Papers

Discrete diffusion models offer the ability to re-draft, revisiting and correcting earlier tokens throughout generation. This capability depends on the forward corruption process that defines what the denoiser learns to correct. Masked diffusion models fix tokens once they are unmasked, while uniform diffusion permits revisions but relies on uniformly random token substitutions. We instead learn which substitutions are most useful for training the denoiser to re-draft. We introduce Variational Stackelberg Discrete Diffusion (VSDD), a framework for learning a semantically aware corruption process. VSDD formulates training as a leader-follower game: the leader defines a Markovian corruption process parameterized by the denoiser's token embeddings, while the follower optimizes a variational denoising objective with the corruption process held fixed. The leader rewards corruptions based on how much the denoiser improves after learning from them, rather than on how easily the current denoiser can reconstruct them. We measure this improvement under a fixed reference corruption process, approximate the follower's response with a one-step gradient update, and optimize the leader using a score-function estimator. We evaluate VSDD across molecular, text, and playlist generation. VSDD substantially improves molecular validity over uniform and masked diffusion, reduces text perplexity relative to uniform diffusion while remaining competitive with masked diffusion, and achieves sizable improvements in offline playlist recommendation metrics.

We present the first calculation of the leading perturbative quantum chromodynamics (QCD) correction to the nonleptonic weak flavor-equilibration rate in dense quark matter. The computation is based on the Kadanoff-Baym kinetic equation and performed using real-time thermal field theory. The resulting QCD correction is both gauge invariant and fully finite. In particular, all soft and collinear divergences cancel after combining real and virtual corrections, in accordance with the Kinoshita-Lee-Nauenberg theorem. In the low-temperature regime relevant to compact-star dynamics, the correction enhances the tree-level rate by $30 - 40\%$ over the high-density range where perturbative QCD is applicable. The corrected rate improves on the tree-level result used in phenomenological calculations for more than three decades, providing a more accurate microscopic input for bulk-viscosity calculations in dense quark matter.

The relaxation-time Boltzmann equation governs relativistic matter far from equilibrium, but existing approaches rely on finite-order truncations that break down in this regime. We introduce ShARK (Stochastic Advection Relaxation Kinetics), the first fully 3D Monte Carlo kinetic framework to capture exact, far-from-equilibrium ultrarelativistic dynamics for arbitrary initial conditions while directly enforcing target transport coefficients, converging to the exact Anderson-Witting solution. Validated against the Bjorken and Gubser flows, the framework reproduces the exact kinetic result where truncated methods fail.

Quantum circuit Born machines (QCBMs) built with instantaneous quantum polynomial (IQP) circuits have loss functions that can be efficiently estimated classically, but require quantum computers during inference. Thus, classical computers can train large IQP QCBMs that exceed the qubit numbers of available quantum hardware. This motivates research into how qubit-limited quantum computers can be used to sample from larger IQP QCBMs. Leveraging insights from classical simulation of noisy IQP circuits, we propose a deployment protocol for IQP QCBMs that injects noise at the logical level to probabilistically decompose the original circuit into smaller clusters that can be sampled on qubit-limited quantum processors. We also apply lightweight resampling of known noisy qubits to mitigate the impact of the introduced error. Numerical simulations of our protocol applied to shallow IQP QCBMs and IBM Heron experiments deploying a 316-qubit IQP circuit onto 156-qubit hardware show that mitigation improves the similarity of the empirical distribution to the target distribution. The proposed deployment protocol also crucially preserves correlation structures more accurately than a classical baseline that has accurate single-qubit marginals.

Mass gathering events (MGEs) can create intense, short-lived transmission hotspots whose effects propagate through urban populations. Quantifying these effects requires linking fine-scale interactions during gatherings with epidemic transmission across metropolitan mobility networks. We developed a hybrid multiscale simulation framework coupling empirically derived individual-based contact networks for MGEs with a stochastic metapopulation susceptible-exposed-infected-recovered (SEIR) model. Using population and commuting data from Madrid, Spain, we investigated how event-related transmission affects epidemic dynamics and district-level invasion. Across the scenarios examined, MGEs increased epidemic peak incidence and accelerated epidemic progression, although the magnitude of these effects depended on event-specific contact patterns and background transmissibility. Dense contact patterns produced greater epidemic amplification, whereas measures reducing close-contact interactions weakened these effects. MGEs also advanced epidemic invasion across districts and increased the temporal synchronization of arrival times, producing a more synchronized metropolitan epidemic wave. The association between MGEs and earlier district-level invasion was partly explained by the size of the event-generated epidemic seed. The main patterns persisted across the epidemiological parameter ranges examined. The framework provides a multiscale approach for investigating how mass gathering management may influence metropolitan epidemic dynamics.

We present an in-depth investigation of how a form of Bayesian reasoning about common causes can emerge as a cross-contextual generalization in small, tractable transformers. Incrementing on recent work, our set-up (i) disentangles causal mechanisms in the model from the causal structure of the true data-generating process, (ii) orients more towards natural language prediction by considering inference of latent common causes, and (iii) considers whether and how Bayesian evidence accumulation for latent common causes can be implemented in representations and mechanisms that allow for cross-context generalization to novel test cases.

A classical solution concept in fully observable nondeterministic (FOND) planning, is the strong policy (aka winning strategy in the closely related area of reactive synthesis), i.e., such a policy ensures that the goal is reached in an adversarial environment. When strong policies are not available or there is no evidence that the environment is adversarial, one can resort to best-effort policies, which always exist, and which follow the classic decision-theoretic principle that an agent should not use a dominated strategy. A typical positional best-effort policy works as follows: from every state, it follows a strong policy if one exists from that state (such states are called “strong-winning”), else a weak policy if one exists from that state (“weak-winning”), and else is unconstrained (“losing”). In this work, we introduce a sound and complete planner for both best-effort planning and strong planning. The algorithm that underpins the planner is quite simple: it represents certain sets of states, such as the winning regions, by their $\subseteq$-minimal elements. The algorithm returns uniform policies, i.e., it returns a policy $\pi_t$ that is a strong solution starting in every strong-winning state, and it returns a policy $\pi_w$ that is a weak solution starting in every weak-winning state, and it provides a certificate for the set of losing states. We implemented the algorithm with some simple optimizations (calling it FONDANT), and evaluated it on a benchmark set consisting of the instances that were used in the evaluation of leading strong planners PR2 and FOND-SAT, and the best-effort planner BeSyftP. On coverage, our implementation is at least as good on all domains, and outperforms on some domains; and on wall time, it is slower on small and medium-sized instances, and outperforms on larger instances.

Identifying regimes in which interacting fermionic systems admit efficient classical simulation is a crucial task in quantum many-body physics. In this work, we rigorously establish the Gaussianity and classical simulability of sufficiently weakly interacting fermions. For the Gibbs state of a local interacting fermionic Hamiltonian $H=H_0+V$ at any finite temperature and in any spatial dimension, where $H_0$ is quadratic and $V$ is the interaction term, we show that there exists a positive interaction strength below which the Gibbs state is a convex mixture of fermionic Gaussian states. Moreover, based on this structure, we construct a polynomial-time classical sampler for the Gibbs state via Metropolis updates on a tree whose leaves are Gaussian states.

Multi-turn rollout dominates the cost of agentic reinforcement learning (RL). Asynchronous execution and elastic GPU resources can accelerate this stage, but adding rollout replicas yields diminishing returns while training GPUs remain idle between updates. We observe that effective resource use also depends on the prefill–decode (PD) configuration. Both the choice between colocation and disaggregation and the optimal PD ratio vary with the workload, making resource scaling and PD configuration interdependent. Exploiting this opportunity requires selecting effective configurations and realizing their benefits within transient resource-availability windows despite reconfiguration costs. We present PEARL, an asynchronous agentic RL system that coordinates external resource elasticity, temporary reuse of idle training GPUs, and adaptive PD execution. PEARL maintains a unified GPU–worker–role state and uses runtime profiles to predict rollout batch completion time, accounting for environment-induced reductions in decode concurrency. It selects the PD mode and ratio under the current GPU budget and translates each decision into an incremental transition plan that minimizes worker and role changes. Cost-aware switching and borrowing policies suppress transitions with insufficient expected benefit while ensuring timely return of training GPUs. Our evaluation show that PEARL achieves $2.17$–$2.79\times$ the throughput of fixed-resource ROLL across different LLMs. Compared with RLBoost+, throughput improves by up to approximately 26.9% for Qwen3-8B and 36.3% for Qwen3-30B-A3B.

We study a one-parameter family of double octic Calabi-Yau threefolds with two degenerations of type $K$. For each of these degenerations we construct a semistable model and compute the limiting mixed Hodge structure. We show that the $\ell$-adic Galois representation on the limiting cohomology is isomorphic (up to semisimplification) to the direct sum of the Galois representation attached to a weight-three modular form associated with a singular K3 surface and its Tate twist. This gives an explicit realization of the expected relation between type $K$ degenerations of Calabi-Yau threefolds and K3 surfaces.

A principal makes decisions relying on the private information held by a biased agent. Lacking a prior over the circumstances that may arise, she seeks a delegation rule that minimizes worst-case regret relative to the complete-information benchmark. The optimal delegation mechanism governs two distinct margins: whether to depart from a common default action, and how much discretion to grant conditional on departure. This rule pairs the default action with an action gap, followed by a discretionary interval with randomized boundaries designed to balance regret across states. This structure is globally optimal within the class of all stochastic, incentive-compatible mechanisms.

Retrieval-augmented generation (RAG) systems use external documents for facts and rules that guide answers. We study rule-scope misapplication, where a rule valid within its documented scope is applied to a question outside that scope. Correct facts can then support an incorrect answer even when the rule's boundary is visible. This exposes the limits of document-validity checks. We introduce LENS (Local Evidence, Non-local Steering), an LLM-agent-based attack framework for constructing auditable instances of this failure. Its Planner identifies a grounded rule, its legitimate use, intended misuse, and document roles. Proposer and Verifier agents construct and repair attack documents using local retrieval and reader feedback while preserving factual support and explicit scope boundaries. Across four QA benchmarks and three target generators, LENS achieves 47.1% target-error rate over 4,000 fixed attempts, including construction failures. A separate fixed-attempt audit yields 36.1% joint target-error rate under independent validity checks. Fact-preserving interventions link errors to rule scope. For 76.2% of target-matching default responses, the same model correctly judges the rule inapplicable when explicitly probed. With matched information and budgets, an applicability audit reduces target error from 70.8% to 16.4%, compared with 31.2% for task-alignment checking. These findings motivate checking and enforcing rule applicability during answer generation.

Dimensionless descriptions of complex physical systems can sometimes be reduced further when multiple input groups combine into a single composite scaling coordinate. Identifying such structure from noisy response evaluations is difficult because apparent departures may reflect either genuine response structure or evaluation noise. Here we develop equal-response probing with replicate calibration to assess whether two dimensionless inputs support a one-group representation under prescribed directional and collapse tolerances. Equal-response geometry provides a candidate coordinate without direct numerical differentiation, while a second stage evaluates the accuracy of the resulting one-dimensional collapse. Replicate evaluations estimate the noise contribution to both stages, enabling reducible, not-red and unresolved outcomes. We evaluate the framework across eight model-based benchmark regimes spanning rough-pipe flow, cell-monolayer wound closure, microfluidic droplet generation and column buckling. Six regimes satisfy both criteria and two fail both, while known reduced coordinates are recovered in exact-reduction benchmarks. Controlled synthetic tests assess false-alarm behaviour, sensitivity to structural departures and noise calibration. These results show how replicate-calibrated equal-response probing distinguishes regime-specific scaling structure from variation attributable to evaluation noise while retaining an unresolved outcome when evaluation noise precludes a definitive verdict at the prescribed tolerances.

We study a non-stationary queueing system with multiple parallel servers and a finite buffer capacity, over a finite planning horizon. Motivated by an operation that is central to an EV (electric vehicle) battery-swapping station – how to best manage the charging of batteries swapped off from EVs, we develop a continuous-time MDP (Markov decision process) approach to the problem. The main technical challenge is to allow both the arrival rate and the service-cost rate to be functions of time with jumps at discrete time points, key features of the EV swapping station's operation. We derive an exact and explicit threshold structure that characterizes the optimal control policy. Moreover, we give conditions under which the thresholds are monotone over time, and with respect to key system parameters such as service and buffer capacities. Extensive numerical experiments are carried out to supplement as well as validate the theoretical results, and also to illustrate the robustness of the threshold structure in practical implementations.

Successive weaving segments (SWSs) on urban expressways are bottlenecks prone to recurrent congestion and collisions, requiring fine-grained active traffic management (ATM). Existing approaches struggle to balance the adaptive performance of data-driven optimization with the resilience and transferability of model-based control. We propose a hybrid framework to coordinate lane-level variable speed limits (VSLs) and ramp metering across SWSs. First, we reconstruct L-METANET, a lane-level macroscopic traffic flow model that captures free and forced lane changes. Second, we combine XGBoost-SHAP with a random-parameters binary logit (RPBL) model to derive analytical equations for merging and diverging collision risks and formulate system cost and reward functions. Third, we develop MPC-STMAPPO, a hierarchical controller integrating model predictive control (MPC) and multi-agent reinforcement learning (MARL). Its upper MPC layer uses L-METANET for long-horizon rolling optimization and generates baseline commands; its lower spatiotemporal MAPPO (ST-MAPPO) layer, enhanced with Mamba cells and graph attention, produces residual actions for short-horizon adjustment. Real-world experiments on the 18-km Eastern Expressway in Changchun, China, show that L-METANET accurately reproduces lane-changing-induced flow redistribution and capacity drops, with state evolution aligned with ground truth. XGBoost-SHAP-RPBL achieves AUCs above 0.80 in most tasks, outperforming conventional logit models. MPC-STMAPPO converges faster and performs better across multiple metrics than MPC- and MARL-based baselines. Under randomly fluctuating demand, it also significantly outperforms pure MARL in generalization, demonstrating strong potential for industrial deployment.

STEM student competitions aim to identify capable, science-interested students and further support them in developing their STEM-related abilities and interests. When such students do not participate, they may appear to be missed not only by the competition but by this form of support altogether; unless they are reached through other competitions. The present study therefore investigated to what extent students who would likely have succeeded in the German Physics Olympiad are missed by it. The study sample comprised 282 Olympiad participants and 1,103 non-participants from academic-track secondary schools, all assessed on 31 indicators spanning sociodemographic background, cognitive abilities, self-related beliefs, motivational variables, personality traits, vocational interests, and prior participation across other STEM competitions. Three machine learning models (elastic net, random forest, gradient boosting) were trained on the participant sample to predict first-round success based on those indicators. The elastic net was found to perform best. Although predictive accuracy was modest, the model proved well-calibrated, supporting valid group-level inferences. Only a few variables predicted success: mathematics skills, physics- and engineering-related skills, and having skipped a grade were positive predictors, whereas conventional vocational interests were a negative predictor. When applied to the non-participants, the model identified 43 students (3.9%) as potentially successful. Crucially, 74% of these students had previously entered at least one selective STEM competition, leaving only 11 students (ca. 1% of all non-participants) entirely unreached by the selective STEM competition system. These findings indicate that the majority of potentially successful non-participants are not absent from the selective STEM competition system but are engaged by it through other domains.

Successful intercultural communication requires more than grammatical competence. It demands sensitivity to culturally embedded social norms whose violation triggers subtle but meaningful nonverbal responses. For German learners of Mandarin Chinese, acquiring this sensitivity is critical yet poorly supported by existing language-learning agents. We present a Wizard-of-Oz (WoZ) study design and supporting real-time system for collecting multimodal behavioral data from native Chinese speakers reacting to social norm violations by German learners. The system features a photorealistic MetaHuman avatar driven by Live Link face capture and MediaPipe upper-body tracking, a wizard console for real-time behavior selection, and synchronized multimodal logging across agent and learner streams. A layered annotation framework, based on psychological theory and covering non-observable socioemotional reactions, norm interpretation, verbal, and observable behavior thereof, and future supervision targets enables the corpus to support training of future automated cultural interpretation and behavior generation models. Four ecologically valid interaction scenarios, developed with cultural and pedagogical experts, provide the methodological and technical foundation for a culturally adapted conversational agent for Chinese language learning.

Deploying, migrating, or scaling an agent changes its model, support system, application, and users. Calibration should align the agent with destination requirements without requiring source imitation or universal superiority. We propose three standard families: basic capability and operational reachability; technical and professional precision, including terminology, units, reporting periods, program fidelity, and report dependencies; and user context covering roles, use cases, authorized values, permissions, and budgets. A versioned mapping distinguishes requirements to preserve, transform, add, or retire. Four instruments—benchmarks, positive/negative examples, skills, and role-conditioned cost/utility—supply tunable reward channels within a fixed acceptance contract. Consultants annotate methods, branches, bounded loops, checkpoints, relative comparisons, and report refinement; a registry records applicable evidence. Adaptation revises configurations or trains a separate controller or skill while freezing the foundation executor. Independent hard gates govern qualification; RL and preference methods optimize candidates without supplying factual truth. Qualification and held-out improvement over the unchanged target are separate outcomes. Supplied intent labels permit a limited paired recount: small prompt/retry gains coexist with regressions and failed gates. FinQA supplements match prior summaries and retain provenance conflicts. These observations motivate destination-specific evaluation but do not establish calibration efficacy. We specify matched-budget tests, clean reload, complete-denominator reporting, and dependency revalidation. This is a methodological proposal with partially reconciled pilot evidence; confirmatory validation and the greater-than-80% zero-shot task-context acceptance target remain outstanding.

We show that the Anderson-Witting relaxation-time approximation (relativistic Bhatnagar-Gross-Krook (BGK) equation) emerges as the $N\rightarrow \infty$ limit of a stochastic $N$-body particle gas, in which collisions perform a full microcanonical momentum redraw that enforces local energy-momentum conservation at every event. Using the RAMBO algorithm to sample the Lorentz-invariant phase space, we derive the finite-$N$ single-particle momentum spectrum and show that it converges to the Jüttner-Boltzmann equilibrium distribution as the local particle number grows. We couple this relaxation kernel to an advection-relaxation splitting scheme to construct ShARK (Stochastic Advection Relaxation Kinetics), a 3D Monte Carlo relativistic solver for the relaxation time approximation Boltzmann equation. The framework is conceptually related to lattice Boltzmann methods, but formulated in continuous momentum space to avoid the accuracy loss caused by finite-order momentum discretizations far from local equilibrium. We validate the numerical results against analytical solutions for two highly symmetric conformal expansions, the Bjorken and Gubser flows. The framework provides a first-principles route to far-from-equilibrium relativistic transport, with applications ranging from heavy-ion collisions to expanding astrophysical plasmas.

We study the exact Euclidean ball-proximal point method for proper, closed, convex functions, where each iteration minimizes the objective over a ball centered at the current point. Retaining the objective gap in the decrease of squared distance yields sharper bounds on objective values, stationarity, and the symmetric Bregman distance to a minimizer. For constant radius $t>0$ and initial distance $D_0>0$ to the solution set, the objective gap after $K$ iterations is at most its initial value multiplied by $\exp(-2K^2t^2/D_0^2)$. We characterize convergence for arbitrary positive radius sequences. If their sum diverges, the method reaches a minimizer in finitely many iterations whenever one exists; otherwise, the objective values converge to the infimum and the iterates escape every bounded set. If the radii are summable, the iterates converge to a possibly nonoptimal point. Self-contraction gives finite trajectory length for bounded iterates and a constant-radius termination bound $O_d(1+D_0/t)$, whose implicit constant depends only on the dimension. A polyhedral family shows that the dimension-independent quadratic bound remains asymptotically sharp. We also identify a minimum successful geometric decay factor and analyze adaptive radius rules based on subgradients or objective gaps, epigraph reformulation, and relaxed updates. Together, these results strengthen the foundations and convergence guarantees of the method without assuming smoothness or strong convexity.

We present a theoretical framework for handling quantum nonlinear processes in epsilon-near-zero (ENZ_ nano spheres, based on macroscopic quantum electrodynamics and the resonant states (RSs) expansion of the Green's tensor in nano cavities. Using third-harmonic generation (THG) as a case study, we derive analytical expressions for the generation efficiency, and the quantum fluctuations of the generated nonlinear field, as a function of different quantum states of the impinging pump field. Finally, we discuss how to generalise this theoretical framework to arbitrarily quantum nonlinear processes in photonic nanostructures.

Sabine's foundational theory established the cornerstone of modern architectural acoustics. However, it fails to predict zero reverberation time in perfectly absorptive rooms. Eyring subsequently addressed this issue by proposing a new formula, which Knudsen later extended to incorporate air absorption. Nevertheless, Eyring's underlying approach contains fundamental theoretical contradictions. To resolve these flaws, the author previously introduced macroscopic and microscopic models for a revised reverberation theory; however, their mathematical unification, rigorous derivation from differential equations, and the mechanism behind Eyring's underestimation remained unclarified. This paper presents a comprehensive generalization and mathematical foundation of the revised theory. The macroscopic model is derived directly from fundamental energy differential equations, incorporating air absorption and a generalized mean residual free path. By reformulating the microscopic model as a sequential convolution process, its fundamental consistency with the macroscopic model is mathematically demonstrated. Crucially, it is shown that Eyring's underestimation can be interpreted as stemming universally from the structural omission of temporal variance expansion ({\sigma}_{n}^{2}=n{\sigma}^{2}) inherent to multiple reflections, regardless of the assumed probability distribution. Finally, ray-tracing simulations validate the scale-invariant accuracy of the proposed theory. Ultimately, this mathematically consistent framework establishes the theoretical limit for future generalized theories in non-diffuse sound fields.

We compute the potential-energy surfaces for molecular ions up to H$_2$O$^{3+}$ with any combinations of outer-valence, inner-valence and core holes. To obtain the potential-energy surfaces, we consider several ab initio techniques based on the CASSCF method and benchmark them with respect to the literature. We thus identify the techniques which produce accurate potential-energy surfaces as a function of the two bond lengths and the angle between the two bond lengths. These potential-energy surfaces will be used in future studies of the interaction of water with free-electron laser pulses. Furthermore, the techniques employed in this work in the context of water can be used to obtain the potential-energy surfaces of other triatomic molecules.

AI agents increasingly carry out long-horizon professional work, but their evaluations rarely require a finished creative deliverable. To this end, we introduce Timeline-Bench, a benchmark of 56 real video-editing tasks, each asking an agent to turn raw production material into a finished video. Tasks range from selecting dialog takes and shaping interview footage into a story to cutting commercials from product shots, voiceovers and graphics. Every task provides a brief, source assets, a container and a set of tests. A task is resolved when the output passes every test. The tests check the delivery format, the content and the brief's explicit requirements, and include a quality test calibrated on 2,582 blind judgments by 43 video editors. We evaluate 16 agents that pair frontier models with coding-agent harnesses such as Codex, Claude Code and OpenCode. The best, GPT-6 Astra in Codex with curated editorial guidance, resolves only 15 of the 56 tasks (26.8%), and the average agent resolves 14.0%. Human editors prefer the reference edit in 83.5% of judgments. Most unresolved runs (562 of 771) fail only the quality test: agents perceive footage through stills and transcripts and check their renders for defects, not craft. We release the tasks, verifier and per-run results at https://timelinebench.tensortest.com.

Aviation authorities worldwide expect Advanced Air Mobility (AAM) traffic management to be decentralized among service providers, requiring AAM flights to autonomously plan trajectories by predicting other flights' control inputs rather than relying on centralized coordination. Game-theoretic approaches that formulate multi-agent collision avoidance as an exact dynamic potential game can efficiently find open-loop equilibria, but they assume that agents exactly follow their equilibrium trajectories—an unrealistic assumption given uncertainties in actuation, perception, and computation. We propose a strategically robust formulation where each agent protects against a fictitious adversary that, for each timestep, perturbs other agents' control inputs within a bounded budget to minimize distance at that timestep. We show that, under reasonable assumptions on agents' distance cost and robustness levels, the strategically robust game remains an exact dynamic potential game and admits a quasi-closed-form solution to the inner adversarial problem for linear dynamics, which limits computational overhead. Experiments with up to eight agents using logarithmic distance costs show that strategic robustness selects more robust trajectories in high-collision-risk configurations while leaving low-risk trajectories nearly unchanged, with only a modest increase in runtime.

We propose a reinforcement learning (RL) framework that tunes both Real-Time Optimization (RTO) and Economic Nonlinear Model Predictive Control (ENMPC) to address plant–model mismatch in process systems. Drawing on modifier-adaptation concepts, the method parameterizes the dynamic model, stage costs, constraints, and RTO modifiers, and uses Q-learning to adjust these parameters at two timescales: a fast update for the ENMPC layer and a slow update for the RTO layer. The framework is experimentally validated on a laboratory rig emulating a three-well subsea oil-production network. Using plant measurement data, the proposed RTO-RLMPC scheme achieves 8.6% higher economic profit than nominal ENMPC, preserves input feasibility of the deployed ENMPC policy and empirically satisfies the path constraints under disturbances and measurement noise, and drives the learned model parameters toward reference plant values, independently identified from experimental data, along the directions that most affect the economic objective. This work provides one of the first experimental demonstrations of RL-tuned ENMPC integrated with RTO.

Reinforcement learning (RL) post-training often uses distinct GPU kernels for rollout and policy update. In synchronous PPO and GRPO, numerical disagreement can perturb ratios between current token probabilities and those assigned during rollout. Recomputing rollout log-probabilities with the policy-update backend avoids this discrepancy but adds a forward pass. Bitwise-consistent unified kernels permit reuse when the policy snapshot and probability processing match the objective. Their optimization must preserve agreement across distinct execution regimes. We present KernelBraid, an agentic framework starting from a hand-tuned, bitwise-consistent implementation. Its optimization intermediate representation (IR) organizes source-code search by linking implementations and modifications to numerical requirements, workload measurements, and derivation history. The agent coordinates changes and retains verified intermediates for further exploration; promotion requires passing correctness checks and improving aggregate latency within per-workload limits. Across 12 end-to-end training configurations on H20, KernelBraid achieves 1.10x average throughput relative to AReaL with log-probability recomputation, and the mean training-reward ratio rounds to 1.00x. Isolated-layer profiling yields 1.40x average speedup in summed phase time across 15 model-GPU pairs. Operator-level evaluation covers correctness and performance for 10 operators on A100, H20, and H200, all passing the prescribed bitwise checks. Unified-attention search achieves 2.52x speedup in summed workload latency over the starting implementation using 7M LLM tokens; ablations assess the contributions of retained evidence and branch exploration to search efficiency and attained performance. Our code is open-sourced at https://github.com/areal-project/AReaL-TIK.

Multi-document retrieval-augmented generation (RAG) requires a language model to process multiple retrieved text chunks before answering a question. Precomputing each chunk's KV cache independently and concatenating the caches when the chunks are retrieved can accelerate this step. However, the assembled cache lacks cross-chunk attention information, reducing answer quality. Selective recomputation methods recover the missing cross-chunk context by rerunning the target LLM on selected tokens, incurring substantial online computation. We introduce CacheRepair, a lightweight network that learns the difference between independently computed KV caches and those produced by processing the chunks together. The network combines compressed KV features with token embeddings and uses attention that is bidirectional within each chunk and flows from earlier to later chunks. Each repair block receives the compressed cache features, and the predicted residual is added to every document token's cache. Each repair network is trained for a specific frozen target LLM on a generic retrieval corpus and reused across downstream datasets. Our analysis shows that repair reduces KV errors both near chunk boundaries and throughout chunk interiors. Evaluation across three target LLMs and four downstream datasets places CacheRepair on the measured answer-quality-latency Pareto frontier in eleven of twelve model-dataset combinations. Reported time to first token (TTFT) includes online cache transfer and repair. Across all twelve combinations, the largest repairers achieve 1.69-4.61$\times$ speedups in median TTFT over full prefill and improve mean F1 by 2.1-26.1 percentage points over direct cache reuse.

Latent world models predict future states for goal-directed planning using action chunks spanning multiple primitive steps. Existing methods typically use fixed-length chunks and either omit goal-conditioned action generation or limit their supervision to short goal spans. We introduce FlexiWorld, a JEPA-based world model that combines mixed-span goal supervision with variable-length action chunks to improve long-horizon control. During training, we sample varying goal spans and randomly partition the actions into variable-length chunks. We jointly train the world model with a causal action encoder that embeds variable-length chunks and an autoregressive actor that generates primitive actions sequentially. Student Forcing reduces exposure bias by training on generated action prefixes. For planning, Actor-Residual Cross-Entropy Method (ARCEM) combines action-residual search with within-chunk autoregressive feedback and chunk-boundary latent prediction. Across four benchmarks and goal distances, FlexiWorld with ARCEM achieves 89.29% mean success, compared with 83.98% for the strongest baseline. PushT ablations show improved direct control from mixed-span supervision, variable-length chunks, and Student Forcing. Without retraining, FlexiWorld supports different planning chunk lengths: longer chunks accelerate ARCEM by approximately $1.3\times$ on average while maintaining comparable average success.

AI advice increasingly enters human decision workflows before costly information acquisition and final action. This paper studies how such advice affects decision quality, measured by the gross utility of the final action. Advice shifts beliefs before the decision maker chooses an acquisition experiment with uniformly posterior-separable (UPS) costs. The analysis characterizes gross effects through the continuation gross value generated by net-optimal acquisition. Convexity of this value function is necessary and sufficient for every AI signal to weakly improve gross utility at every prior, while nonconvexity allows a binary signal to cause gross harm. The criterion gives tight state-space results. Every AI signal is weakly gross-improving in binary-state problems under arbitrary actions, payoffs, and UPS costs. A minimal three-state problem can generate gross harm through concentrated shutdown, where valuable acquisition stops. For classification problems with Shannon acquisition costs, the criterion becomes an active-set slope test. It guarantees weak gross improvement for homogeneous classification and all three-state weighted classification problems, while a minimal four-state failure arises from attention dilution, where acquisition shifts away from high-stake distinctions. Along a Blackwell-increasing mixture path, gross utility can first decrease and then increase. With endogenous interpretation, cheaper interpretation can reduce gross utility.