Papers

Boson sampling with array of on-demand single photons (discreet variable Fock states) as resource states provides distinct advantages in complexity scaling compared to single photon resource states mimicked by the states (continuous variable) generated in nonlinear photonics. Recent progress in realization of regular ordered array of mesa top single quantum dots (MTSQDs) enabling on-demand generation of many photon resource states with high single photon purity and indistinguishability [1,2] has opened the possibility of implementing chip-scale Boson sampling with single photons. An important question towards scale up of such circuits is how the dephasing and decay timescale of individual on-demand single photons affect the final probability distribution of the N-photon entangled state. In this work we report a framework that provides direct link between the physical sources of decoherence to the outcome of multiphoton correlation measurements. We explicitly incorporate the symmetrization of the N-photon states while accounting for the shape of the individual wave packets and dephasing. Using this framework, we show predictive estimates of the probability of generation of a 4-photon NOON state using an optical circuit initialized with 4 identical single photon wave packets, mimicking the emitted photons from MTSQD single photon sources reported in [2]. Specifically, we show how the probability of generating various 4-photon entangled states depend on the two-photon interference visibility which is determined by the pure dephasing timescale in relation to the radiative decay timescale. The results provide quantitative guidelines to measure 4 photon and higher number of entangled photon states generated in an optical circuit with on demand single photon sources and provide a link between physical processes such as temperature and the output state of the quantum optical circuits.

Least squares summarizes the joint distribution of a response and regressor through second-order information. When an observed regressor is composed of latent components carrying different relationships with the response, this summary can average over distinct linear effects. We study the additional linear information contained in higher-order mixed cumulants and formalize it through the latent slope spectrum. Ordinary least squares is a variance-weighted barycenter of this spectrum, whereas higher-order cumulants can separate its components when the spectrum is sufficiently parsimonious. In the scalar one-confounder model, fourth-order cumulants identify the unordered pair $\{\beta,\beta+\gamma\}$ through a $2\times2$ linear system and a quadratic equation. We develop estimation and root-$n$ inference under regular separation and show that detection and estimation diverge near degeneracy: slope multiplicity remains detectable on an $n^{-1/4}$ local scale even when the individual slopes are not uniformly regularly estimable. In multiple regression, a quadratic tensor recurrence yields a test of one versus multiple effective coefficient vectors that does not require the vectors themselves to be identified; multiplicity can therefore be consistently detectable even when the recurrence operator is noninjective. The same quadratic diagnostic can retain signal in directions where a first-order higher-cumulant restriction cancels exactly. The results distinguish representation, identification, detection, and estimation of latent linear heterogeneity, while making explicit the limits of structural and causal interpretation from observational data alone.

A recommendation request on Reddit draws several replies, each naming different titles. When an LLM recommender is trained on such threads with reinforcement learning (RL), which replies should a generated list be rewarded against? The state-of-the-art recipe, Rank-GRPO, picks one reply per training example, so the advantage mostly records which reply was drawn (85% of its variance) rather than how good the list is. We argue that RL should train on the request, not on each answer, and propose ReMaRk: it scores every list once against the union of all replies' titles and samples each request equally. This makes RL cheaper. On seven backbones and three datasets, ReMaRk reaches the validation recall of 2,000 per-reply RL steps in 25 to 53% of the steps and leads at that budget; trained until neither improves, it is never significantly behind Rank-GRPO and in four of seven pairs reaches Rank-GRPO's best with fewer GPU-hours. The same reward helps on ProtoQA, where about 100 people answer each question, which points to a simple rule for RL with several valid human answers: reward them jointly.

The 3He(alpha,gamma)7Be reaction takes place in several distinct astrophysical scenarios. First, in solar hydrogen burning, 3He(alpha,gamma)7Be controls the competition between the pp-I chain, on the one hand, and the pp-II and -III chains, on the other hand. Its rate directly affects the solar model predictions for the well-measured 7Be and 8B solar neutrino fluxes, which can in principle be used as solar thermometer. Second, in Big Bang nucleosynthesis, 3He(alpha,gamma)7Be is the main production channel for primordial 7Li, the decay product of 7Be. Finally, the reaction is believed to produce radioactive 7Be in astrophysical novae. The cross section of this reaction has been studied previously using different methods, namely 7Be activation, 7Be recoil detection, and in-beam {\gamma}-ray detection. The latter studies have in many cases been performed with just one detector in close geometry. Here, for the first time a comprehensive measurement of the 3He({\alpha},{\gamma})7Be {\gamma}-ray angular distribution is reported. Altogether 22 high-purity germanium detectors were used, arranged at 27-145 degrees angle with respect to the beam direction. Data were taken at 442-1213 keV center-of-mass energy. In addition, one data point for the astrophysical S-factor was obtained with the activation method. The angular data are compared with the predictions from a hybrid theoretical approach, where the E1 transitions are modeled using a cluster-based effective field theory (EFT) while the M1 and E2 transitions within an EFT-inspired framework. Good agreement is found, if the interference among these multipolarities is taken into account in the theory.

We give two types of examples where two non isomorphic finite quandles have isomorphic enveloping groups. This implies that two non-isomorphic finite quandles can share isomorphic representation categories (as additive and tensor categories simultaneously).

A model trained on readers' highlights can show a newcomer where earlier readers converged on a page, even on pages with few readers. A year of stronger language models suggests two upgrades: use them directly, or use them as teachers. We compare both with a third, more reader data, on one social highlighting platform, with the comparisons, the selection rules and a sealed set of fresh pages fixed in advance. On 215 development pages, seven language models released by October 2026 rank sentences zero-shot at an average precision of at most 0.3744 against the sentences half the readers marked most, and 0.3931 when fused with reading position, while a 3-billion-parameter model trained on reader highlights alone reaches 0.4346. Used as the teacher that breaks ties among readers' counts, the stronger fusion changes a trained model by +0.0013 [-0.0012, +0.0040], whereas going from half to all of its training pages adds +0.0242 [+0.0179, +0.0301]. A 27-billion-parameter backbone on the reader-only recipe adds +0.0225 on the development pages; on 131 fresh pages, read once and too few for a confirmatory comparison, its difference from the 3-billion model is unresolved, both score above the fusion, and in an exploratory split their lead over the previous model sits on pages from domains that their fresh training pages also cover. In this recipe, more reader data improves a trained model where a stronger tie-breaking teacher yields no material gain, and frontier models used as rankers still trail a small model trained on readers.

Large reasoning models (LRMs) achieve strong performance by externalizing explicit reasoning traces before producing the answer, yet suffer from overthinking challenge that allocates uniformly heavy computation to queries of varying difficulty. While proprietary models mitigate this via opaque routing, open-source LRMs still lack an efficient mechanism to internalize adaptive reasoning due to both expensive training cost and limited disclosure of training recipes. In response, we introduce RPO (Root-token Policy Optimization), a framework that enables LRMs to self-determine when to reason by training only the initial root token (e.g., whether to invoke the <think> tag) via group relative reward and group-wise advantages. By focusing on this pivotal branching point, RPO drastically reduces training overhead and VRAM usage. Across multiple model families and scales, RPO learns difficulty-aware adaptive thinking at just ~2% of the training compute of prior adaptive-reasoning methods.

The objective of this paper is to study the existence and uniqueness of positive ground state solutions of the following Schrödinger problem \[ -\Delta u=u\log u^2+\mu\abs{u}^{p-1}u\qquad\text{in }\mathbb {R}^N, \] where $N\ge2$, $1<p<2^*-1$ and $\mu>0.$ We prove that there exists a positive ground state solution for every $\mu>0$ and every such solution is smooth, radially symmetric with respect to some point, strictly decreasing in the radial variable. Moreover, as $\mu\downarrow0,$ the positive ground state solutions centered at the origin will converge, uniformly over the ground-state set, to $U_0(x)=\exp(\frac{(N-\abs{x}^2)}{2})$ in the logarithmic Orlicz–Sobolev space and in $C^{2,\alpha}(\R^N)$ for every $\alpha\in(0,1),$ respectively. Furthermore, for sufficiently small $\mu,$ we also prove that the positive ground state solutions are unique up to translations and the kernel space of the real scalar linearized operator can be spanned by the translation derivatives. The proof combines minimization on the Nehari manifold with uniform compactness and an exterior coefficient estimate that prevents normalized difference and kernel sequences from vanishing locally.

Generative AI has achieved remarkable success in de novo protein design, yet protein engineering typically begins with existing, characterized proteins. The primary demand is targeted local editing, e.g., engineering a specific region to optimize affinity, alter immunogenicity, or improve stability, while strictly preserving core catalytic motifs and overall scaffold integrity. However, no standardized framework currently exists for carrying out and evaluating such edits. In current generative-design workflows, researchers often need to manually specify complex low-level parameters and rely on repetitive, trial-and-error sampling and evaluation across generative backends to search for viable conformations. While large language models (LLMs) offer an intuitive interface to express these goals using natural language, prevailing structure-generation models cannot directly interpret high-level human instructions. To bridge this gap, we introduce PROTEAN, a standardized benchmark for text-guided protein structural editing. PROTEAN comprises 380 sub-tasks spanning six intervention families and evaluates a two-stage pipeline in which an LLM translates natural-language goals into backend-specific geometric constraints and a generative model remodels the protein backbone. A two-tier evaluation separates geometric compliance, including target attainment and off-target scope fidelity, from biophysical realizability assessed by inverse folding and refolding consistency. Across five frontier LLMs paired with three backends, strict end-to-end biophysical success starts at just 7.0%. Ablation studies show that LLM constraint compilation is already highly accurate, whereas the main bottlenecks are backend-specific scope leakage and refolding failure. PROTEAN provides an open testbed for developing autonomous, multi-backend macromolecular design agents.

When an agent leaves a spatial network, its replacement removes old contacts and may transport population to a new location. The frequency of replacement alone therefore does not determine how turnover changes spatial organization. We study diffusing agents joined by persistent attractive contacts, with the total pull on each agent bounded. At fixed population, successors appear at the departure position, within a surrounding disk, or anywhere in the system. We ask whether small differences between dense and dilute regions grow or fade. The theory separates an exactly calculable diffusion and replacement contribution from an approximate attraction contribution obtained by averaging the contact network. Paired simulations find similar attraction contributions under the three rules at all nine sampled wavelengths. Changing only the replacement location reverses the broadest tested pattern: it grows with replacement at the departure position but fades under both displaced rules. An independent radius scan traces this change from growth to decay. Under displaced replacement, intermediate wavelengths grow between sampled long and short scales with decay. Rates measured in the earliest finite window are compatible with the prediction at long and intermediate wavelengths; discrepancies are already present at short wavelengths. Replacement distance thus helps determine which spatial density contrasts are initially amplified.

Long-term memory supports agents across sessions. Repeated retrieval and write-back can let one historical claim occupy multiple retrieval positions and influence current answers without independent evidence. Existing work examines retrieval, updating, and question answering, leaving same-source record accumulation and post-correction persistence underexplored. Motivated by research in social cognition and memory psychology on repeated exposure, retrieval, and revision, we define Memory Echo as the excess retrieval exposure gained by historical information from the same source without additional independent evidence, and introduce MEchoBench to evaluate its emergence, correction resistance, persistence, and effect on answers. To address this phenomenon, we introduce the Family–Provenance–Authority framework (FPA). Specifically, FPA selects one representative per source family, conditions the eligibility of old records for the answer context on verified corrections, and constrains final answers with authoritative evidence. In paired replay on 190 fixed questions across four memory frameworks, FPA raises accuracy from 0.825 to 0.920 with DeepSeek V4 Flash and from 0.782 to 0.916 with Kimi K2.6; mean Decision Contamination Effect falls from 0.106 to 0.030 and from 0.109 to 0.038, respectively. These results show that agent memory can preserve traceable history without allowing repeated historical information to govern current decisions. The code is available at https://github.com/OrcasSam/memory-echo-benchmark.

Utility-scale photovoltaic (PV) inspection is dominated by aerial surveys, which are effective for identifying module-level hot cells but provide limited visibility of critical components located under the array, including connectors, cabling, fasteners and structural elements. These assets are difficult to inspect consistently at scale, and their defects are often spatially small, partially occluded and difficult to localise from isolated images. We present an automated 3D visual-thermal inspection system (Tricorder) for under-panel PV asset inspection. The approach combines LiDAR-inertial Simultaneous Localization and Mapping (SLAM) with geometrically calibrated RGB and thermal imagery and machine-learning-based object detection and segmentation. Detected assets are associated with the reconstructed 3D environment, allowing visual and thermal observations from multiple viewpoints to be spatially fused and localised. We demonstrate the approach on utility-scale PV data for MC4 connector detection and connection-state recognition, thermal anomaly identification, and detection of structural components including bolts, nuts and torque tubes. The results demonstrate a pathway from image-based inspection toward spatially grounded autonomous inspection, where defects are detected and linked to specific assets and their 3D locations within the solar farm.

Automatic Speech Recognition (ASR) system evaluations have consistently revealed performance disparities along lines of race, gender, socioeconomic status, and language variety. While these disparities are well-documented, they are often (implicitly or explicitly) treated as issues that subsequent, larger, and more universal models will correct. We challenge this assumption of steady progress in reducing accent-related performance disparities, conducting a longitudinal evaluation of 16 prominent systems released between 2021-2025 across 8 datasets representing diverse English accent varieties. We find that the performance disparity between Mainstream US English (MUSE) and minority accents is stubborn, not reliably declining as a function of time. Specifically, while systems are reliably performant on MUSE, minority accents are subject to risks of large performance degradations between generations of systems, sometimes worsening by 65% absolute Word Error Rate. We argue that standard ASR benchmarks enact an implicit language policy privileging Mainstream US English varieties. As such, these performance disparities reflect systemic failures of language technology development infrastructure that disproportionately affect underrepresented speaker populations.

Gaussian process bandits exploit similarities among configurations to optimize prompts, retrieval systems, and agent workflows. In these applications, however, inexpensive model-based evaluations can disagree with trusted task outcomes. We study selective verification when an action is chosen before a diagnostic is observed and trusted feedback is informative on an unknown set of diagnostic states. A finite reliability class and a finite RKHS signal dictionary describe these two sources of uncertainty. We construct a fixed audit design that combines reliability coverage with reward separation and establish a finite-time regret bound that accounts for a hard audit budget and the rounds needed to obtain informative observations. A safe adaptive extension mixes this robust design with predictable Chernoff allocations while preserving a uniform identification floor and exact expected acquisition-cost accounting. For signed-coordinate kernels, structured and diffuse reliability classes require different audit scales despite sharing the same reward geometry. Synthetic experiments distinguish these certified designs from an adaptive posterior heuristic and GP controls. Safe adaptive allocation reduces aggregate regret relative to the fixed design under the exact dictionary, but its finite-budget error certificate is numerically loose. Conditional-feedback errors can remove the heuristic's advantage and target perturbations can reverse it. Separate AI-task replays assess feedback correction and fitted-dictionary policies without establishing end-to-end validation of a learned selective-verification method.

In this paper, we focus on an open problem whether every locally compact dcpo is sober for the Scott topology. Specifically, we obtain two sufficient conditions for answering this problem. The main results are (1) every meet continuous locally compact coherent dcpo is sober with respect to the Scott topology; (2) the Scott space of every locally compact second countable dcpo is a sober space.

Multi-continuum models provide an effective macroscopic description of transient flow in fractured and highly heterogeneous porous media by introducing several interacting state variables at each macroscale location. Although high-order multiscale (HOMS) methods reduce the cost of resolving microscopic heterogeneities while recovering local oscillatory features, repeated solution of the coupled homogenized system and repeated high-order reconstruction remain expensive in long-time and large-scale three-dimensional simulations. This work develops a proper orthogonal decomposition (POD) accelerated HOMS framework for transient multi-continuum problems. A vector-valued POD space is constructed from snapshots of the coupled homogenized solution, and the resulting block system is projected onto a low-dimensional space. Two recovery strategies are considered. The first reconstructs the reduced homogenized fields and performs full HOMS assembly at every output time. The second applies the spatial part of the multiscale recovery operator to retained POD modes offline and evaluates the online solution by direct linear combinations of corrected bases and time-derivative bases. An error estimate between the fully discrete homogenized solution and its POD-reduced approximation is established for the coupled system. A recovered POD-HOMS estimate is then obtained by combining the HOMS consistency error with POD truncation errors propagated through value, gradient, inter-continuum exchange, second-derivative, and time-derivative recovery terms. Numerical experiments for two- and three-dimensional heterogeneous configurations show that the reduced solutions retain the accuracy and local oscillatory behavior of the corresponding HOMS solution while substantially decreasing computational cost.

Large language model agents can follow complex human instructions to solve real-world tasks. In practice, however, initial instructions are often incomplete. In our analysis of 100 user-agent trajectories from SWE-chat, 56% of initial instructions omit requirements that later emerge as necessary, and 52% of the useful follow-up requirements could have been inferred from the project. A user states requirements from their own understanding of the task, and these stated requirements often differ from the true requirements that completing the task imposes. An agent should therefore treat the instruction as evidence rather than as the objective, inferring the true requirements from the instruction and the environment and satisfying them even when the user is not yet aware of all of them. We call this ability outcome alignment. In this work, we construct TiGanEval, a benchmark of software engineering and research tasks. In every task the user genuinely lacks part of the requirements, each requirement stays discoverable from the code and runtime behaviour of the environment, and requirements the agent infers are separated from those the user adds after observing delivered work. These properties together make outcome alignment measurable. Evaluating eight frontier models including GPT-5.6-Sol, we find outcome alignment broadly missing: the models lose 53 to 67 points of success rate when moving from the full specification to the partial instruction, and consequence feedback across four submissions does not close the gap. Furthermore, a plan-mode harness cannot compensate for the missing ability of the model itself, which places outcome alignment among the abilities the next generation of models should be developed for.

In answering the main criticism of the classical robust optimization that the size of the uncertainty set has to be determined beforehand, a recently proposed Joint Estimation and Robustness Optimization framework treats the size of the uncertainty set as a variable and maximizes it subject to some robust conditions over the uncertainty set. This paper studies its effect on portfolio selection when both the robust return and risk constraints are built into Joint Estimation and Robustness Optimization framework. Although this framework does not directly pursuit any optimality of portfolio return or risk, we report a surprising result that Joint Estimation and Robustness Optimization framework actually selects efficient portfolios. This is proved by reformulating the robust risk constraint as a tractable form through the S-procedure. Moreover, a closed-form formula was derived when there is only the budget constraint. An efficient algorithm is proposed when more constraints are added. Numerical experiments illustrate the validity of the reported results. This research paves the way for Joint Estimation and Robustness Optimization framework to be applied to the portfolio selection problem with return and risk related robust constraints.

Variational quantum algorithms utilize parameterized circuits many times during training, at new parameter values, over batches of data, and with gradients. Pauli propagation approximates the expectation values of such circuits classically by evolving the observable in the Heisenberg picture as a truncated sum of Pauli strings. We present PADO-Pauli, a GPU engine for Pauli propagation that compiles a circuit and its observables once into sparse matrices in which each rotation angle contributes only a cosine and a sine factor, so every later evaluation reuses the compiled program. It evaluates batches of parameters and data inputs embedded as rotation angles. By default its gradients come from a vector-Jacobian product registered with PyTorch's automatic differentiation, so they flow through the angles to any classical layer that produces them. PADO-Pauli also provides truncation by Pauli weight and coefficient size, and noise channels. Zero-filtering removes terms that do not contribute to an expectation value in the all-zero reference state and makes the batched forward pass about 1.5 times faster. Compared to cuPauliProp on an NVIDIA A100, PADO-Pauli evaluates up to 10 times faster on 4 to 12 qubits and uses up to 3.1 times less memory on 16 qubits. We check its accuracy against exact statevector simulation up to 28 qubits. These features make PADO-Pauli a practical toolkit for training variational quantum algorithms, including quantum machine learning models.

In this paper, we establish vanishing viscosity limit of the 2D stationary Navier-Stokes equations in an expanding annulus with a fixed width. The flow is subjected to a large perturbation of a rotation of rigid body at one boundary and stays at rest at another boundary. Due to this different boundary condition setting, the boundary layer effects are different near the interior and exterior boundaries, which result in different thickness of boundary layer and different leading order boundary layer equations. We will construct an approximate solution to this 2D Navier-Stokes equations by using higher order asymptotic approximation and show the validity of the boundary layer expansion. The leading order of the Euler solution is the Taylor-Couette flow $\left(Ar+B\frac{1}{r}\right)e_\theta$ for some suitable constant $A$ and $B$, which is determined by using the principle of the Prandtl-Batchelor theory. The perturbation of the tangential velocity becomes large as the annulus expands, which causes heavy difficulties when we show the convergence of the approximate solution. Variously weighted energy estimates up to third-order derivatives of the error solution are constructed to close the energy of the nonlinear system.

We investigate the thermal phase structures and kinematics of atomic hydrogen (H I) gas in the Draco nebula to clarify how infalling H I evolves toward cold atomic and molecular clouds. Using the DHIGLS H I 21 cm data, we apply the Regularized Optimization for Hyper-Spectral Analysis (ROHSA) algorithm to decompose the emission into spatially coherent Gaussian components, and classify them as warm neutral medium (WNM), lukewarm neutral medium (LNM), and cold neutral medium (CNM). We identify two kinematically distinct WNM components: WNM1 near $-3$ km s$^{-1}$ and WNM2 near $-30$ km s$^{-1}$. The LNM forms a broad velocity bridge between two WNMs, whereas the CNM is concentrated within a narrower intermediate velocity range. CO emission is also confined to this interval and is associated with CNM-rich regions. The H I mass fractions are $\sim 41\%$, $\sim 23\%$, $\sim 17\%$, and $\sim 19\%$ for WNM1, WNM2, LNM, and CNM, respectively, with the LNM and CNM together accounting for $\sim 36\%$ of the total H I mass. No comparably independent WNM component is recovered at intermediate velocities, although a weak broad component may remain unresolved. The observed phase structures are consistent with a dynamically compressed multiphase interaction layer in which compression increases the gas density, shortens the radiative cooling time, and drives the evolution from WNM through thermally unstable gas toward the CNM. The present interaction may both create new CNM, and further compress or grow pre-existing dense structures. Our results provide the missing link between previous evidence for dynamical interaction and molecular cloud formation in Draco, and identify CNM formation as a key step in transforming gas accreting toward the Galactic disk into cold atomic and molecular clouds.

We investigate a direct two-ends approach to restriction for the truncated cone in $\R^n$, which gives the diagonal restriction range \[ p>2+\frac{4(7n+15)}{11n^2-n-24},\qquad n\ge10. \] This improves the range obtained by applying the result of Wang–Wu and then slicing.

Large language model (LLM)-based multi-agent systems (MAS) are commonly organized around two coordination regimes. Centralized coordination provides explicit control but can create a bottleneck at a single coordination locus, whereas decentralized coordination offers local autonomy but must achieve effective collaboration through distributed decisions and experience. These trade-offs motivate a systematic study of the space between the two regimes. We introduce Degrees of Freedom (DoF), a continuous configuration framework that quantifies the size of the decision space delegated to downstream agents. DoF controls the granularity of delegated work and the target probability that each task unit is assigned to the branches. Its endpoints recover the two coordination regimes. Across multiple task domains, DoF reveals a hierarchy of organizational effects. Delegated granularity and within-regime routing produce structured but bounded profile reallocations, whereas task domain and retained experience are associated with stronger differences between the centralized- and decentralized-branch profiles. The resulting performance landscape is task-dependent: different domains favor different DoF configurations, while atomic-target conditions are more consistent across runs. In-domain retained experience brings the two branch profiles closer together. The same controlled sweep reveals a task-dependent coordination landscape: some domains exhibit higher-performing configurations, whereas others show flatter responses and no single configuration dominates across domains. In five of the six evaluated state, an intermediate DoF configuration achieves higher mean accuracy than both endpoints, demonstrating the value of exploring the coordination space between them. Together, these findings provide empirical guidance for studying and designing intermediate coordination regimes in LLM-based MAS.

Robotic assistance for human mobility could benefit care recipients and reduce caregiver burden during physically demanding tasks such as Sit-to-Stand (STS) and Lift-and-Transfer (L&T). Learning robot policies for these whole-body assistive tasks is challenging: real-world data collection for human-robot interaction is costly and safety-critical, making simulation an attractive alternative. Prior work, however, typically treats the human as static or minimally responsive and thus fails to capture the cooperative nature of mobility assistance. We present a simulation-based framework for learning whole-body assistive robot policies. At its core is a human policy trained with an adversarial motion prior that produces realistic care-recipient motion in response to the robot's actions. A human-adaptive scripted controller interacts with this policy to collect diverse robot demonstrations at scale. From these demonstrations, we train 3D diffusion policies that map segmented human point clouds and robot joint positions to whole-body actions for a mobile manipulator, without relying on privileged human state at deployment. The resulting robot policies successfully perform both STS and L&T assistance in simulation. Preliminary experiments further demonstrate sim-to-real transfer for STS after fine-tuning on a small amount of real-world data. Videos are available on our project page: https://rchi-lab.github.io/human-gen-mobility-assistance/

Conventional theoretical descriptions of moiré superlattices, epitomized by the Bistritzer–MacDonald continuum model, are fundamentally constrained by low-energy Dirac cone approximations and first-shell interlayer coupling truncations. While successful for small-angle twisted bilayer graphene near charge neutrality, these frameworks fail to provide universal analytical solutions for strained lattices, high-energy band behaviors, or arbitrary commensurate geometries. Here, we establish a unified, approximation-free analytical formalism for generalized moiré systems based on three pillars: (i) a rigorous two-dimensional-to-one-dimensional supercell cyclic mapping, (ii) exact block-circulant diagonalization of intralayer Hamiltonians, and (iii) quasi-degenerate Löwdin perturbation theory centered on arbitrary energy degeneracy points across the full spectrum. Unlike traditional schemes anchored to the Dirac point, our approach enables systematic perturbative expansions around any band crossing, unifying the characterization of low-energy flat bands with high-energy degeneracies. We demonstrate that this framework perfectly recovers the Bistritzer–MacDonald model in the low-energy limit while extending its reach to heterostrained systems and high-energy regimes. By incorporating high-order interlayer coupling shells, we reveal an intrinsic scale self-similarity in moiré lattices and predict the emergence of higher-order magic angles and tunable flat bands driven by secondary degeneracies. This work eliminates the long-standing lack of universal analytical tools for diverse moiré platforms, providing a full-spectrum theoretical foundation for band engineering and the exploration of exotic correlated and topological states in van der Waals heterostructures.

Policies trained offline from teleoperated demonstrations can degrade outside their training distribution. This gap is particularly acute in long-horizon whole-body mobile manipulation, where long horizons amplify compounding errors and coordinated arm, torso, and base motion makes exhaustive demonstration coverage impractical. We introduce I3L, an integrated hardware and learning system that combines whole-body correction with iterative policy refinement to address this last-mile reliability gap. Its low-cost, open-source JoyLo+ interface uses policy-tracking leader arms and touch-triggered takeover to enable coordinated corrections across the body parts, while iterative deployment and retraining allow operators to target failures of the updated policy. We evaluate I3L across three robot embodiments and eight simulated and real-world tasks, including four long-horizon whole-body mobile manipulation tasks. Controlled simulation experiments show that four-round HG-DAgger achieves the highest final success on both tasks. Under this protocol, I3L improves autonomous success over the base policies by an average of 68.3 percentage points and achieves at least 90% success on every task, outperforming demonstration-only training by 35.6 percentage points on average with the same number of training trajectories. Averaged across tasks, takeover duration per episode decreases by 72% over four rounds. Together, these results show that I3L helps close the last-mile gap between offline training and reliable deployment and provides practical insights into collecting and using corrections for long-horizon whole-body mobile manipulation. Videos, visualizations, and appendix are at https://i-3l.github.io/.

Failure-attribution benchmarks ask an LLM judge which step or agent caused a failure. A high score should mean the judge checked the faulty step against the run (a cross-step check). However, spurious signals (surface cues) that predict the annotation can produce a high score. In this paper, we propose four tests to separate checking from cue use. Input restriction asks whether partial input predicts the annotation. Cue following checks whether moved cues change answers. Residual accuracy scores judges where cue-only predictors fail. Aggregation asks whether the failure mix or metric changes rankings. Real observations motivate these tests: on three benchmarks, simple predictors match or beat GPT-4.1. In two Who&When Pro systems, a classifier reading one step's reasoning finds 95-100% of faulty steps, motivating input restriction. In one of them, error injection leaves no reasoning after the faulty step; judges fine-tuned on its labels beat GPT-4.1 by 41 points. Adding task-related reasoning after that step moves this cue: fine-tuned judges lose 31-99 points, but zero-shot judges lose at most 9, motivating cue following. On Aegis, blaming every agent matches GPT-4.1 on micro-F1 but trails by 19 points in exact-set accuracy, motivating aggregation. We recommend reporting these tests when developing failure-attribution benchmarks.

A robot may know how to perform a task yet fail when a human asks for it naturally. We study this human–VLA communication gap by separating an information gap, where task semantics, grounding, or event structure remain unresolved, from a residual execution gap, where even a fully specified task depends on how it is organized and invoked. To do this, we characterize how people naturally express demonstrated objectives and evaluate controlled variants of the same physical tasks, with detailed analysis on $\pi_{0.5}$ and consistent trends on $\pi_0$-FAST and Cosmos3-Edge. Our study reveals that humans increasingly compress instructions as tasks grow, relying more on contextual grounding and implicit execution structure. Current policies are misaligned with this behavior: they tolerate grounding dependence relatively well, but struggle when task semantics or event structure must be reconstructed. Even after the information gap is resolved, execution remains sensitive to command granularity and invocation state. These findings identify concrete targets for human-centered policy learning and interfaces that better bridge natural human objectives to robot behavior. Project website: https://commgap.github.io/.

Can an integrated optical readout improve the useful operating range of a miniature microphone? We address this question with an analytical model of a diaphragm carrying the sensing arm of a Mach-Zehnder interferometer (MZI). The model links pressure-induced bending to waveguide elongation and then to photocurrent, allowing mechanical sensitivity and optical detection limits to be assessed separately. Optical shot noise and back-cavity thermomechanical noise set the pressure resolution, while a 10% total harmonic distortion criterion defines acoustic overload. Two operating regimes are examined: a compact MEMS microphone with 1 mW optical input and a measurement microphone with 100 mW optical input. For the compact case, a 1 mm diameter, 1 micrometer thick, stress-free silicon diaphragm and 100% modulation efficiency give a predicted signal-to-noise ratio of 78 dB and acoustic overload pressure of 124 dB. The higher-power case lowers the modeled noise floor and increases dynamic range relative to the selected measurement microphone benchmark. These estimates rely on favorable assumptions and do not establish a substantial overall advantage in conventional microphone applications. Remote optical interrogation offers a separate motivation for harsh-environment sensing, but the temperature tolerance of the proposed sensing head remains unverified. The analysis identifies the design tradeoffs and measurements needed to test the predicted performance.

We study a problem that unifies two major generalizations of popular matchings in the one-sided preference model. In a bipartite graph in which each vertex, or applicant, on one side has preferences over the adjacent vertices, a matching is called popular if it does not lose a head-to-head election against any other matching. Since the seminal work of Abraham, Irving, Kavitha, and Mehlhorn, who gave the first polynomial-time algorithm for finding a popular matching, numerous generalizations of the popular matching problem have been investigated. Two prominent directions are weighted popular matchings, in which each applicant is assigned a voting weight, and popularity problems under matroid constraints. The former admits a polynomial-time algorithm due to Mestre, while the most general form of the latter is the popular common independent set problem, for which a polynomial-time algorithm is given by Kavitha, Makino, Schlotter, and Yokoi. In this paper, we introduce the weighted popular common independent set problem, which simultaneously generalizes weighted popular matchings and popular common independent sets. Our main result is the first polynomial-time algorithm for this problem. A key ingredient of the algorithm is a solution-independent characterization of weighted popular common independent sets, which we establish in this paper. As consequences of our algorithmic result, we obtain polynomial-time algorithms for several previously unresolved problems, including weighted popular matchings with matroid constraints and weighted popular branchings.