Papers

Many physical properties relevant to robotic manipulation are hidden from vision. A sealed object, for example, may reveal little about its center of mass (COM) or internal contents until it is lifted, shaken, or otherwise dynamically perturbed. This study shows that such interactions can enable a new modality of robotic perception and learning, in which interaction-induced dynamic responses are used to infer object physics that is inaccessible to conventional sensing. We implement this idea using an origami-inspired soft robotic arm that functions as a physical reservoir computer. After grasping an object, the arm is excited by a fixed shaking input at its base, and the resulting ringdown response is recorded through either camera tracking or embedded sensors. Because the input is held constant across trials, hidden object properties, such as the COM position, are encoded through their effect on the dynamics of the coupled robot-object system. A lightweight linear readout can then decode these dynamics to recover interpretable information about the hidden object physics. Using this framework, the soft robotic arm reservoir completed three tasks of increasing difficulty: inferring the orientation of the object's hidden COM, inferring the COM distance from the grasp point, and using the inferred COM information to guide a subsequent regrasp. We further develop a dynamic summary representation of the ringdown response that improves prediction accuracy. Together, these results establish shake-to-learn mechanical interrogation as a promising strategy for robotic systems to convert brief physical interactions into actionable cues about hidden object properties for downstream manipulation.

In non-stationary online learning, dynamic regret has attracted increasing attention as a measure of how well an online learner performs against a time-varying comparator sequence. Despite considerable advances, attaining optimal bounds for strongly convex and exp-concave losses often involves intricate analysis. In this paper, we present a simple framework that reduces dynamic regret minimization to switching regret minimization. As a result, we can derive dynamic regret bounds by using off-the-shelf algorithms with switching regret guarantees. The key idea of our reduction is to construct, for any comparator sequence, an auxiliary random sequence that is unbiased at each round, with the controlled variance and a manageable number of switches. Combining this construction with suitable surrogate losses, we can decompose dynamic regret into the expected switching regret against the random sequence and its controlled variance. Theoretically, for strongly convex and exp-concave losses, we establish the $\widetilde{O}(T^{1/3}P_T^{2/3})$ dynamic regret bounds, where $T$ denotes the time horizon and $P_T$ denotes the path-length of the comparator sequence. Moreover, for general convex losses, the same reduction also recovers the $O(\sqrt{T(1+P_T)})$ dynamic regret bound. Notably, all our findings match the minimax optimal results for these three types of losses, highlighting the versatility of our proposed framework.

Close-proximity pipeline inspection is challenging for standard drones due to high hovering power consumption, airflow sensitivity, and propeller wash interference with gas sensing. To address this, we present AeRove, a compact bimodal aerial-terrestrial robot. AeRove uses its propeller guards as wheels to roll along pipes and employs a spring-loaded bistable mechanism to reconfigure between ground and flight modes in 200 ms without continuous actuator power to maintain either state. The converging propeller-guard geometry also improves measured thrust efficiency. By reserving flight for obstacle hopping and using ground rolling for continuous traversal, current draw is reduced 14x compared to continuous flight (0.7 A vs. 10 A), extending estimated travel distance from 144 m to 2057 m. Autonomous trials on a 51 cm diameter steel pipe demonstrated navigation on straight and curved sections, obstacle jumping, and leak detection of simulated inspection markers. Separate CO2 sensing experiments evaluated gas-detection performance under perched and aerial operating conditions. Perched inspection produced a substantially larger concentration response than hovering under the tested conditions. During flight, an underbody propeller-intake configuration produced a faster and stronger gas-detection response than an extended probe by leveraging propeller-induced airflow. Code and designs are released under the CC-BY license.

The proliferation of harmful internet memes poses a significant societal threat, yet their automated classification remains a formidable algorithmic challenge due to the nuanced, multimodal nature of their content. To address this, we introduce MemeTAG, a novel dual-objective framework that pioneers a keyword-aware approach to meme classification. Our core innovation is a two-part semantic guidance mechanism: first, we leverage a pretrained Vision-Language Model to generate a set of descriptive keywords, that capture the high-level semantics. Second, we introduce the Aggregated Tag Inference Network (ATIN), an attention-based module that distills these keywords into a single, rich semantic embedding. This embedding serves as a target for a novel auxiliary reconstruction loss, which compels the model to learn deeply aligned visual and textual features. This approach, combined with an efficient three-stage training strategy, establishes a new state-of-the-art on the HarMeme, Hateful Memes Challenge (HMC), and PrideMM datasets, decisively outperforming existing state-of-the-art methods.

The EU AI Act mandates that datasets for high-risk machine learning (ML) systems meet strict quality criteria such as soundness and bias mitigation. While Satisfiability Modulo Theory (SMT) solving offers a formal approach to verifying these properties, its scalability in realistic ML settings remains unexplored. To bridge this gap, this work presents the first large-scale empirical study of SMT-based dataset verification on two real-world ML datasets. We systematically evaluate how solver performance is shaped by three key dimensions: the type of data-quality property, the specification style, and the dataset encoding strategy. Our findings demonstrate that SMT-based verification is feasible for practical scenarios, but each dimension shapes it: the property type sets the tractability limit, the specification style drives scalability (exceeding $2{,}000{\times}$ for aggregate properties), and the encoding strategy has a systematic effect, with extracted feature columns performing best.

Performance models of LLM servers support both latency evaluation and the design of controllers for autoscaling against service level objectives (SLOs) and for inference optimization. We model the multiplexed execution of prefill and decode operations with a tractable, approximate queueing model under Markovian assumptions. Three parameters characterize a model-accelerator pair, namely a baseline per-iteration overhead, a per-token compute cost, and a per-token key-value (KV) cache access cost. The model combines a mean-value analysis of per-iteration work with a state-dependent Markov chain for batch occupancy to predict mean time to first token (TTFT) and inter-token latency (ITL). We validate these predictions against measurements and show that the three parameters can be estimated from observed latencies. Over a grid of input and output lengths and arrival rates spanning light to moderate load, the relative error of the average ITL is about 5% for Llama-3.1-8B and 8% for Qwen2.5-14B running on an H100 GPU, and the corresponding TTFT errors are 14% and 16%. We then implement an autoscaling controller that uses the model to adjust inference-server replica counts as the workload changes. On an OpenShift cluster of H100 GPUs it tracks a fourfold load ramp under both latency targets, missing one in 7 of 127 control cycles, and its in-loop predictions carry median errors of at most 5% for TTFT and 9% for ITL. A decode-throughput analyzer from an existing autoscaler, which takes no latency target, misses 27 of 128 cycles under the same controller and load while provisioning 4% and 28% fewer replicas.

Projection-based controllers can overcome fundamental limitations of classical linear time-invariant control by modifying the controller's input-output behavior via projection. A key example is given by the hybrid integrator-gain system, a projected integrator, which has recently found successful application in several industrial systems. While prior work on analysis and design of projection-based control systems has primarily focused on the continuous-time setting and non-incremental analysis, a more refined incremental analysis in discrete-time is needed to better reflect actual digital implementation and obtain more accurate (robust) performance assessment. To address this need, this paper considers incremental stability and convergence analysis of discrete-time projection-based control systems. Our first methodology is based on showing that such controllers preserve the quadratic incremental stability of their nominal (unprojected) dynamics, if the projection metric is well-designed. Building on this, we derive a small-gain condition guaranteeing incremental input-to-state stability for interconnections of projected controllers with general nonlinear plants. A second approach is grounded in a direct Lyapunov-based method for verifying incremental stability in input-affine piecewise-smooth systems, which can be seen as an extension of the classical discrete-time Demidovic conditions. We illustrate our results through several examples, and demonstrate performance quantification via nonlinear Bode plots, with a special focus on first-order projection elements.

We present a novel end-to-end model-based Reinforcement Learning (RL) algorithm for efficient policy synthesis under given Linear Temporal Logic (LTL) specifications (e.g., safety or reachability) in unknown environments. To do so, a Limit-Deterministic Büchi Automaton (LDBA) representation of the LTL task is synchronised with a Bayes-Adaptive Markov Decision Process (BAMDP) representation of the environment, which allows us to leverage an enhanced exploration-exploitation trade-off that is achieved via Bayesian RL, as opposed to traditional non-Bayesian approaches. We further propose a novel Bayes-Adaptive Monte-Carlo Planning (BAMCP) algorithm to allow for approximate Bayes-optimal strategy synthesis in the synchronised BAMDP construct. A range of finite- and infinite-horizon task experiments demonstrate the effectiveness of our approach in terms of both property satisfaction and sample efficiency, when compared to traditional model-free approaches. Additional ablation studies also successfully highlight the value of the novel BAMCP algorithm in comparison to classical BAMCP for LTL task satisfaction. Finally, we also showcase a successful application of our approach for cautious RL, namely to reduce the number of task violations incurred during policy training.

The paper proposes a day-ahead scheduling framework with a novel multi-stage battery degradation modeling method for an electric vehicle (EV) fast charging station (FCS) equipped with a battery energy storage system (BESS). Unlike previous studies, which employ a single battery degradation model to represent the aging process, this paper proposes a novel multi-stage battery degradation modeling method to accurately capture the degradation process across the whole lifespan. Subsequently, the multi-stage model is explicitly integrated into the proposed adaptive optimization framework in a computationally tractable way, thus having important practical implications in the field. The paper provides case studies to demonstrate the effectiveness of the proposed modeling method on a selected cycle aging model in reducing the operation cost of FCS with BESS operating in different stages. As a result, the overall operation cost with the multi-stage model is around 2.1% on average lower than the single-stage model counterpart. In addition, results show that with the increasing number of divided stages, the model error decreases and becomes stable, while the reduced operation cost compared with the single-stage model increases and saturates. Finally, we apply the multi-stage framework considering other conventional degradation models to show the superiority of the proposed method.

Undesired information such as harmful content and private data propagates through Multilingual Large Language Models (LLMs) via direct training and indirect cross-linguistic spread. Multilingual Machine Unlearning (MMU) aims to remove such information, yet its evaluation remains underexplored, leaving unclear whether unlearning truly eliminates target knowledge across all languages. To bridge this gap, we introduce $\mu^2$-Bench, an MMU benchmark that simulates the full pipeline of memorization, unlearning, and evaluation across diverse languages. It 1) spans a broad set of languages, 2) evaluates on both training and hold-out languages, and 3) assesses knowledge as dispersed across multiple languages. We show that successful MMU requires methods that reflect multilingual characteristics, and conduct analysis to provide deeper insights into MMU.

The conventional distinction between forced and natural oscillations treats the oscillation source as external to a fixed power-system boundary. This paper develops a coupling-aware dynamical interpretation by explicitly accounting for the directionality of source-grid coupling. It examines when prescribed forcing and a one-way autonomous source are observationally equivalent, and when grid measurements can distinguish the presence of grid-to-source feedback. Matched interface signals yield exact equivalence under the same deterministic grid initial-value problem and measurement map. Bidirectional feedback enables source-grid eigenvalue interaction in the augmented autonomous model, while a reduced two-mode model shows that the interaction depends jointly on loop coupling and complex modal separation. The resulting framework establishes source-grid coupling directionality as a feedback-based dynamical boundary and clarifies when feedback can be inferred from grid-side measurements and why weak feedback may remain unresolved. Simulation studies on a two-area system demonstrate feedback detection within a calibrated source family, while source-model mismatch can be confounded with feedback. A WECC 243-bus system study demonstrates feasibility on a large system and shows that feedback detection need not require exact candidate selection.

LLMs increasingly assist scientific peer review. This creates a feedback loop when generated reviews enter the training data of future LLMs. While recursive training has been studied broadly, its effects on scientific judgments remain poorly understood. We empirically study this effect in a controlled setting by varying synthetic-review exposure while fixing other training conditions. After one recursive step, increasing exposure from $0\%$ to $100\%$ makes recommendation ratings more concentrated, independently generated reviews of the same paper about $11\%$ more similar, and reviews across different papers about $5\%$ less diverse. A further step of recursive training amplifies homogenization when synthetic supervision is reintroduced. We characterize this task-specific phenomenon as $\textbf{scientific-judgment collapse}$. We also explore a training-free method using contrastive activation steering with a limited reference reviews. Controlled experiments show a partial recovery in recommendation agreement and judgment diversity across different papers, but not among independently generated reviews of the same paper. To explore this approach beyond controlled experiments, we develop $\textbf{TrustReviewer}$. We train the model on a large-scale curated data and apply activation steering at inference time. On held-out papers, steering increases exact recommendation match from $73.85\%$ to $75.40\%$, compared with $73.10\%$ for OpenReviewer on the same LLM backbone, while maintaining relatively diverse scientific judgments. Our findings quantify judgment homogenization under recursive training and motivate further research on preserving judgment diversity.

In the classical prophet inequality, an algorithm observes a sequence of random variables with known distributions in an online fashion, and it must select one of the random variables with the goal of maximizing the expected value of its selection. The performance of the algorithm is compared to an omniscient prophet who observes all of the random variables before having to make its selection. Combinatorial extensions of the classical prophet inequality in which the algorithm gets to pick a subset of the random variables (constrained to belong to some family of feasible sets) have been studied extensively. We study prophet inequalities with a $k$-matchoid constraint (a common generalization of a $k$-matroid intersection constraint and a $k$-bounded hypergraph matching constraint) in two common online arrival models. We give guarantees with respect to the ex-ante fractional relaxation of the omniscient prophet, obtaining an ex-ante competitive ratio of $\frac{1}{k+1}$ in the adversarial order case, and $\frac{1-e^{-k}}{k}$ in the random order case. Using the duality framework of Lee and Singla \cite{Lee2018}, this also yields online contention resolution schemes in these settings. Our adversarial-order prophet inequality can be viewed as a generalization of a recent $\frac12$-competitive matroid prophet inequality by Kalantarzadeh and Pashkovich, 2026. This generalization introduces a new framework: coordinated weighted principal partitions across multiple matroids. Our random-order prophet inequality is a generalization of the $k=1$ matroid case of Lee and Singla, 2018. The two results improve previously known competitive ratios for $k$-matroid intersection, which were $\frac{1}{(e+o(1))k}$ and $\frac{1}{k+1}$, respectively.

In this work, we study the neural scaling laws of RydbergGPT, an autoregressive transformer model trained on qubit projective measurement data gathered from interacting Rydberg atom arrays. The quantum system is known to exhibit a finite-size remnant of a critical point as the laser detuning parameter is varied. We find that near the critical point the transformer loss as a function of training dataset size is well described by a power-law with a loss floor correction. However, away from criticality the quality of the power-law description is substantially reduced. We then compare the statistical structure of both Rydberg measurements and natural-language corpora using an entropy-normalised, finite sample corrected mutual information "two-point" function. We find that near-critical statistics of the two point functions are closest to those observed in natural-language, whilst other qubit configurations far from the critical point have two-point functions that decay more rapidly. This supports the hypothesis that multi-scale dependence contributes to stable neural scaling, and that scaling behaviour should be viewed as a property of the model-data pair.

Multiplayer PC gaming on Linux faces a structural problem: the anti-cheat systems that publishers require operate as proprietary Ring 0 kernel modules that are architecturally incompatible with Linux's security model, GPL licensing, and stable ABI guarantees. We argue the right response is not to port these invasive modules to Linux, but to replace them entirely. This paper presents TPM-Attest, a hardware-rooted remote attestation framework that uses the Trusted Platform Module (TPM) 2.0 and the Linux Integrity Measurement Architecture (IMA) to prove, cryptographically, that a client booted cleanly and ran only authorised software – without any kernel driver, without proprietary code, and without scanning player memory. The system intercepts Epic Online Services (EOS) SDK calls via a userspace LD_PRELOAD hook, gates session access on a live TPM quote bound to a server-issued nonce, and constructs an index-prefixed Merkle tree over the IMA log that is immune to duplicate-leaf collision attacks. Across 500 constructed tamper sessions we achieve a 100% detection rate; incremental leaf caching reduces repeat-attestation latency to under 3 seconds on real TPM 2.0 hardware. A controlled red-team evaluation against a live demo game confirms all four file-backed attack vectors are blocked while precisely characterising the two confirmed bypass conditions. The full implementation is released as open-source software.

Quasars are luminous objects in the universe that exhibit stochastic brightness variations encoding information about the supermassive black holes powering them, and modeling these variations from ground-based survey data time series, known as light curves, is a statistical challenge. This paper reviews how stochastic differential equations (SDEs) have been adapted with neural network parameterizations to overcome this challenge in history. We create the Continuous-Delayed-Memory Stochastic Gradient Descent which depend on the past state of the discrete iteration process. We performed the simulation on some 2-dimensional landscape and observed some wider-exploration and more precise convergent behavior compared to Vanilla SGD by adjusting hyperparameters. Besides, we proposed a reinforcement learning structure with continuous time policy gradients for exploratory policies without solving HJB PDE, and we show that its optimality conditions recover the Gibbs policy of previous works.

In this work, we present an uncertainty-driven training framework for three-dimensional computed tomography (CT) lung nodule classification, where validation-based uncertainty estimates guide loss reweighting to enhance predictive performance and probability calibration. Two Uncertainty Quantification (UQ) methods are considered: Monte Carlo Dropout (MCD) and Evidential Deep Learning (EDL). Both provide per-class uncertainty estimates that modulate the loss and encourage focus on hard or unreliable classes. The framework is evaluated with ResNet, DenseNet, EfficientNet, Vision Transformer (ViT), and Swin Transformer backbones on two datasets: the clinical LIDC-IDRI cohort and the NoduleMNIST3D benchmark. Uncertainty-driven training achieves classification performance similar to conventional training while substantially improving calibration, with an expected calibration error (ECE) reduced by up to 65% on LIDC-IDRI. EDL attains competitive performance on shallower architectures with single-pass inference, whereas MCD is more robust on deeper networks. Analysis across architectural families reveals that uncertainty-driven training benefits convolutional backbones more consistently than transformer-based architectures: EDL in particular degrades on ViT, suggesting that the Dirichlet evidence parameterisation may interact unfavourably with attention-based architectures at lower input resolutions. A posteriori temperature scaling proves highly effective across all configurations, indicating that a simple scalar calibration can be competitive even without explicit uncertainty-aware training. Our results indicate that integrating UQ into the training loop can significantly improve probabilistic calibration and support more trustworthy deployment of three-dimensional medical imaging models.

Hybrid motor-imagery brain-computer interfaces (MI-BCIs) combining EEG and fNIRS can outperform EEG-only systems by exploiting complementary electrophysiological and hemodynamic information. To obtain such hybrid information when paired EEG-fNIRS acquisition is unavailable or inconvenient, recent studies have focused on EEG-to-fNIRS cross-modal generation. However, existing methods still suffer from slow generation and often require pretraining, limiting their use in real-time MI-BCI scenarios. Although one-step generative models offer an attractive route to low-latency synthesis, removing the iterative refinement process can reduce generation fidelity and introduce non-physiological artifacts. To address these problems, this paper proposes Bio-MF, a latent-free one-step MeanFlow framework for EEG-conditioned fNIRS generation. Bio-MF performs direct signal-space x-prediction, converts this signal-space output into MeanFlow velocity supervision, and completes inference with one network evaluation. To preserve task-relevant hemodynamic structure under heterogeneous sensor layouts, Bio-MF integrates Spatial-Temporal Interactive 4D Encoding, cross-modal classifier-free guidance, and noise-level-gated FFT regularization. On Dataset 1, EEG + synthetic fNIRS improves ACC over EEG-only by 3.37 and 4.15 percentage points for HbR and HbO, respectively. On Dataset 2, the corresponding gains remain 2.98 and 2.50 percentage points under the unseen 64-channel EEG montage. On an RTX PRO 6000 GPU, Bio-MF generates one fNIRS trial in 7.0 ms, corresponding to an 857x speedup over the 1000-step SCDM latency. These results show that Bio-MF enables fast EEG-to-fNIRS synthesis while preserving task-relevant generation quality for downstream hybrid MI decoding. Our code is available at https://github.com/psychosiwa/Bio-MF.

Motivational interviewing (MI) is a collaborative approach to elicit autonomous motivation for health behavior change. Generative AI (GenAI) offers new ways to deliver MI via conversational systems, but evidence on their design, assessment, and translation into interventions remains fragmented. This scoping review characterized evidence on GenAI-MI chatbots across system design, safety, MI quality, user perceptions, and intervention outcomes. We conducted a PRISMA-ScR scoping review. Nine datasets were searched for studies published or publicly available from January 1, 2015 to June 2, 2026 that used GenAI to generate MI chatbot responses or counselor utterances. Data were extracted using a predefined framework and synthesized descriptively. Forty-seven reports (48 studies) were included. Twenty (41.7%) focused on system design without direct participant use; 28 (58.3%) involved direct interaction. Most systems were text based and disembodied; 23 (47.9%) incorporated dynamic adaptation. Safety measures were unevenly reported. Among studies with direct use, 21/28 (75.0%) reported informed consent or user education. Thirty (62.5%) assessed MI quality, generally suggesting MI-consistent interactions. User perceptions were favorable, especially empathy, usability, helpfulness, and intention to use, though measures were heterogeneous. Eighteen (37.5%) reported intervention outcomes, mostly after a single session. Positive findings were more consistent for short-term motivation than sustained behavioral or functional change. GenAI-MI chatbots can deliver MI-consistent interactions perceived favorably, but evidence for sustained behavioral or functional change is limited. Future research should strengthen runtime safety monitoring, standardize MI quality assessment, and use longer-term comparative designs with behavioral and functional outcomes.

We introduce a new class of backward logarithmic orthogonal functions and generalized backward logarithmic orthogonal functions, constructed by applying a terminal-endpoint logarithmic mapping to generalized Laguerre polynomials. These functions are designed for backward spectral approximations of problems whose solutions exhibit weak singularities at the terminal endpoint. The proposed basis functions generate non-polynomial weighted approximation spaces with nodes naturally clustered near the singular endpoint, and therefore provide an effective framework for resolving algebraic and logarithmic endpoint singularities. We develop the basic approximation theory for these backward logarithmic orthogonal functions, including recurrence relations, derivative formulas, orthogonality, Sturm–Liouville characterization, mapped Laguerre–Gauss quadrature rules, weighted projection estimates, backward Lagrange interpolation estimates, inverse inequalities, and stability properties in weighted Sobolev-type spaces defined through a terminal logarithmic pseudo-derivative. A generalized version of the basis is also introduced by incorporating an algebraic scaling parameter, which improves the flexibility of the approximation space and allows singular factors to be represented more effectively. The error analysis and numerical results show that the proposed backward logarithmic basis is particularly suitable for weakly regular functions with terminal-endpoint singularities and can recover exponential or high-order convergence rates that are typically lost when usual polynomial approximations are applied directly.

Satellite communications are an essential component of sixth-generation (6G) mobile networks, which provide ubiquitous connectivity for global services. However, the satellite uplink remains a critical bottleneck for ground devices: their limited transmit power and antenna apertures result in low data rates and high packet errors. To overcome this bottleneck, this paper advocates a novel relaying paradigm termed generate-and-forward (GF) communications, where satellites exploit on-orbit generative artificial intelligence (AI) to robustly reconstruct corrupted data prior to forwarding. Specifically, we propose SpaceDiffusion, an over-the-orbit diffusion framework for satellite-assisted image transmission. The core of this framework is a channel-distortion-aware diffusion theory developed using the following approach. By formulating the recovery of compressed and lost image tokens as an inverse problem, this theory incorporates a channel-distortion correction term directly into the conventional denoising diffusion implicit model (DDIM) update. As a result, this design enables a single pretrained diffusion model to adapt dynamically to varying packet-loss patterns and compression distortions without retraining. Furthermore, we analytically characterize the progressive token-reconstruction error and derive a diffusion-step activation threshold that predicts when SpaceDiffusion is expected to outperform conventional decode-and-forward (DF) relaying. Building on these theoretical insights, we further develop an energy-aware early-exit policy to efficiently deploy SpaceDiffusion in orbit. Experimental results demonstrate that SpaceDiffusion achieves lower end-to-end latency compared to DF scheme with retransmission protocol and saves approximately 15 dB of uplink transmit power at a target perceptual quality.

We study a minimal classification problem: Given independent labeled observations $X\sim P$ and $Z\sim Q$ from two unknown distributions $P,Q$, and given an independent target $Y$ drawn with equal probability from $P$ or $Q$, can one classify $Y$ strictly better than chance whenever $P\neq Q$? The one-nearest-neighbor rule succeeds for every pair of multivariate Gaussian distributions with distinct means and a common positive-definite covariance matrix but can perform strictly worse than chance even for smooth densities on the real line. We construct a fixed randomized kernel rule whose expected accuracy is exactly $1/2+\operatorname{MMD}_k^2(P,Q)/4$, and obtain characteristic kernels on countably generated measurable spaces from countable families of measurable binary questions. We also prove that a deterministic order rule on $\mathbb R$ beats chance for every pair of distinct Borel probability measures. A measurable encoding then gives a deterministic distribution-free rule which beats chance on every countably generated measurable space, in particular every separable metric space. Finally, we show that no rule works for every distinct pair of distributions and every unknown unbalanced class prior; under adaptive target-class selection, every rule other than a fair coin is strictly worse than chance for some finitely supported pair.

Compact summaries are a key tool for approximate query processing over large datasets. For range-query workloads, an $\varepsilon$-net provides a small summary that hits every sufficiently large range. However, classical $\varepsilon$-nets only guarantee range validity and do not control the group composition of the selected tuples. As a result, the summary may be range-valid but poorly representative, which can propagate imbalance to downstream query results. Motivated by recent work on fair $\varepsilon$-nets and fair geometric hitting sets \cite{dehghankar2025fair}, we study fairness-aware range summaries under prescribed target group ratios. Different from previous sample-and-repair approach, we propose a target-stratified sampling method. For demographic parity (in which the ratio of fairness is determined by group proportion), our sample size is $O(A_{\varepsilon})$, coinciding with the standard $\varepsilon$-net bound, improving previous bound of $O\!\left(A_\varepsilon\log\frac{k}{\varphi}\right)$. For custom-ratio targets (in which the ratio of fairness is determined by manually defined proportion), our sample size is $O(A_{\Gamma})$, where $\Gamma$ is a parameter measuring the gap between the customized ratio and the demographic parity; we prove that this dependence on $\Gamma$ is unavoidable, with a worst-case lower bound of $\Omega(\Gamma/\varepsilon)$. Using our target-stratified sampling method, we could improve the previous approximation ratio for the fair geometric hitting set problem by a logarithmic factor, and making use of this result, we could in turn improve the size of custom-ratio fair $\varepsilon$-net. Experiments on real and synthetic datasets demonstrate that our method constructs smaller fair summaries than existing approaches, scales to large datasets and fine-grained group constraints, and improves downstream range query processing.

We analyze the composer embeddings learned by a composer-conditioned Transformer as a continuous latent space of compositional style, rather than merely as an internal representation for generation. A model that recursively predicts melody continuations was trained on melodic sequences extracted from MIDI data, conditioned on composer identity (124 composers). Principal component analysis of the learned composer embedding matrix (124x128) shows that the first principal component correlates strongly with composer birth year (r = -0.884, p < 0.001, n = 123), a stronger correlation than we obtain by applying the same PC1-birth-year analysis to existing general-purpose audio-text embeddings (CLAP, MuQ-MuLan) trained on unrelated audio-text corpora, not on symbolic melody generation. A shuffle test (2,000 permutations) confirms that the Silhouette score for stylistic-period labels is statistically significant (0.0110, p < 0.001). We further show that vector arithmetic in the embedding space captures meaningful stylistic relationships between composers. These results suggest that composer embeddings, learned without supervision beyond composer identity, form an interpretable latent space that captures musical-historical structure.

Closed-loop visual servoing requires predictions that indicate whether an action reduces task error, not only whether the action is plausible. We call this gap the prediction-control mismatch and introduce WM-VS, a target-centric progress-aligned world-model framework for closed-loop visual servoing. Offline target-region DINOv2 correspondences define a signed four-dimensional servo coordinate for translation, scale, and in-plane rotation. Stage 1 aligns action-conditioned latent transitions with this coordinate; Stage 2 freezes the world model and trains a reactive joint-velocity policy with action imitation, consequence supervision, and short imagined rollouts that favor error contraction. Deployment is RGB-only and reactive, without online trajectory optimization. On a real 7-DoF eye-to-hand system, WM-VS reaches a corner RMSE no larger than 10 percent of its initial value in 30/30 trials and retains this criterion at the final valid frame in 25/30 (83.33 percent). Removing future-error alignment reduces retention to 26.67 percent. The learned progress signal agrees with an external AprilTag corner error not used for training or control (mean Spearman rho = 0.8778). Without retraining, two unseen 3D targets achieve translation-error reductions of 86.48 percent and 90.27 percent and rotation-error reductions of 70.01 percent and 65.70 percent. These results link progress-aligned action consequences to repeated closed-loop correction and transfer. Code and data will be released as open source.

Coding agents have emerged as a promising paradigm for robot manipulation: a language model writes the robot controller as a program, and agents built in this way now operate robots without robot-specific training. Whether this paradigm is also safe, however, has not been asked. We evaluate coding agents under a safety constraint, where each task pairs a manipulation goal with an obstacle the robot must not touch. The agent pursues the goal but collides with the obstacle in most cases, treating task completion as its sole objective. The agent reasons about the obstacle in its traces, and the prompt already forbids touching it, so neither perception nor instruction is at fault; the fault lies in the planning, where the stated constraint never becomes a priority. By decomposing manipulation into a route phase and a contact-rich moment, we locate the source of the failure. Along the route, the model cannot prioritize the safety constraint, having no notion of a clearing route and none of replanning once a chosen route becomes infeasible. At the contact, it is unaware that contact execution is bounded by the same constraint. To close this gap, we present SafeHarness, which equips the model with two obstacle-aware harnesses. Obstacle-aware route planning grounds the objects as bounding boxes and draws candidate routes over them as sequences of waypoints. The agent then plans a route in advance, verifies it, replans when necessary, and only then executes it. Obstacle-aware contact execution instead selects the contact position so that the contact itself avoids the obstacle. SafeHarness attains 81.2% task success and 91.9% collision avoidance with GPT-6-Astra, surpassing the previous SOTA by 13.7 and 23.0 points, and the same agent without harnesses by 31.2 and 57.5 points, respectively.

Embedding spaces define notions of semantic similarity and distance. We study whether those embeddings reflect physical measurements of mass, distance, time and volume, which admit a unique, objective notion of semantic equivalence and distance. We find that physical measurement is only weakly modeled in the embedding space, and that instead quite peculiar measurement patterns can be observed. Further analysis indicates that embedding representations of physical measurements are strongly influenced by superficial string similarity, and recalibration of similarity does not substantially improve the alignment.

Complex robotic manipulation tasks frequently require a long-term memory of past events and actions. As conditioning on full histories renders policies prone to spurious correlations and degrades performance, many approaches to policy memory involve compressing historical information through expensive VLM queries in-the-loop to process only task-salient information. In this paper, we propose an alternative approach in which computationally intensive VLM queries are made during train-time to learn a lightweight latent memory that can be efficiently queried at deployment time. Our representation, which we call the workspace token, is trained by (1) using a VLM to identify current and historical information necessary for completing a task, then (2) distilling these into the workspace token using a set-reconstruction decoder loss. In both simulation and hardware, we show that the workspace token can be used as a drop-in replacement for observations during deployment, enabling policies to solve memory-intensive tasks without the need for VLM reasoning in-the-loop, in effect serving as a latent harness for distilling a stronger reasoning models ability to solve long-horizon tasks to a reactive robotic policy. We further demonstrate that the workspace tokens are not only more lightweight, but also lead to better policy performance compared to conditioning policies on explicit modalities like curated past image frames, motivating a latent approach to history curation and reasoning model harnesses more broadly.

Perceiving and remembering the visual world is fundamental to navigating and interacting with our environment. Current 4D foundation models, such as camera-controllable video models or 4D reconstruction models, can perceive and reconstruct dynamic environments, but how well they remember what they have perceived remains an open question. Existing benchmarks largely rely on pixel-level metrics and lack ground truth for objects once they leave the field of view, making them unable to evaluate visual memory in an object-centric manner against references. To fill this gap, we introduce PersistBench, a dataset and metric suite that leverages 360{\deg} videos as omniscient ground truth and proposes three evaluation aspects: object permanence, motion continuity, and appearance preservation. Evaluating various models across diverse categories reveals that current models can only maintain short-term consistency that degrades significantly once objects leave the field of view. Our findings highlight the gap between current model capabilities and robust visual memory ("seeing is not remembering"), providing guidance for future development of 4D foundation models. Dataset and code are available on the project page: https://guangzhaohe.com/persistbench.

A splash lives for a fraction of a second: sheets tear into ligaments and droplets, appearance is view-dependent and nearly textureless, and little persists long enough to track. Reconstruction research has consequently focused on smoke, synthetic liquids, or gently deforming surfaces. To our knowledge, no synchronized multi-view dataset of splashing liquids exists. We therefore introduce a benchmark of 20 real scenes, from coherent streams to violent splashes, captured by seven synchronized, calibrated 4K cameras at 60 fps, with manually refined per-view liquid and container masks and fixed evaluation splits. We further present SplashSplat, built on a single principle: impose physical structure only where the observations can constrain it. Per-frame liquid SDFs fused from the masks provide the geometry, level-set transport between consecutive SDFs yields a coarse velocity field, and Lagrangian carriers advected along this flow, corrected against each new observation and reseeded where coverage is lost, decode local Gaussians for differentiable rendering. SplashSplat outperforms state-of-the-art dynamic Gaussian splatting methods on our real captures and on a synthetic benchmark, with physically more plausible motion and a lower training cost. The same representation supports temporal interpolation and style transfer without re-optimization.