive distortions caused by sparse feedback.
arXiv:2607.11671v1 Announce Type: new
Abstract: Near-field radiative heat transfer can exceed the blackbody limit through evanescent-mode coupling across nanoscale gaps. This enhancement underpins applications including thermophotovoltaic energy conversion, electroluminescent cooling, thermal rectification, and photon absorption in plasmon-assisted photodetection. These systems most often involve photon- or heat-exchange between dissimilar interfaces, particularly between a semiconductor and a metal. Despite the prevalence of this asymmetric configuration, no closed-form description of its near-field interaction exists. Here, we derive a closed-form analytical description of photon tunneling that clarifies the roles of material properties, namely the plasma frequency, optical loss, and semiconductor absorption, in the thermal exchange. We show that the dominant in-plane wave vector of the radiative heat transfer is an approximate average of the corresponding values for two symmetric reference systems: a plasmonic-plasmonic cavity and a semiconductor-semiconductor cavity. These results establish a compact analytical framework for near-field heat transfer between dissimilar materials.
arXiv:2606.13843v2 Announce Type: replace
Abstract: GenUI is an emergent class of AI tools that use large models to generate UI mock-ups based on users' high-level descriptions, promising to democratize UX design exploration for a broader audience. Most GenUI designs to date tend to inherit the conventions of conversational large models, such as ChatGPT and Gemini, where a user describes their design needs primarily via an unstructured prompt, and the tool then takes a depth-first approach, delving into the design right away and producing a high-fidelity prototype. In this research, we rethink how well this unstructured, depth-first, and high-fidelity GenUI design can support early-stage, 0-to-1 design exploration. To probe this question, we propose a contrastive design with structured input, breadth-first exploration, and low-fidelity generation. We then conducted a comparison study with 24 UX designers and product managers who conducted mini design exploration exercises using an existing GenUI tool and our contrastive GenUI tool. Findings reveal participants' perceived benefits and trade-offs of the two GenUI designs: structured input surfaces key facets but requires more work, raising entry barriers to start exploration; breadth-first workflow reveals more possibilities, but previewing UX ideas spanning many screens remains hard; and though low fidelity has value, professionals favor high fidelity because it fits practice and GenAI heightens fidelity expectations. We conclude with design implications for GenUI and similar AI-powered creativity support tools.
arXiv:2606.13894v2 Announce Type: replace
Abstract: AdamW is a default optimizer for modern deep learning, but its first and second moment states add roughly two parameter-sized buffers to training memory, increasing the already substantial cost of large-scale pretraining. We propose Gefen, a memory-efficient optimizer that automatically shares second-moment estimates across parameter blocks and quantizes the first moment using a learned codebook, thereby reducing AdamW's memory footprint by ~8x while maintaining the same performance, corresponding to a reduction of 6.5 GiB per billion parameters. The method is motivated by a theoretical result showing that large mixed Hessian entries constrain the ratio of squared gradients toward one, suggesting that Hessian-aligned parameters are natural candidates for sharing second-moment statistics. Since computing Hessians is impractical at scale, Gefen infers block structure from the initial squared gradients, requiring no architecture-specific metadata or hyperparameters beyond AdamW defaults. Gefen learns an exact histogram-based dynamic-programming quantization codebook and reuses the same blocks for first-moment scaling. Across diverse pretraining experiments, Gefen achieves the lowest peak optimizer memory among the compared AdamW-like methods while maintaining AdamW-level performance. In single-machine or distributed training, the reduced memory footprint enables larger microbatches and improves throughput significantly over AdamW, providing a practical drop-in replacement with lower memory usage that can increase throughput and enable training larger models or using larger global batch sizes. We provide the complete Python implementation, including fused CUDA kernels at https://github.com/ndvbd/Gefen
arXiv:2606.14617v2 Announce Type: replace
Abstract: Safe physical human--robot interaction on floating-base robots requires interaction regulation under changing contact constraints. We develop a contact-consistent normalization in which the residual end-effector channel is represented as a linear double integrator in acceleration coordinates. Both discrete prediction matrices are independent of configuration and support mode; posture and contact enter only through task-inertia force recovery and constraints. The controller combines a constant-Hessian receding-horizon QP, an acceleration-disturbance observer, and a priority-consistent realization. Classical operational-space impedance is shown to be the unconstrained infinite-horizon limit. MuJoCo experiments on a 17-DOF biped and a Menagerie-derived Unitree G1 model evaluate sustained forces, transmitted shocks, and scheduled contact-model changes. Disturbance estimation is the dominant source of fixed-stance accuracy, while covariance inflation gives only scenario-dependent transient benefit. Dynamic walking and hardware validation remain outside the present evidence.
arXiv:2606.16278v2 Announce Type: replace
Abstract: Long-tail hazardous scenarios are essential for safety-oriented autonomous driving, yet they are difficult to collect and reproduce at scale. Editable 3D Gaussian Splatting (3DGS) simulation offers a promising alternative by reconstructing real driving scenes and supporting controllable scene editing. However, edited 3DGS-rendered videos still suffer from a significant Sim-to-Real gap, including rendering artifacts, degraded foreground assets, inconsistent illumination, and temporal flickering. Existing restoration and video generation methods are insufficient for this task, as they often fail to jointly repair 3DGS-specific artifacts, improve visual realism, and ensure temporal consistency. To fill this gap, we propose RealityBridge, a structure-preserving and asset-aware Sim-to-Real framework for edited 3DGS driving videos. RealityBridge uses multimodal controls, including rendered videos, foreground masks, edge maps, and semantic masks, together with a lightweight GateNet for adaptive condition allocation across backbone layers. We further construct targeted training data and introduce autoregressive long-video training with reward-guided post-training to improve restoration quality, temporal stability, and hallucination suppression. Extensive experiments on internal and public driving datasets show that RealityBridge outperforms existing methods in artifact removal, illumination harmonization, and long-sequence temporal consistency.
arXiv:2604.15865v2 Announce Type: replace
Abstract: Series and parallel elastic actuators offer complementary but mutually exclusive advantages, yet no existing actuator enables real-time transition between these topologies during operation. This paper presents a novel actuator design called the Dual-Topology Elastic Actuator (DTEA), which enables dynamic switching between SEA and PEA topologies during operation. A proof-of-concept prototype of the DTEA is developed to demonstrate the feasibility of the topology-switching mechanism. Experiments are conducted to evaluate the robustness and timing of the switching mechanism under operational conditions. The actuator successfully performed 324 topology-switching cycles under load without damage, demonstrating the robustness of the mechanism. The measured switching time between SEA and PEA modes is under 33.33 ms. Additional experiments are conducted to characterize the static stiffness and disturbance rejection performance in both SEA and PEA modes. Static stiffness tests show that the PEA mode is 1.53x stiffer than the SEA mode, with KSEA = 5.57 +/- 0.02 Nm/rad and KPEA = 8.54 +/- 0.02 Nm/rad. Disturbance rejection experiments show that the mean peak deflection in SEA mode is 2.26x larger than in PEA mode (5.2 deg vs. 2.3 deg), while the mean settling time is 3.45x longer (1380 ms vs. 400 ms). The observed behaviors are consistent with the known characteristics of conventional SEA and PEA actuators, validating the functionality of both modes in the DTEA actuator.
arXiv:2604.19753v2 Announce Type: replace
Abstract: We propose a feature-free approach to algorithm selection: instead of hand-crafted instance features, we use pretrained text embeddings. Our method, ZeroFolio, proceeds in three steps. First, it reads the raw instance file as plain text. Second, it embeds it with a pretrained embedding model. Third, it selects an algorithm via weighted k-nearest neighbors. The key to our approach is the fact that pretrained embeddings can distinguish problem instances without any domain knowledge or task-specific training. Hence, we can apply the same three-step pipeline (serialize, embed, select) across any problem domain with text-based instance formats. We evaluate our approach on 11 ASlib scenarios spanning 7 domains (SAT, MaxSAT, QBF, ASP, CSP, MIP, and graph problems). Our experiments show that this approach outperforms a random forest trained on hand-crafted features in 9 of 11 scenarios, robustly across serialization seeds (every seed, not just a favorable one) and often by a substantial margin; it still wins 8 of 11 against a per-scenario-tuned random forest. On the three scenarios with published AutoFolio results from the 2015 ASlib competition, ZeroFolio comes within a small margin of AutoFolio without any per-scenario configuration tuning. Our ablation study shows that inverse-distance weighting, line shuffling, and Manhattan distance are the key design choices, and we analyze the sensitivity of the selector to the serialization seed. On SAT12-ALL, where both selectors are competitive, combining embeddings with hand-crafted features via soft voting yields a further improvement.
arXiv:2606.17034v2 Announce Type: replace
Abstract: Post-hoc context erasing over the KV cache is challenging because a local edit has a global consequence: once a span has been processed, its influence propagates into the cached states of all subsequent tokens. This issue arises naturally in long-context LLM applications, where stale retrieved facts, incorrect tool observations, retracted user preferences, or harmful prompt injections may be identified only after prefill. Exact erasing must then recompute all tokens after the deleted span, making its computational cost depend on suffix length rather than erased-span length. We introduce KVEraser, a learned KV-cache editing method for efficient localized context erasing. Given a processed context and a span to remove, KVEraser replaces only the KV states of the erased interval with learned steering states while reusing the remaining cache unchanged. To learn a transferable erasing mechanism, we build a two-stage training pipeline: generic span-neighbor pre-training teaches the eraser to suppress the influence of the erased span, while task-specific fine-tuning adapts this capability to downstream scenarios. Experiments show that KVEraser nearly matches full recomputation in post-erasure performance on in-domain tasks across 1K--32K context lengths, while its latency increases by only 24% compared with a 17.6x increase for full recomputation. KVEraser also generalizes to unseen long-document QA tasks with harmful factual distractors, achieving the best performance among approximate baselines with a 3--4x speedup over full recomputation.