Papers

Polyethylene (PE) is one of the most commonly used synthetic polymers. While the synthesis and processing protocols for PE are well established, precise experimental assignment of microscopic structures at atomistic resolution (i.e., the position of each atom) remains largely limited to highly crystalline systems. This gap is often addressed via computer simulations using empirical interatomic potentials, which use approximate but efficient descriptions of interatomic interactions to reach the length and time scales needed to describe macromolecules. These empirical potentials typically perform well for bulk and/or collective properties but face challenges with chemical realism for complex systems, e.g., during reactive processes. In this work, we address this challenge by combining the computational efficiency of a deep potential (DP) machine-learning force field and the chemical realism of first-principles van der Waals (vdW) corrected hybrid density functional theory (DFT) enabled by a SeA high-throughput framework. Using this approach, we study the structure and dynamics of PE oligomers and polymers in an ethylene solvent under common high-pressure (supercritical) radical polymerization conditions. We found that the local solvation environment of radical-containing PE oligomers converges for chain lengths greater than (n~6), suggesting extensibility of our oligomer-trained MLFF to significantly longer polymers. We then confirmed the extensibility of these models to long PE chains by characterizing the molecular weight scaling of single-chain structure and dynamics, which showed classic good solvent behavior. Our PE MLFF retained a consistent level of fidelity and stability across a wide range of thermodynamic state points and chain lengths, at full atomistic resolution, therefore paving the way towards first-principles-based polymer structure and property prediction.

Vision-Language-Action (VLA) models are typically trained with behavior cloning (BC) on expert demonstrations. However, BC provides only positive supervision for expert actions, without explicit negative supervision indicating which actions are instruction-inconsistent or otherwise inappropriate. Reinforcement learning (RL) can provide such corrective signals, but often relies on externally specified rewards or curated non-expert data, both of which are costly to obtain in robotics. We show that offline RL for VLA models need not rely on curated non-expert trajectories: successful expert demonstrations alone can be transformed into dense corrective supervision through instruction relabeling. Specifically, by pairing expert actions with mismatched alternative instructions, we synthesize counterfactual instruction-observation-action tuples from the dataset and combine them with adversarial discriminator training to learn an instruction-grounded reward model for offline RL, without collecting additional rollouts or annotations. On the robustness-focused LIBERO-PRO benchmark, our method improves robustness to object position and task perturbations over a strong state-of-the-art baseline. It also outperforms competitive baselines in real-robot experiments on the TX-G2 (compatible with AGIBot G2). More broadly, our results suggest that, for data-constrained VLA learning, extracting denser supervision from each demonstration can complement collecting additional data.

We study the rounding of a supplied half-integral feasible solution of the root-assignment bidirected cut relaxation for Steiner Forest (Forest-BCR). Byrka, Grandoni, and Traub [IPCO 2025] proved a $16/9$ guarantee for a recursive framework that normalizes the LP point, selects a vertex set of maximum projected LP density, buys a minimum spanning tree on that set, contracts it, and recurses. We prove that the same framework has guarantee $8/5$. The new analysis keeps the orientation and the root label of each projected half-unit of LP mass. In a simple projection, the cut constraints at a terminal of degree two determine the root-assignment vector of every demand incident with it, and half-integrality leaves only two possibilities: a unit assignment to one root, which forces excess outdegree inside that root's support, or a split assignment to two roots, which forces overlap between their supports. For every connected component $C$ of the split-root graph this yields $ \beta_C\ \ge\ \frac{L_C}{2}$, where $\beta_C$ is the circuit rank of the union of the root supports in $C$ and $L_C$ is the number of its vertices of degree two in the full projection. Balancing the density certificate obtained from this inequality against the ordinary degree sum gives a vertex set of density at least $5/8$, and the inherited contraction lemma turns that into the $8/5$ rounding. For every $q\ge3$ we also construct a normalized half-integral point whose maximum projected density is exactly $5q/[2(4q-1)]$, so the universal projected-density bound is asymptotically tight.

Rare unsafe behaviors in large language models can remain practically significant even when their probability is extremely small, particularly at deployment scales involving millions or billions of interactions. Twisted Sequential Monte Carlo (SMC) provides a principled framework for rare-event probability estimation by learning twist functions that guide generation toward a target event. However, the standard twist learning framework relies on positive samples from the rare-event target distribution, which may be nearly absent before an informative twist has been learned, resulting in unreliable rare-event estimation. We propose Adaptive Multilevel Twisted SMC, which learns the rare-event twist through a sequence of progressively rarer intermediate events. At each level, the learned twist provides more informative positive examples for learning the next twist, ultimately leading to a more accurate final twist for the target rare event. Experiments across diverse tasks and model scales show that the proposed method produces more accurate rare-event probability estimates. By enabling more reliable discovery of hard-to-observe unsafe behaviors, our method provides a practical tool for strengthening the evaluation and safety alignment of deployed language models.

This paper presents a safety-critical bilateral teleoperation framework for omnidirectional aerial manipulators that integrates visual and force-sensorless haptic wrench feedback. Unlike existing approaches that either rely on onboard force/torque sensors or use model-dependent wrench estimates, which may become unreliable under model uncertainties or induce unintended feedback during free-flight, our method implements a hierarchical safety filter based on control barrier functions to avoid such limitations. The safety filter, being the key contribution, explicitly accounts for tracking errors arising from physical interaction between the aerial manipulator and its surroundings while enforcing thrust limits, a factor overlooked despite its critical importance for flight safety. This safety filter adjusts the command from the operator to ensure safe and stable aerial manipulation and avoid motor saturation. The adjustment made by the filter is mapped to haptic feedback, which is intuitive to the operator and conveys information on physical interaction and impending motor saturation. By actual experiments with a hexarotor-based omnidirectional aerial manipulator, we demonstrate that the proposed method avoids haptic feedback during free-flight, provides directionally consistent feedback under physical interaction, and can be operated for diverse manipulative tasks. Moreover, an ablation study further shows that the saturation filter improves interaction stability by explicitly preventing motor saturation and informing the operator of corrective actions.

Missing lane-count observations can distort pressure-based signal decisions even when neighboring detectors remain operational. We propose a lane-movement probabilistic graph neural network (LMP-GNN) that uses the movement relations involved in pressure computation to predict a mean and standard deviation for each lane. Three rules convert these outputs into replacement counts using the mean alone, a fixed uncertainty discount, or a staleness-dependent discount. Observed counts remain unchanged, and the completed state is supplied to an unchanged Signed Max-Pressure controller. Evaluation covers reconstruction and uncertainty calibration, decision-time diagnostics, and closed-loop traffic performance. Across 4,333,392 masked lane events, reconstruction achieved a mean absolute error of 0.7873 vehicles per lane. In a separate stored-trace audit of 372 network-outage-seed cells, pressure-score error was strongly associated with phase disagreement, with a Spearman correlation of 0.929, identifying pressure fidelity as a key decision-level diagnostic. Across five fixed-demand CityFlow networks, the fixed-discount rule reduced accrued average travel time by up to 13.74% relative to road-level mean imputation under correlated missingness. It also reduced travel time at 60% random missingness, whereas mean imputation performed better at 80% and 90%. The selected model has 63,362 parameters and a median single-thread inference time of 0.983 ms on a central processing unit. These results support lightweight probabilistic lane reconstruction as a practical input to pressure-based control, with traffic benefits that depend on the missingness regime.

Emerging switched CXL memory pooling systems, albeit promising, suffer from significant performance interference due to the shared but performance-uncontrolled data path among concurrent memory streams between a host core and a remote DIMM. We systematically characterize a memory pooling appliance based on XConn's Apollo CXL switch and identify three issues: intra-host contention, in-fabric congestion, and unmanaged host-remote DIMM interaction. This paper presents a new transport layer, MemChannel, which provides the mchannel abstraction to manage end-to-end fabric bandwidth among competing memory flows and enable application-specific traffic for switched CXL memory pooling. Our key idea is to build a sender-driven, fabric-informed transport protocol, inspired by Core-Stateless Fair Queueing (CSFQ), that admits just the right amount of CXL requests to each mchannel based on the estimated core-to-CXL-DIMM bandwidth availability. To address CXL-induced idiosyncrasies, MemChannel introduces time-based rate control, host-side admission control, cross-host bookkeeping, new congestion signals, rate estimation based on the fluid model, and delay-based link-capacity adjustment. We build MemChannel from scratch and support unmodified applications. Evaluations over switched memory pooling demonstrate its effectiveness from performance-isolation, scalability, and multi-tenancy perspectives.

We study low-degree testing for group-valued functions over a Boolean slice. Specifically given a degree parameter $d$ and oracle access to a function $f:\{0,1\}^n_{n/2}\to G$ where $\{0,1\}^n_k$ denotes the set of vectors in $\{0,1\}^n$ of Hamming weight $k$ and $G$ is an Abelian group, the low-degree testing problem asks us to distinguish the case where $f$ is a polynomial of degree at most $d$ (with coefficients from $G$) or is $\varepsilon$-far from the set of all such polynomials. Classical works in this area considered functions with domain $\mathbb{F}_q^n$ and range $\mathbb{F}_q$. More recent works have considered the setting where the domain is the Boolean cube [Bafna, Srinivasan, Sudan (Random Struct. Algorithms 2020), Amireddy, Srinivasan, Sudan (RANDOM 2023)], or when the domain is the slice (i.e., $\{0,1\}^n_{k}$) and the range is $\mathbb{F}_2$ [David, Dinur, Goldenberg, Kindler and Shinkar (SIAM J. Comput. 2017), Kalai, Lifshitz, Minzer and Ziegler (FOCS 2024)]. Each of the changes introduces new challenges in designing and analyzing low-degree tests and this happens again in our setting with domain being a slice and range is general. Our main theorem gives a test that makes $O_d(1)$ queries to $f$ and accepts degree-$d$ functions while rejecting functions that are $\varepsilon$-far with probability $\Omega(\varepsilon)$. The central proof idea is to reduce this low-degree testing problem to the problem of low-degree testing on the cube. Specifically we show how to randomly embed the $n/2$-dimensional cube $\{0,1\}^{n/2}$ in the $n$-dimensional slice while nearly preserving the proximity of $f$ to the space of degree-$d$ polynomials on this cube. While the embedding is simple and natural, the analysis involves a careful induction with a novel use of a basis of degree-$d$ polynomials on slices (from a work of Anstee, Rónyai and Sali (Graphs and Combinatorics 2002)).

Simulation-Based Inference (SBI) serves as a vital framework for parameter inference in scientific fields where simulators involve intractable likelihoods, yet while amortized generative models offer rapid posterior estimation, they are often restricted by the specific priors used during training, thereby limiting their flexibility as prior knowledge evolves. To address this prior dependency, PriorGuide was introduced as an inference-time guidance method, but due to its intractable formulation, it relies on Gaussian approximations of the reverse transition kernel and Gaussian mixture model fitting for the prior ratio, both of which introduce systematic bias. Motivated by these limitations, we propose an unbiased test-time guidance framework that leverages Density Ratio Estimation (DRE) to learn a score guidance term, effectively decoupling the inference process from the prior training. Moreover, our framework remains agnostic to the specific density ratio estimators, making it a general and flexible framework for handling prior changes. Experimental results across multiple tasks demonstrate that our method matches or outperforms PriorGuide on C2ST and MMD in most tasks while maintaining robustness even under limited overlap between the training and target priors. Furthermore, we apply our method to Bayesian updating for parameter inference from planetary light-curve data, where it also demonstrates strong effectiveness and robustness. Code is available at https://github.com/a-chenchen/dre-based-prior-guidance .

Reinforcement learning with verifiable rewards (RLVR) is deployed to make models better at reasoning tasks, but its side effect on what models will divulge is under studied. Here we show that RLVR on facts increases extraction of personally identifiable information (PII) the instruct model had already memorized. We first confirm that instruct models have already memorized PII but leave them latent, rarely surfacing one when asked. We then apply RL on benign factual data that contains no PII of any kind, and re-probe: a targeted probe over name->email pairs, and an untargeted free-recall prompt that simply asks the model to list the addresses it knows. PII extraction rises sharply under both: on DeepSeek-V3.1, verbatim recall@k increases from 0.155 to 0.370, a 2.4x gain. The effect scales with model size: across three models spanning 8B to 671B parameters, absolute leakage is largest in the biggest model. Meanwhile model's reasoning abilities and refusal rates are retained, indicating that RL selectively changes which memorized information is accessible rather than broadly altering the model. In summary, memorized private data can be made markedly more extractable by training that never touches it. This gives an adversary a route to memorized data that requires no privacy-relevant training signal and no access to the data itself – only the ability to fine-tune on something innocuous.

We introduce a continuous data assimilation method for particle-in-cell simulations of the Vlasov-Poisson equation when only hydrodynamic moments are observed. The forecast state is an empirical measure on phase space, whereas the observed fields (density, bulk velocity, and temperature) constrain only a few velocity moments and leave the velocity-space shape of the distribution undetermined. We construct the moment feedback as a Wasserstein gradient flow of a moment-mismatch functional over phase-space measures. The resulting drift acts directly on particle positions and velocities, couples the density, momentum, and energy residuals through a single variational structure, and vanishes on the entire moment-compatible set. Under the standard Wasserstein metric, the energy residual produces a position correction that grows quadratically with the particle speed, and the particle system falls outside standard well-posedness theory. Our primary formulation pairs the quadratic moment mismatch with a velocity-weighted Wasserstein metric that penalizes spatial transport at large peculiar velocity relative to the observed bulk flow, which removes this growth. A direction-split variant retains the plain metric instead. Under the same weighted metric, an alternative moment-relative-entropy functional yields an affine, shape-preserving velocity correction and explicit global moment balances for the space-inhomogeneous system. We prove that the finite-particle scheme with linear Lenard-Bernstein collisions is globally well posed. In 1D1V and 2D2V experiments with several collision models, the nudged formulations reduce bulk-velocity and temperature errors by up to two orders of magnitude relative to an unassimilated run.

Correcting health misinformation in dialogue requires more than producing a factual rebuttal: users differ in what they know, what they believe, and what they need to hear, so an effective intervention often depends on first asking the right clarifying question. Yet existing methods either respond immediately or probe indiscriminately, treating clarification as either unnecessary or always beneficial. We propose Reward-Optimized Probe-and-Respond (RO-PnR), a framework that learns when asking is worth its cost. At each turn, RO-PnR chooses between probing for more information and committing to a final correction, guided by a turn-level reward that weighs the expected gain from probing against its interaction cost. To capture how user heterogeneity affects probing value, we model each simulated user with a latent state along health literacy and belief commitment. Experiments show that RO-PnR achieves the highest cost-adjusted utility across three health-misinformation datasets and three base models, using 30% fewer turns than always-probe baselines.

AI inference clusters are increasingly constrained by instantaneous power, not just energy: grid operators condition new capacity on demand response, imposing time-varying power caps. Existing LLM serving systems optimize a static energy objective or shed fixed priority tiers under load; either way, goodput collapses when the power envelope moves. An LLM pipeline is not a uniform load: compute-bound prefill loses throughput almost linearly with GPU frequency, memory-bound answer decode sustains it down to $0.57\times$ nominal, and reasoning's thinking phase couples KV-cache capacity to scheduling – so a cap should be steered to where each watt costs the least performance. PowerSlider does so with a new Flex SLO contract that turns bounded user slack into an optimization constraint, prefill–think–answer disaggregation exposing per-stage frequency and KV control, and a Karush–Kuhn–Tucker (KKT) online solver re-solving within 7.7 ms of every cap change, backed by a consolidated fail-safe that power-gates drained instances when DVFS bottoms out on static power. On SGLang with production traces, \sys{} sustains 78.3% online goodput at a 30% cap reduction versus 47.6% for the best of five baselines ($1.64\times$), holds latency-critical tails within $1.3\times$ of nominal (baselines: $2.3$–$6\times$, up to $12\times$), and delivers 92% mean goodput through a replayed CAISO grid-emergency day bottoming at $0.41\times$ (54% at the trough; every baseline below 7%).

As an emerging technology, Extended Reality provides end-users with an immersive experience of interacting with virtual and physical environments. Unlike traditional software, the execution of XR applications involves more computationally complex operations, such as 3D scene rendering, real-time animation, and process simulations. Inefficient coding practices during the software development of XR applications may cause various performance bugs, degrading user experience and even causing motion sickness. Thus, it is an urgent need to develop an automated program repair framework for fixing performance bugs in complex XR programs. However, it is non-trivial to achieve this goal due to several technical challenges: (1) a lack of a real-world XR codebase and bug dataset, (2) no accurate bug detection tool, and (3) no effective bug-fixing tool designed for XR performance bugs. To tackle these challenges, we present a novel large language model-based framework, namely XRFix, to repair performance bugs for open-source XR programs. We first construct a corpus of domain-specific performance bugs built with a codebase from 23 open-source XR projects and a dataset of XR-related bugs containing 104 real-world bugs. Then, we tailor two static analysis tools for accurately detecting bugs in both C# scripts and asset files. Last, we design different prompts to instruct LLMs to fix XR bugs in three types of bug scenarios with different complexities, i.e., single-line level, function level, and class level. We conduct extensive experiments on five off-the-shelf LLMs to evaluate the bug-fixing performance of XRFix. We also compare our XRFix with three SOTA APR approaches. Through static analysis, reference answer comparison, and manual inspection, we demonstrate that our XRFix can effectively fix XR bugs, outperforming SOTA APR methods.

Recent page-image retrievers such as ColPali have improved retrieval over visually rich documents, yet little is known about how they behave in cross-lingual, low-resource settings. We introduce LëtzCross, a benchmark for cross-lingual page-level retrieval over Luxembourgish PDF documents, with document pages indexed as images and queries provided in English, French, German, and Luxembourgish. The benchmark combines text-focused QA pairs with visually grounded QA pairs, covering both textual and visual retrieval needs in PDF-based RAG. We use LëtzCross to compare OCR-based text-only retrievers with ColPali-style page-image retrievers and find that the latter perform better across query languages in this system-level comparison. We also examine single-language and multilingual fine-tuning. Fine-tuning transfers across query languages, with French yielding the highest mean performance on Luxembourgish queries among the single-language settings. In the multilingual setting, including Luxembourgish gives the strongest results and substantially improves retrieval for Luxembourgish queries.

Reasoning-augmented text-to-image models such as GoT-R1 emit an explicit textual plan - object names, attributes, and bounding boxes - before generating image tokens. When such a model fails a compositional prompt, is the plan wrong, or is the plan right and the decoder unfaithful? Because the plan is machine-readable it can be edited before decoding, which makes the two separable. We first validate the ruler. Swapping the two bounding boxes inside the model's own chain demonstrably flips the generated layout: detector-based accuracy falls 0.75 -> 0.48 (p<1e-3), while a widely used VQA-based spatial metric rises. A five-rater human study agrees with the detector on 81% of items and with the VQA judge on 57%. All spatial results therefore use geometric scoring. Under sound measurement the decoder is a faithful executor: 94% of generated layouts realize the planned relation, and object-box binding survives reordering of the plan's object segments. The planner is the bottleneck. It writes wrong relations for phrasing-dependent reasons - 98% accuracy on "left" against 54% on "right" for semantically identical layouts, a raster-order bias we isolate with a mention-order control - and cluttered geometry that the decoder faithfully reproduces. Editing the plan therefore fixes the image without retraining: symbolic verification with resampling gives +5.0 points (p<1e-3), minimal in-place repair +6.0 (p=.02), rewriting only box geometry +10.7 (p<1e-4), and replacing the plan outright +13.3 (p=1e-4). Gains are indifferent to the plan's prose style and to its likelihood under the planner, but not to its geometry. Modular planner-decoder designs are therefore viable, provided the plan is internally consistent: box-text contradictions induce object duplication and identity fusion. We release the plan-fidelity evaluation protocol, all plans, and 12k generated images.

Transformer architectures for clinical prediction from electronic health records (EHRs) often require costly manual tuning and may vary across hospitals and tasks. We introduce ATHENA (Agentic Transfer across Hospitals for EHR Neural Architecture Search), a knowledge-guided agentic neural architecture search framework for Transformer-based EHR modeling. A weight-sharing supernet is pretrained once per hospital, allowing candidate architectures to inherit weights and be evaluated through fine-tuning rather than independent pretraining. A two-layer cross-hospital prior supplies high-performing source architecture examples and task-specific component preferences summarized by SHapley Additive exPlanations (SHAP)-based meta-regression. These priors guide a multi-agent large language model search using target-hospital validation feedback. Across six clinical prediction tasks at one held-out OneFlorida+ site and one external MIMIC-IV site, ATHENA significantly outperforms all four baselines in 9 of 12 site-task evaluations under a strict equal-compute comparison. In a separate comparison using a common pretrained AutoFormer supernet to isolate search strategy performance, ATHENA ranks first in 9 of 12 evaluations at a maximum search budget of 30 candidate evaluations. It also shows more consistent architecture selection across repeated searches. ATHENA supports more efficient architecture selection for Transformer-based clinical prediction models.

With the advance of deep neural networks, the quality of disparity maps obtained through stereo matching has steadily improved. However, existing stereo matching methods still struggle to preserve fine-grained geometric details, resulting in blurred edges and over-smoothed predictions in challenging regions. To address these limitations, we propose StereoDiffuer, an iterative diffusion-based stereo matching framework that explicitly models geometric details and progressively refines disparity estimates. The framework incorporates a Saliency Attention Perception (SAP) module to extract salient geometric cues, including object boundaries, thin structures, and sharp edges. Confidence-guided SAP features are combined with the initial disparity estimate to condition an iterative denoising diffusion process, which corrects residual disparity errors and restores geometric details suppressed during cost-volume regularization and upsampling. Experimental results on the Scene Flow and KITTI benchmarks demonstrate the effectiveness of the proposed framework and its competitive performance relative to the compared stereo matching methods.

Diffusion models are widely used as priors for linear inverse problems, yet endpoint quality does not reveal when measurement information enters reverse denoising or how it is allocated across signal directions. We study this process through the smoothed likelihood force, the difference between exact posterior and prior scores at each noise level. For a fixed measurement, its expected squared norm gives both posterior–prior relative-entropy dissipation and reverse-path relative-entropy growth. Averaging over measurements yields an information–minimum mean-square error (I-MMSE) identity linking information gain to denoising-error reduction. Under finite second moments, the force energy and its ratio to prior-score energy decay quadratically in the noising kernel's signal coefficient at high noise. Solvable models show that conditioning removes class separation already explained by the measurement, reduces a uniform index entropy over \(n\) empirical samples from \(\log n\) to \(H(I\mid r)\), and makes assimilation depend on operator–prior alignment even for identical singular values. Experiments in models with tractable posteriors evaluate these predictions. In a separate illustration with a frozen FFHQ model, masks sharing the same spectrum yield different prior-normalized null-space trajectory statistics.

The randomized extended Kaczmarz method is an effective iterative framework for solving large-scale inconsistent linear systems. In this paper, we extend this framework to third-order inconsistent tensor linear systems under the t-product and propose the Tensor Greedy Double Block Extended Kaczmarz (TGDBEK) method. At each iteration, TGDBEK dynamically constructs active blocks of row and column slices via a residual-based greedy selection strategy, prioritizing the slices associated with the largest residual norms. Unlike existing tensor block Kaczmarz variants that rely on static, predefined partitions –the tensor randomized extended block Kaczmarz (TREBK) method, its greedy counterpart (TREGBK), and the tensor randomized extended average block Kaczmarz (TREABK) method – TGDBEK adapts the active block sizes dynamically at each step using a single intuitive threshold parameter $\eta$. We establish the theoretical linear convergence of TGDBEK to the unique minimum-norm least-squares solution $\mathcal{A}^\dagger * \mathcal{B}$. Extensive numerical benchmarks on synthetic dense and sparse tensor systems, as well as multidimensional multichannel color and 3D volumetric MRI image deblurring problems, demonstrate that TGDBEK substantially outperforms state-of-the-art tensor Kaczmarz solvers in both iteration count and CPU running time.

Retrieval-Augmented Generation (RAG) grounds LLM generation on retrieved documents, but the standard terminal retrieval stage–dense-vector similarity, optionally followed by reranking–often returns documents that share keywords with the query without containing the needed information, a failure mode that grows with the knowledge base. We trace it to a conceptual gap: similarity captures only associational relations, whereas the documents that matter are linked to the query causally. We model the terminal retrieval stage with a causal graph grounded in Reichenbach's common cause principle: the keywords shared by the query and a retrieved document form a latent common cause A, and the document's residual keywords form a latent set B linking the document to the ideal output. Since a retrieved document is a collider (A -> d <- B), retrieval itself opens an associational path between the query and B, which licenses a training-free, attention-style re-scoring rule: the cosine similarity between the query embedding and the weighted centroid embedding of B. Unlike causality-enhanced RAG variants that model causal relations inside the knowledge content, our graph models the causal structure of the retrieval process itself. On a real 471-document enterprise knowledge base, the method promotes a relevant guideline from rank 6 to the top 3; on a controlled diagnostic corpus reproducing the keyword-stuffing regime, it improves the mean target rank from 2.88 to 1.25, while a trained cross-encoder reranker barely helps (2.63). Conversely, on three BEIR benchmarks the score underperforms the similarity baseline, delineating the applicability boundary: the method guards the keyword-stuffing regime of growing proprietary knowledge bases and complements neural rerankers; a corpus-level calibration gate selects the correct regime with >= 95% reliability. A fully local testbed demonstrates deployability.

Continuous-time flow and diffusion models are widely used across many application domains, from large-scale deployment in computer vision and protein folding to emerging adoption for modeling language, time series, and quantum states. After training, inferring statistical properties from continuous-time models is costly. Wavefunction flows target this cost by recasting learned transport as unitary evolution, whose final Born distribution approximates the target distribution. This prepares a coherent amplitude encoding (a qsample) that can be post-processed by quantum algorithms offering a quadratic advantage over Monte Carlo sampling. We present the first numerical study of these flows, in which we represent time-dependent potentials and states as tensor networks. At spatial dimension $d=8$, storage falls by $\sim 10^7\times$ relative to the dense grid of $N^d$ points, and evolution wall-clock time falls by $\gtrsim 10^3\times$ against a baseline extrapolated from the measured $d\le 5$ scaling. We validate our pipeline by reproducing the $O(1/\sqrt{p_{\rm rare}})$ scaling of rare-event sampling.

Modern flapping-wing insect-scale air vehicles display agility similar to that of their insect counterparts; however, these impressive maneuvers are only possible with off-board sensors like optical tracking cameras. In this manuscript, we introduce two embedded proprioceptive sensors for insect-scale aerial robots: thin film piezoelectric polymers integrated directly into a driving actuator and a pitching hinge which track stroke and pitch angle, respectively. We fabricate the aforementioned size-agnostic mechanically intelligent structures (sensor-actuator, sensor-flexure) using laminate stack fabrication methods. Chirp experiments with our sensors integrated into an insect-size flapping-wing robot show accurate tracking of stroke (RMSE = 0.44 deg) and pitch (RMSE = 2.44 deg) angles in the relevant frequency range. As the first step towards demonstrating the utility of these sensors for enabling numerous onboard autonomy applications, including closed-loop wingbeat control and sensor fusion with existing insect-scale sensor suites for more accurate proprioception and localization, we show one application for each sensor. The proprioceptive hinge enables collision detection, reducing the chance of permanent damage if the robot's wing collides with an object. The proprioceptive actuator enables asynchronous flapping, which is hypothesized to increase adaptability and efficiency in insects and robots alike. A microrobot equipped with our proprioceptive actuator allows us to test these hypotheses with potential for improving flapping aerial robot performance. We foresee proprioceptive sensors having an important role in progressing both the fields of insect-scale aerial robots and robo-physics due to the bio-inspired nature and high integration level of our sensors.

Generating expressive facial behavior in virtual humans (VHs) remains a central challenge in affective computing and character animation. This paper presents a novel approach based on Conditional Variational Autoencoders (CVAEs), trained on real human facial expression data, to synthesize controllable emotional expressions at varying intensities. Using a dataset comprising six basic emotions represented at two intensity levels (low and high), we train a CVAE model to generate synthetic facial expression data while preserving semantic consistency with real human expressions. Despite the limited amount of training data (only 7,680 facial expression samples), the proposed approach learns meaningful latent representations and generates coherent emotional variations. Our method enables control over emotional intensity, making it suitable for animating virtual characters without requiring actor performances or manual artistic intervention. Our research aimed to evaluate whether the method (CVAE) preserves the characteristics associated with the different intensity levels present in the dataset. Results show that the proposed model preserves key expressive characteristics across intensity levels while supporting generalization across emotional intensity levels, contributing to the creation of emotionally expressive virtual characters from relatively small datasets.

Mixture-of-Experts (MoE) LLMs scale model capacity efficiently through sparse activation, but their large expert parameter footprint, routing imbalance, and long-context KV-cache growth make deployment difficult on commodity hardware. Practical deployment often requires stacking multiple compression techniques: expert pruning removes redundant experts, weight quantization lowers model memory footprint, and KV-cache compression reduces long-context memory pressure. However, these techniques are typically evaluated in isolation, leaving open how they interact when applied together in realistic deployment pipelines. In this work, we present MoEXBench, a systematic benchmark for evaluating composable MoE compression as an end-to-end deployment workflow. MoEXBench studies 10 MoE models ranging from 30B to 235B total parameters across standard-attention, hybrid linear-attention, and sliding window attention architectures. It evaluates 20%-50% expert pruning rates, 1 to 16 bit weight-quantization schemes, and multiple KV-cache precision settings, applied both individually and in combination. MoEXBench introduces an eight-module evaluation suite that jointly measures composable-compression quality, workload and architecture robustness, pruning/quantization/KV cache sensitivity, and deployment efficiency on commodity hardware. Our results reveal non-trivial interactions among compression methods: composable compression cannot be predicted from standalone techniques, compression rate alone does not reliably predict quality loss or runtime gain, expert pruning is the dominant degradation source, and average quality can hide workload and architecture-specific failures. By releasing normalized module scores, compressed artifacts, and reproducible scripts, MoEXBench enables practical accuracy-memory-latency comparison across MoE families and hardware backends.

Physics-Informed Neural Networks approximate the solution of boundary value problems by minimizing the strong-form residual, their accuracy is often limited near corner and crack singularities, sharp layers and localized features and the enforcement of Neumann boundary conditions remains challenging. In hp-Variational Physics-Informed Neural Networks, the problem is formulated in a Petrov-Galerkin weak form, reducing the differentiation order applied to the neural network, while boundary constraints are imposed through penalty terms. The variational residual is evaluated using fixed Gauss quadrature points and the strong-form residual on a fixed collocation set, which may fail to capture regions of rapid variation. To address this issue, we integrate hp-Variational Physics-Informed Neural Networks with Residual-based Adaptive Refinement. During training, the strong-form residual is computed on a dense candidate pool and points with the largest residuals are added adaptively to the collocation set. Since the weak formulation is defined in $H^{1}(\Omega)$, it avoids the $H^{2}(\Omega)$ regularity requirement that fails on cracked and non-convex domains. The proposed method is validated on four two-dimensional elliptic problems and consistently achieves substantial reductions in relative $L^{2}$, absolute $L^{\infty}$ and relative $H^{1}$ semi-norm errors through adaptive residual-driven point placement.

Joint-Embedding Predictive Architectures (JEPAs) learn world models by predicting future representations rather than reconstructing observations. Variational JEPA (VJEPA) gives this prediction probabilistic semantics and permits flexible conditional predictive families. We propose MDN-VJEPA, a multimodal probabilistic JEPA that instantiates the latent predictor as a mixture density network (MDN), representing future states with a Gaussian mixture whose components encode alternative futures, their probabilities, and within-mode uncertainty. For sequential prediction, we address two additional structural requirements: multi-modal futures should remain temporally coherent across a trajectory, and predictive beliefs should remain consistent when propagated to the same horizon through different temporal path compositions, as motivated by the Chapman–Kolmogorov equation. Together, these considerations define a three-level hierarchy of multimodal prediction, coherent multimodal trajectories, and compositionally consistent multimodal dynamics. Controlled experiments show that MDN-VJEPA closely recovers known branching predictive laws, that retaining the complete multimodal belief rather than its conditional mean can lead to materially different and substantially better planning decisions under branching futures, and that trajectory-level coupling and distributional consistency improve temporal coherence and multi-horizon agreement. These results support a simple principle: a probabilistic JEPA should preserve not only how uncertain the future is, but which distinct futures constitute that uncertainty, how they form coherent trajectories, and how those beliefs compose through time.

Representation quality is a central determinant of PINNs' performance, yet standard training leaves representations to emerge implicitly while fitting the final solution. We introduce representation curriculum, an ordered process in which representations are explicitly learned, transferred independently of their predictors, and progressively refined. We realize it with Staged Depth Training (SDT), which trains a shallow prefix under a temporary physics-informed head, discards the head, and freezes the learned prefix while adding depth, without equation-specific encodings or changes to the final architecture. Across the 20 default forward problems in PINNacle with three backbones, SDT improves 40 of 59 equal-budget problem–backbone cells by at least 5% and remains within that band in the rest, with a 32.8% geometric-mean error reduction on a PirateNet-style backbone. Mechanistic ablations suggest that the gain is not explained by optimizer restarts or shallow warm-starting alone. Representation visualizations and hyperparameter-basin analyses provide diagnostic evidence on representation geometry and local sensitivity to shared hyperparameters. On Poisson–Boltzmann 2D, SDT also more than doubles the fitted depth-scaling exponent for both backbones. These results support representation curriculum as a promising training strategy for improving PINNs while preserving the deployed architecture and inference cost.

A recorded dependency graph cannot certify that it contains no omitted edge. For learned artifacts, graph-scoped invalidation therefore cannot by itself justify admission after authority changes, especially when output testing misses a provenance-stale derivation. We separate recorded lineage, which proposes impact scope and a dependency explanation, from independent current-contract revalidation of every retained target, which supplies soundness relative to the declared contract. We evaluate this design in two evidence layers. Four Qwen3-14B LoRA packages bind 1.028 GB of adapter bytes to training and authority records. Stale and full-fresh adapters trained on mutually exclusive versioned supervision emitted identical plans on all 80 tested authority-neutral inputs. Byte-derived reconstruction and a sealed current-contract ledger nevertheless refused the stale and naive-append packages and admitted full-fresh and lineage-selective packages. In a controlled harness with six role-separated amendments, transitive lineage recovered all 40 semantically invalidated amendment-target pairs while nominating 56 impact candidates, compared with 120 under document-wide invalidation. All 56 had a declared path to an amended clause, although 16 remained content-valid. Omitting one load-bearing edge per amendment reduced semantic recall to 34/40 while independent revalidation refused all six affected packages. The results establish properties of a controlled governance mechanism, not an enterprise invalidation rate, unlearning result, operationally optimal workflow, or measured total-cost reduction.

Designing a high-order structure-preserving discretization is a constrained mathematical search: conservation, a positive discrete inner product, physical spectral behavior, boundary accuracy, bandwidth, and partial differential equation (PDE) error must hold simultaneously. We test whether large language models (LLMs) can help while remaining non-authoritative. The motivating MOLE implementation of the Corbino-Castillo staggered operators satisfies a general discrete Gauss identity and conserves exactly, yet its order-six and order-eight Dirichlet blocks develop four non-real boundary-localized modes. An endpoint-supported positive-definite identity would instead force a real non-positive spectrum, so the search changes the closure and norm architecture. An LLM proposes only a typed construction program; a deterministic linear-program compiler generates coefficients; an independent verifier tests algebra, positivity, physical modes, conditioning, and manufactured PDEs; and coupled rational reconstruction provides exact certificates. Across 1,200 solver evaluations, an externally fixed verifier accepted 55.0% of full-metric-feedback proposals and 53.3% of illumination-archive proposals, versus 13.3% for uniform random search. Four leading LLM-originated programs were reconstructed exactly. The strongest order-six-interior, order-four-boundary candidate lowers the prior positive-diagonal spectral-radius constant by 25.4% and its PDE error 62.7-fold. Its certified heat-equation energy is contractive, whereas the order-six reference exhibits 6.6% transient growth; both have the same RK4 stability limit. The LLM proposes structural hypotheses; deterministic mathematics determines validity.