Papers

Multimodal models are increasingly shifting toward unified architectures that understand and generate text, images, and other modalities within a shared conversational context. This design enables fluid interaction across modalities, but it also changes the privacy threat model: Information revealed in one part of a conversation may remain accessible when the model later generates content in another modality. This risk is particularly concerning in settings where users rely on locally deployed models for privacy, assuming that sensitive interactions remain confined to their device. We introduce Privacy-Leaking Watermarks (PLWs): invisible, trigger-dependent watermarks that a malicious model provider can condition on prior chat history. With this adversarial intervention, the usual separation breaks: a sensitive keyword or semantic cue mentioned earlier in the conversation can cause a later, unrelated image to carry a hidden yet detectable watermark. PLWs pose a novel threat to users of unified multimodal models: A poisoned model can retain utility while covertly turning image generation into a channel for privacy leakage, even when deployed locally. Across 13 sensitive-attribute triggers and two model families, PLWs reach up to 100.0% TPR at 1% FPR. For example, across all tested conversational separations, OmniGen2 detects every prior disclosure of depression while falsely flagging only 1% of images generated without such a disclosure.

Accurately estimating judgment reliability is a central challenge in adapting LLM judges to newly verified feedback while preserving previously learned behavior. However, existing approaches often rely on output-level confidence, which can be overconfident and poorly aligned with judgment correctness. We propose VERA, a VErdict-conditioned Reliability Axis that estimates reliability from hidden activations by distinguishing correct from incorrect judgments within each predicted-verdict group. Using VERA as a control signal, we develop a VERA-guided periodic adaptation framework that integrates reliability-ranked corrective updates, reliability-residual replay, and periodic refresh of the reliability directions. After VERA-guided adaptation on Chatbot Arena, 8B- and 14B-parameter judges outperform the strongest baseline on each of four held-out public benchmarks, with relative gains of up to 23.01%. The framework also improves focal-class recall by up to 16.1% relative to the strongest adaptive baselines on a separate proprietary temporal auditing task.

It was recently shown by Paolini and Shelah, with another proof by Laskowski and Ulrich, that torsion-free abelian groups are Borel complete. This means that isomorphism for torsion-free abelian groups is as complicated as possible, so that torsion-free abelian groups do not admit any meaningful complete isomorphism invariants. We give a proof following the ideas of the Paolini-Shelah argument but with several expository advantages. In particular, the combinatorial part of the argument is much simplified. We make use of the ideas of this proof to show that there is a torsion-free abelian group with no computable copy but with a copy computable from every non-computable set. This answers a long-standing question of Downey and Goncharov.

Preference-based alignment methods such as reinforcement learning from human feedback (RLHF) and Nash learning from human feedback (NLHF) aggregate pairwise preferences to learn an LLM policy, but a natural goal is maximizing social welfare (average cardinal utility), which comparisons alone need not identify. Gölz, Haghtalab, and Yang (GHY) measure the gap by distortion: the worst-case ratio between the welfare of the best fixed lottery (distribution over responses) and of the learned lottery. They show NLHF is optimal when every user receives the same lottery. Account-based LLMs, however, have information about their users and can serve different lotteries to different people. We give an efficient algorithm, GLHF, that learns a single group-conditioned policy from one comparison per user. Under individual Bradley–Terry comparisons, GLHF asymptotically matches GHY's optimal population distortion bound simultaneously on every group in a prespecified, possibly overlapping collection, with sample complexity growing logarithmically in the number of groups and inversely with the smallest group mass. A sharper guarantee for groups with similar preferences approaches distortion of one when members share a feasible favorite response. In experiments using human coffee ratings and synthetic LLM-generated ratings, GLHF lowers distortion in every evaluated group and substantially reduces worst-group distortion relative to NLHF and other group-agnostic baselines.

We revisit an old non-diagonal modular invariant of $F_4$ at level $6$, discovered in 1989. At the time of its discovery, it was noticed that it could not be obtained from a conformal embedding of affine Lie algebras. Later, it was shown that it can be obtained from an extension of a vertex operator algebra which is not of Lie type. To this modular invariant, one can attach a full CFT defined (or formulated) as a module category $\mathcal M$ for the fusion category $\mathcal A = \mathcal C(F_4,6)$. More recently, it was recognized that $\mathcal M$ is associated with an étale algebra of $\mathcal A$. In other words, the resulting quantum module is a quantum subgroup (it is also said to have self-fusion, to be of type I, or to be flat). We study the dual $\mathcal O$ of $\mathcal A$ with respect to $\mathcal M$. In particular, we describe the algebra of quantum symmetries (we obtain the chiral generators and display the Ocneanu graphs for the fundamental irreps of $\mathcal A$). The chiral graphs are not connected, each one containing two pairs of isomorphic components. The vertex sets of these components are identified with the simple objects, up to equivalence, of two module categories $\mathcal M$ and $\mathcal M^\prime$ sharing the same modular invariant. $\mathcal M$ enjoys self-fusion, whereas $\mathcal M^\prime$ does not. The restriction of the action of the Verlinde algebra $A$ to the associated abelian groups $M$ and $M^\prime$ defines two families of matrices with non-negative integer entries (nimreps) and their adjacency graphs. As a by-product, we obtain the matrices describing the defect lines of the theory, the integer and rational fusion polynomials attached to $A$ and its modules, and the relevant numerical invariants, quantum dimensions, quantum masses, and induction-restriction rules.

Quantum conference key agreement (QCKA) enables multiple spatially separated users to establish a common key with composable information-theoretic security by exploiting multipartite quantum correlations. Greenberger–Horne–Zeilinger (GHZ) states are a natural resource for this task, but their distribution over optical fiber is degraded by transmission loss, polarization decoherence, and detector imperfections. We investigate the finite-key performance of an $N$-party GHZ-based QCKA system using the $N$-BB84 protocol, where each optical link is modeled by extending the bipartite erasure–Pauli channel to incorporate detector dark counts. We characterize the impact of this quantum channel on the shared GHZ state and derive closed-form expressions for the error rates. The analytical results reveal that the local bit-error rate depends only on transmissivity and dark counts, while the global phase-error rate compounds over all $N-1$ links and therefore degrades faster with both distance and the number of parties. These error rates are integrated into a finite-key secret-key-rate (SKR) framework, in which we optimize the parameter-estimation probability via Brent's derivative-free method. Numerical results show that increasing fiber distance and network size reduce the achievable SKR. Furthermore, the optimized parameter-estimation probability consistently increases the SKR and extends the maximum secure distance relative to a fixed-allocation baseline, with comparison to the asymptotic limit quantifying the residual finite-key penalty.

We report the results of the 1.2 mm continuum and molecular line $\sim$0.1 parsec-scale observations with the Atacama Large Millimeter/submillimeter Array of two fields in the star-forming region N83 in the Large Magellanic Cloud (LMC). This study is third in a series of investigations exploring molecular complexity in a low-metallicity and high UV flux environment of the LMC. We determine the physical and chemical properties of five sources detected with ALMA based on spectral modeling with XCLASS and the analysis of the continuum emission. The temperatures exceeding 100 K, small sizes, and high H$_2$ volume densities of sources N83A-mm A and B indicate that they are hot cores. N83A-mm A/B are associated with complex organic molecules: methanol (CH$_3$OH), methyl cyanide (CH$_3$CN), and dimethyl ether (CH$_3$OCH$_3$). We also found CH$_3$OH in three cold cores, and extended cold CH$_3$OH emission in both fields. We detected SO, $^{33}$SO, SO$_2$, OCS, H$_2$CS, CS, C$^{33}$S, HC$^{15}$N, H$^{13}$CN, HNCO, HDCO, H$^{13}$CO$^{+}$, SiO, and CH$_2$CO in at least one core. Molecular abundances with respect to H$_2$ fit into the trends previously reported for the Magellanic hot cores: the CH$_3$OH abundance varies significantly, while the abundances of S-bearing species show small variations. We found evidence for strong UV radiation and complex kinematics in both ALMA fields in N83. The feedback-produced shocks are important drivers of the chemistry of S-bearing species, SiO, and CH$_3$OH in and outside hot cores. Local effects have a large impact on the chemical composition and abundances of star-forming cores in the LMC.

Off-policy hierarchical reinforcement learning must estimate the values of high-level decisions while the low-level policy changes. HIRO adapts replay data through subgoal relabeling, but after a label change, the value update targets the relabeled subgoal instead of the subgoal the high-level policy originally needed to update. We propose Hierarchical Time-aware Bootstrapping (HTB), which evaluates specified subgoals under the current low-level policy while retaining accumulated task rewards. Remaining execution time distinguishes subgoal continuation from a new high-level decision. Together with primitive-action conditioning, it enables off-policy Bellman updates based on the stationary environment transition law. HTB combines these one-step updates with multi-step suffix returns and truncated relabeling, reducing dependence on intermediate value estimates. A shared value component supports learning across actions, while nonnegative residuals constrain upward corrections relative to that component. At a fixed mixture weight of 0.95, HTB achieves 32.8% AntFall success versus 9.6% for matched local HIRO over five paired seeds at 10M environment steps. Ablations identify contributions from recursive continuation and mixed supervision; fixed-policy tests show more accurate predictions for actions whose returns were excluded from fitting.

We study the latency tradeoffs of signature-free asynchronous reliable broadcast with optimistic two-step delivery. We consider $n$ parties, of which at most $f$ are Byzantine. An optimistic fault budget $b\le f$ guarantees two-step delivery with an honest broadcaster and at most $b$ faults, while correctness tolerates $f$ faults. We write $(d_o,d_s,d_t)$ for optimistic latency, honest-broadcaster latency with up to $f$ faults, and totality delay-the elapsed time from first to last honest delivery. For $3f+1\le n<4f$, our first construction improves Opt-RBC's $(2,4,4)$ guarantees to $(2,3,4)$, preserving its maximal optimistic budget $b_*=\lfloor n/2\rfloor-f$. Thus maximal optimism is compatible with the optimal three-step good case. We prove a complementary lower bound: for $3f+1\le n<4f$, every deterministic reliable broadcast protocol in our asynchronous model with two-step optimistic budget $b\ge n-3f$ has totality delay at least three, even when the broadcaster is honest. Our second construction achieves $(2,3,3)$, matching this lower bound at maximal optimism when $f\ge2$ and $\lceil(10f-3)/3\rceil\le n<4f$. Both constructions use quadratic messages and tolerate adaptive corruption. In the matching range, maximal optimism therefore costs exactly one totality step relative to Bracha's $(3,3,2)$, without increasing good-case latency.

Religiously branded platforms raise important questions about trust, usability, and autonomy when technology is built around a specific faith. Prior HCI work on Islam and Muslim technology has focused on single-purpose tools such as prayer, scripture, and health apps, leaving infrastructures like browsers, which shape a user's relationship with the Internet, unexamined. We address this gap through semi-structured interviews with 16 users of Kahf Browser, a faith-aligned browser designed to support Muslims' online activities. Findings show religious identity motivates adoption but does not sustain it. Trust is not fixed by religious branding but shifts over time based on the browser's functionality, and users take pride in Muslim-built infrastructure. Building on these insights, we introduce porous digital segregation: a model where users seek not a sealed alternative internet but a protected default with selective exit mechanisms. We conclude by connecting findings to broader issues in faith-aligned technology and offering design recommendations.

The extension of nanoscale quantum materials to the third dimension brings new possibilities to geometrically tune emergent properties, emulating or indeed going beyond the physics of bulk materials, with exciting implications for high-density technological integration. Realising such three-dimensional architectures requires nanofabrication techniques that not only enable complex 3D geometries, but also deliver functional quantum materials with emergent phenomena. Recent advances in focused electron- and ion-beam induced deposition (FEBID and FIBID) have made 3D nanopatterning of functional materials in complex geometries increasingly feasible, paving the way for the experimental realisation of new phenomena and device prototyping. In this review, we discuss the use of FEBID and FIBID 3D nanopatterning of superconducting and magnetic materials. We address the underlying mechanisms, and capabilities of these techniques, focusing on the physical phenomena that become accessible. Considering the resolution of the techniques in the context of the fundamental lengthscales of the materials, we discuss the emergent effects that arise due to three dimensional geometries in magnetic and superconducting nanostructures, and highlight perspectives for future developments of the techniques.

The recursive-line and signed Zarankiewicz numbers maximize the number of squares in augmentations of a maximum $C_4$-free base, subject to two sufficient irreducibility criteria. The count includes one square per base cell and one per selected pair of unused cells. We compare these parameters with the second-order number, which uses irreducibility itself. Every maximum $m\times n$ base admits a recursive-line augmentation with at least $mn/2-C\max(m,n)$ squares, for an absolute constant $C$. Combining this bound with a two-column extension of known fixed-width families, we show that all three parameters are asymptotically equivalent, uniformly as the larger dimension tends to infinity. For individual displays, a transfer graph shows that once the signed closure identifies every selected pair, the signed criterion is equivalent to irreducibility. We determine the second-order number for every six-column rectangle and give eventual exact formulas for all three numbers at widths seven, nine and eleven. We also prove signed and recursive-line equality at $8\times7$ and, together with earlier values, whenever the shorter side is at most six, except possibly at $14\times4$. The exact-value results are computer-assisted, using exhaustive enumeration, checked propositional refutations and symbolic certificates with a proved lifting argument for arbitrary lengths.

Multiparameter persistent homology is a rapidly developing branch of topological data analysis that improves the robustness of single-parameter persistent homology to outliers, while still capturing the metric characteristics of the data. However, a notable limitation is its lack of scalability. In this paper, we introduce a novel approach for efficiently computing 2-parameter persistent homology on large point sets. Our work extends the Flood filtration, originally developed for single-parameter persistence. Our construction, called the sublevel Flood bifiltration, offers a scalable approximation of the sublevel offset bifiltration. We show that it benefits from theoretical stability properties and describe how to compute it efficiently. We demonstrate the performance of our approach in classification tasks on low-dimensional synthetic datasets, where density awareness is critical, as well as on real-world time series datasets.

We present Guanaco, an algorithm for performing conjunctive query evaluation where, for each Boolean conjunctive query, and positive epsilon, the algorithm achieves polynomial time with exponent equal to the submodular width plus epsilon. The algorithm and its running time generalize smoothly to general conjunctive queries. We believe the algorithm and its analysis to be notably simple, indeed, together we believe they form a highly simple argument that conjunctive query evaluation can be performed in essentially submodular width time. In the case of Boolean conjunctive queries, the algorithm is based on interleaving three simple primitives: a subroutine for establishing a form of consistency; a subroutine for establishing global uniformity, which, briefly speaking, partitions relations as needed to control discrepancies between average degree and maximum degree; and, a simple step that joins pairs of existing relations to form new relations.

We establish a maximum-variance inequality for discrete $\alpha$-concave random variables in the regime $-1/3<\alpha<0$, extending corresponding results beyond the discrete log-concave setting. As applications, we obtain a reverse entropy power inequality and bounds for the Lévy concentration function. We further establish polynomial-type concentration inequalities, reflecting the heavy-tailed nature of the class. A main tool in our approach is a convex majorization principle that reduces the relevant extremal problems to suitable $\alpha$-affine distributions.

For each fixed $n$, we prove a positive lower bound for the diameter of a compact simple geodesic $n$-gonal monotile of the hyperbolic plane. We also prove a positive lower bound for the area of such a tile in a finite monohedral tiling of a closed hyperbolic surface, independent of the topology and metric. Both bounds become independent of $n$ for vertex-proper tilings, in which every genuine tile vertex belongs to at least three distinct tiles. This includes tilings by convex polygons. Reflex angles and non-edge-to-edge incidences are allowed. After qualitative proofs, we obtain explicit constants from a sharp gap estimate for packing polytopes with arbitrarily coupled nonnegative integer constraints. Exact corner balance applies on closed surfaces; in the plane, covering duality and ball counts give a boundary factor depending on the tile diameter. Following Zare, we give, for each integer $q\ge2$, a geodesic construction with $2q+3$ sides and diameter less than $3/q$, showing that side counts cannot be unrestricted without an additional condition. The closed-surface results extend to regular curved sides. The geometric constants are effective but are not claimed to be optimal.

Computer-use agents need to capture procedural knowledge of how people use software. User telemetry offers a scalable source of this knowledge. However, learning reusable skills from these logs requires addressing three challenges: (1) Goal Underspecification, since logs do not record the goal behind each action; (2) Non-Replayability, since past activity cannot be replayed to evaluate skill updates; and (3) Interleaved Trajectories, since logs may mix several tasks without marking their boundaries. To address these, we introduce TeleTune, a framework for learning a textual skill library from offline logs without recorded goals, cannot be replayed during optimization, and may interleave tasks. TeleTune uses action-prediction errors on logged trajectories to propose library edits and keep only those that improve held-out action-prediction accuracy, which we call skill-guided progress. The learned workflows also enable retrieval of demonstrations that cover the subgoals of a new task. At test time, the agent is provided with the learned library and the workflow-based retrieved demonstrations. Experiments on WorkArena and Online-Mind2Web show that TeleTune outperforms random retrieval, Agent Workflow Memory (AWM), and their combination. We find that the best baseline varies by setting, whereas TeleTune achieves average success rates of 77.1% and 80.6%, respectively, improving over the strongest baseline on each benchmark by 6.7% and 7.7%. Under the heaviest perturbation of the WorkArena training data,TeleTune keeps the highest average success rate at 68.5%, 6.3% above the strongest baseline. Our analyses show (1) skill optimization and workflow-based retrieval are complementary, (2) optimizing on fixed logs costs 5 to 75 times fewer tokens than validating the same edits with live episodes, (3) skill-guided progress tracks the live success rate.

A scale-separated monopole method is developed for full-field prediction of precipitation under conditions where particle-scale diffusion fields coexist with solute variations over much larger distances. Precipitates are represented by monopole Green's-function solutions, providing a particle-based description of interacting particle growth and dissolution. The direct quasistatic formulation is validated against transient finite-difference calculations and a conventional Kampmann–Wagner numerical model, but becomes inappropriate when spatially heterogeneous particle populations generate slowly relaxing long-wavelength concentration fields. An Ewald decomposition is therefore used to separate the diffusion field by spatial scale. Screened particle-scale fields are treated quasistatically, while the complementary long-range field evolves by transient diffusion. The resulting hybrid formulation retains explicit particle-level precipitation kinetics while allowing mesoscale solute redistribution to evolve on its physical timescale. Application to Al–Sc demonstrates that a transient depletion field generated by localized precipitation can suppress subsequent nucleation and produce persistent spatial variations in precipitate number density and volume fraction.

Finite-temperature quantum-gravitational corrections to the cosmological constant introduce two additional density parameters, $\Omega_{\Lambda_2}$ and $\Omega_{\Lambda_3}$, the associated terms of which scale as $a^{-4}$ and $a^{-2}$, respectively. In previous work (Papers I and II), these parameters were studied using machine-learning methods that identified a best-fit point. Here we complete the picture with two complementary analyses. First, we perform a Bayesian MCMC analysis using a modified version of class against the Planck 2018 likelihoods. This tells us what the data can constrain once all parameter uncertainties have been marginalized over. Both parameters are consistent with zero, with $\Omega_{\Lambda_3}$ constrained only from one side. Second, we build a CMB pipeline from scratch, following Weinberg's kinetic-theory formulation and validating it against class. This allows us to study the underlying physics. The response analysis carried out with it shows that $\Omega_{\Lambda_3}$ is almost entirely absorbed by a shift in $h$ over the well-measured acoustic multipoles. The part of the signal that cannot be absorbed resides at low $\ell$, where cosmic variance dominates. Meanwhile, the $\Omega_{\Lambda_2}$ signature is nearly degenerate with the overall amplitude of the spectrum. The effects of both parameters on the CMB power spectrum are small, since they represent higher-order thermal corrections to the cosmological constant. The resulting constraints are correspondingly weak: the induced shifts are largely absorbed by degeneracies with $h$ and with the overall normalization of the spectrum. This is the reason why neither parameter is detected.

We investigate singularities of radial solutions of the classical two-phase and one-phase Stefan problem. In a remarkable result, M.A. Herrero and J.J.L. Velazquez show existence of a solution such that the free boundary described by $g(t)$ satisfies in dimension $n\geq 3$ $$g(t)=c_n \sqrt{t} |\log t|^{\frac{-1}{n - 2}} (1 + o(1)), t\to 0,$$ and they conjecture that these asymptotics hold in general. Here we prove the conjecture for all radial solutions, which means complete asymptotic rigidity in that class. In order to obtain these precise asymptotics for the two-phase Stefan problem, for which frequency formulas and monotonicity formulas are unknown, we apply the Potential Reduction to ODE method introduced in arXiv:2608.18913 for the Neumann Bernoulli problem. The philosophy of that method is that the asymptotic shape of the free surface is governed by a precise ordinary differential equation. The fact that we can apply it in this paper to an equation of different type with quasilinear principal part, which is in our regime not the perturbation of a linear operator, confirms the versatility of the method.

We discuss surface-pinned, topologically charged, nonsingular defects (SD) in light scattering by a plasmonic object. They follow from exact Maxwell boundary conditions and are purely topological—unaffected by surface forces. At optical frequencies, in (quasi) 2D, they appear in the electric and Poynting vector fields for $H^{(z)}$ polarization, not for $E^{(z)}$. Exact solutions for a circular cylinder reveal topological charge-conserving bifurcations involving SDs with bulk singularities or SDs alone, and contraction of defect regions under loss.

Recent advances in vision-language modeling have substantially improved multimodal encoding, retrieval and reasoning. Yet for multimodal recommendation, encoding rich item vision-language semantic interactions remains a long-standing bottleneck, which hampers accurate item representation learning and user-item matching. Mainstream approaches primarily adopt independent encoding of vision and language modality followed by rigid late fusion such as concatenation, inherently omitting native vision-language interactions and introducing cross-modal semantic distortion. To address this challenge, we propose OpticalRec, the first visual-space unified encoding paradigm for multimodal collaborative filtering, a fundamental recommendation setting. Instead of isolated modality-specific encoding, OpticalRec renders item textual metadata as visual glyphs, enabling native image-text interaction within the visual encoder - the perceptual encoding level. The resulting representations are further processed by the language decoder - the semantic encoding level, allowing OpticalRec to exploit the dual-attention mechanism of modern vision-language models that previous encoding methods omitted. OpticalRec's efficacy is theoretically supported by mutual information analysis and empirically demonstrated through superior performance across strong baselines and benchmarks. As a plug-and-play module, OpticalRec (1) introduces minimal cost, (2) is robust against rendered text font, color and layout, etc., and (3) integrates seamlessly into existing multimodal collaborative filtering models.

Molecular optimization must improve target activity and satisfy developability constraints within limited evaluation budgets. Classical methods require tailored rules or training to incorporate chemical instructions and property feedback. Frozen language models can condition edits on this information, but need explicit constraint control and relevant experience. We therefore present PharmAgent, a constraint-aware molecular search method driven by adaptive external state. Its Lagrangian controller translates violations in accepted states into accumulated constraint pressure, keeping this history separate from current property measurements. Structure-indexed replay complements this feedback with relevant evaluated transitions that guide subsequent proposals. As a curriculum progressively activates constraints, candidates and the incumbent are compared under the same current objective, and the accepted state determines the next multiplier update. We derive an exact identity that characterizes how accepted-state violations accumulate in the controller's multipliers. Across five tasks with five independent runs, PharmAgent achieves a summed area under the target-score curves (AUC) of 3.9208 in target-only search, improving over MOLLEO by 37.3%. With online constraints, it achieves a property-adjusted AUC of 0.7076, improving over the strongest online baseline, ExLLM, by 53.8%. These results rank first among all evaluated methods in both target-only and constraint-aware search. The online comparison covers all five baseline frameworks. The full system leads every ablation variant in target quality, property-adjusted performance, and Pareto hypervolume. All five molecular cases reach feasible final states, documenting target gains and trade-offs.

We give a classical polynomial-time algorithm that strongly refutes random quantum $3$-SAT at sufficiently large constant constraint density, thereby disproving the quantum analogue of Feige's random $3$-SAT hypothesis. This stands in sharp contrast to classical random $3$-SAT, for which polynomial-time strong refutation is known only at constraint density $\Delta\gtrsim n^{1/2}$. Although quantum $3$-SAT shares the pairwise-independence barrier of its classical counterpart, our SDP-based refutation overcomes this barrier by exploiting the noncommutativity of quantum constraints.

A Constant Distance Gap (CDG) spacing controller for Cooperative Adaptive Cruise Control (CACC) designed to increase the capacity of oversaturated signalized intersections is presented. In a departing platoon of production passenger vehicles, the controller maintains close inter-vehicle distances below 3 m within limited platoon lengths, using one-vehicle-look-ahead (OVLA) communication only. Extensive real-world experiments on heterogeneous, retrofitted legacy production cars demonstrate robust operation. To the best of our knowledge, this is the first experimental demonstration of such close distances under this combination of conditions. A CDG policy can substantially increase discharge rates at oversaturated signalized intersections when vehicles preserve their standstill spacing during startup, as shown in our previous simulation study. Because this benefit unfolds only at high penetration rates, it calls for retrofit-capable solutions deployable in existing combustion-dominated fleets. However, practical realization in such fleets remains an open challenge. The proposed MPC-based controller combines a first-order longitudinal prediction model with precise powertrain feedforward employing OEM-specific torque models, without exchanging confidential vehicle parameters. It avoids explicit platoon coordination to minimize air-interface load in dense traffic. To mitigate disturbance amplification and increase feasible platoon length at very small gaps, two controlled spacing-relaxation strategies are compared: (i) a micro-time-gap policy near the theoretical string-stability bound, and (ii) an acceleration-dependent spacing relaxation based on a virtual mass-spring-damper model. We introduce the Desired Acceleration Overshoot (DAO), a time-domain metric relating local actuator-demand amplification to achievable platoon length.

We study learning in continuous games when players receive only pairwise preference feedback, revealing which of two actions is preferred but neither payoff values nor preference magnitudes. We first show that standard external regret and coarse correlated equilibria (CCE) are not identifiable from this ordinal information: the same sequence of play can incur zero and linear regret in two ordinally equivalent games, while the distributions that remain CCE across all cardinal representations consistent with the same preferences are exactly those supported on pure Nash equilibria. Motivated by this gap, we develop a first-order ordinal theory based on normalized unilateral preference directions, introducing an ordinal directional regret benchmark and corresponding equilibrium notions. We show that block-normalized pseudogradient dynamics achieve sublinear ordinal regret and, under additional structure, Nash-convergence guarantees. We then use a single-comparison estimator to implement these dynamics from finite pairwise comparisons. With one comparison per player and round, the resulting algorithm achieves sublinear finite-resolution ordinal regret against arbitrary opponent behavior and, in ordinal potential games, almost-sure last-iterate convergence to the Nash set. Our results provide regret, dynamics, and equilibrium guarantees directly from preference feedback without reconstructing cardinal utilities.

Let $X$ be a smooth proper integral variety of dimension at most four over a field of characteristic zero. We construct a finite sequence of blowups along smooth centers with projective endpoint, preserving the intersection of all maximal quasi-projective opens of $X$. The construction depends only on a fixed functorial principalization procedure and is natural under isomorphisms, including those over different ground fields. The proof has two parts. A colength argument on regular local surfaces gives, in every dimension, a finite preparation after which some maximal quasi-projective open has complement mapping into codimension at least three in $X$. In dimensions at most four these images are points or curves. Ample numerical classes on the normalized components of that complement give finite bounds for summing base ideals of linear systems. The resulting ideal descends to the original field; geometric integrality is not required.

Purpose: To develop and evaluate an integrated acquisition and reconstruction framework for distortion-free, high-resolution prostate diffusion-weighted imaging (DWI). Methods: PROPELLER-DWI combines pulsed-gradient spin-echo diffusion preparation, an interleaved gradient-and-spin-echo PROPELLER acquisition with golden-angle blade rotation, and self-supervised joint reconstruction and denoising. It was compared with conventional NUFFT reconstruction and MPPCA denoising in a NIST/QIBA diffusion phantom, 25 volunteers, and 4 clinical patients undergoing 3T prostate MRI across three spatial-resolution settings and two b-values (50 and 1000 s/mm2). Reader studies compared geometric fidelity, image quality, and overall preference with clinical single-shot EPI-DWI. Results: PROPELLER-DWI preserved prostate geometry in cases where EPI-DWI showed susceptibility-induced distortion, signal loss, and metal-related artifacts. Self-supervised reconstruction improved noise suppression, with the greatest benefit at higher spatial resolution. Across all resolutions, PROPELLER-DWI had significantly higher geometric fidelity than EPI-DWI, and radiologist preference increased with the severity of EPI distortion. ADC measurements remained consistent across resolutions and comparable to clinical EPI-DWI, whereas EPI-DWI ADC varied with increasing resolution. Retrospective reduced-shot reconstruction also supported shorter scan times. Conclusion: PROPELLER-DWI enables distortion-free, high-resolution prostate DWI while preserving quantitative diffusion measurements, providing a practical alternative for technically challenging prostate MRI and applications requiring improved geometric fidelity.

Automated checks of radiology reports may rely on AI-generated checklists that leave findings unmentioned. We used reinforcement learning to train a vision-language model to fill in a 12-finding checklist from a chest radiograph without seeing the sentence under test; a separate checking model judged the sentence from the checklist. On held-out patients, a rule-based check and an independent medical checker, neither used in training, measured discrimination gains (Youden index) of 12.6% and 11.8%; only the rule-based check met the prespecified false-alarm criterion. Switching to the training format, which fixes finding order and enters unmentioned findings as absent, raised the training checker's measured gain and lowered the independent checker's, a prespecified comparison that yielded 6.2% (95% interval 2.0% to 10.5%) and, post hoc on held-out patients, 7.7%. Across 8 checking models, acceptance of a label-consistent negative statement about an unmentioned finding ranged from 1.0% to 97.0%. Labels were report-derived, not radiologist-adjudicated.

We present a qualitative case study of a human-supervised team of specialized AI agents supporting an astronomy research project. The Bot Spectroscopist team, described in the project record as operating through Grok Bot powered by xAI Grok, divided work among orchestration, numerical implementation, data curation, spectroscopy critique, literature search, figure preparation, manuscript editing, and review. Human decisions determined scientific scope, model changes, and manuscript inclusion. The scientific application is CondGen, a deterministic label-to-spectrum network extending our previous work to five individual elemental abundances. The reported baseline assessment achieved a mean absolute error (MAE) of 0.003681 in continuum-normalized flux on 21,657 synthetic test spectra. Existing binned diagnostics show that reconstruction error varies with effective temperature and abundance. The bots also flagged a non-monotonic cool-star magnesium response for further investigation; a matched SYNSPEC abundance sweep is needed to assess its physical fidelity. Building on a pre-existing spectral database, notebooks, and research program, the authors estimate that the reported analysis, diagnostics, figure preparation, and initial manuscript drafting were assembled in less than a week through human prompting and exchanges among agents. No comparison with a human team starting from the same resources was performed, so this elapsed-time account does not establish a productivity or labor saving. The study illustrates task delegation, artifact exchange, critique, and correction under human supervision.