Papers

Batch black-box optimization is important for expensive systems where many candidates can be evaluated in parallel, yet only a few rounds of evaluation are affordable. In this regime, effective batch construction must identify multiple promising regions while avoiding redundant evaluations, and conventional Bayesian optimization can require increasingly difficult batch-selection procedures as the batch size grows. We introduce Flow Matching for Thompson-Sampling Refinement (FMTR), an amortized approach that learns offline a continuous, history-conditioned proposal distribution from which variable-size batches can be sampled. FMTR is trained offline by conditional flow matching on Thompson-sampling argmaxes refined toward high-value regions via a horizon-dependent Langevin pushforward map, adapting the exploration-exploitation tradeoff over the optimization horizon. At test time, the model parameters remain fixed and batch generation requires neither task-specific surrogate refitting nor numerical acquisition optimization. FMTR achieves strong performance against batch BO and in-context baselines on synthetic and real-world benchmarks, while substantially reducing the computational cost of batch selection.

Current video object removal methods suffer from prohibitive latency (minutes per video) and non-causal processing that prevents real-time interaction. We present StreamErase, enabling sub-second latency with up to 26 FPS streaming generation on a single GPU. Our approach first augments a text-to-video model with mask input to generate high-quality videos that erase target objects and synthesize coherent backgrounds, but does not perform inference on the fly. As such, we progressively transfer the erasing capability from a powerful bidirectional foundation model to an efficient unidirectional streaming eraser, enabling long-horizon erasing without sacrificing visual fidelity. Several key challenges arise when generating videos over long, potentially infinite horizons: (1) bridging the domain gap from training on finite length and extrapolating to long horizons, (2) sustaining high quality by preventing error accumulation, and (3) maintaining fast inference, without increasing computational cost as the context grows. To address these challenges, we carefully designed sliding-window causal attention with attention sinks. By incorporating self-rollout with attention sinks and KV cache rolling during training, we simulate inference-time extrapolations with a fixed context window, enabling constant-speed generation of arbitrarily long videos. Finally, we establish a dedicated benchmark, LongErase, for streaming long-video object removal. Our model achieves strong performance compared to existing video object erasing models and generalizes well to real-world video scenarios, uniquely enabling infinite-length streaming. With StreamErase, users can input a mask and see the erasing results unfold in real-time, delivering a truly interactive experience.

Vision-language models (VLMs) are moving onto edge devices, but little is known about how much energy they draw on single-board computers with NPU or GPU accelerators, or where that energy goes. We measure total-system energy, latency, and thermal behavior of Qwen2-VL-2B-Instruct and its text-only sibling on a Raspberry Pi 5 with a Hailo-10H NPU and on a Jetson Orin Nano Super, on both CPU and accelerator, and separate the fixed energy of each request from the energy that grows with output length. Because the accelerators use different quantization and serving software, we compare deployment stacks rather than silicon. Three findings stand out. First, NPU offloading makes the Pi a much more efficient VLM platform: it cuts energy per generated token by about two-thirds, leaves the host CPU largely free, and avoids the thermal throttling that the CPU backend reaches in sustained use. Second, an image adds both fixed and per-token energy, but at typical response lengths generating the answer costs far more than processing the image. Third, the Jetson GPU favors short responses through its lower fixed cost per request, whereas the Pi NPU favors longer ones through its lower cost per output token; their energy curves cross near 40 output tokens.

Tactile sensing is becoming central to robust, adaptive, and dexterous robotic manipulation. Beyond its traditional role in contact detection and reactive control, touch is increasingly used to learn transferable representations, condition manipulation policies, and support embodied physical reasoning. Yet existing surveys largely emphasize tactile hardware and perception, leaving the rapid shift toward learning-based tactile manipulation insufficiently synthesized. This paper reviews tactile-guided skill learning from a learning-centric perspective, organized by the functional role of touch across the learning and control pipeline. We examine tactile representation and policy learning, tactile foundation models, predictive and adaptive control, and generalization across objects, tasks, sensors, embodiments, and simulation-to-real transfer. Across these directions, we identify common design principles and remaining challenges in learning transferable physical representations, integrating semantic reasoning with contact-level control, and scaling tactile learning across heterogeneous data and hardware. Finally, we outline emerging directions toward tactile foundation and world models that can predict, reason about, and adapt physical interactions. Together, this review characterizes the transition of touch from a task-specific feedback signal toward a core modality for general-purpose embodied intelligence. An accompanying database of the reviewed literature and associated resources is available at https://sites.google.com/view/touch-in-robot-learning-review.

We present eMOOD, an entropy-guided evolution of the Multidimensional Optimal Order Detection (MOOD) framework for high-order finite-volume methods on unstructured meshes. A novel troubled-cell indicator based on numerical entropy production replaces the classical relaxed discrete maximum principle. Using entropy generation as a physically motivated measure of non-smoothness, the detector targets shocks, contact discontinuities and under-resolved flow features with a compact stencil. Compared with conventional MOOD detection, eMOOD reduces flagged cells by up to 75%, improving efficiency and reducing memory-access overhead. The detector is used within both the classical MOOD cascade of progressively reduced-order schemes and a novel MOOD-CWENO hybrid, which retains inexpensive linear reconstruction in smooth regions and activates nonlinear high-order reconstruction only where needed. Implemented in the open-source parallel solver UCNS3D, the methodology is assessed on the Shu-Osher shock-entropy-wave interaction, double Mach reflection, supersonic forward-facing step, shock-helium-bubble interaction and triple material point problems. These benchmarks encompass strong shock interactions, shock-induced instabilities, multi-material interfaces and vortex-dominated flows, testing robustness, shock capturing and accuracy on unstructured meshes. Results show that eMOOD preserves the accuracy and robustness of reference CWENO and MOOD formulations while achieving speed-ups of up to 20% without loss of solution quality. Its compact detection stencil, reduced memory footprint and improved localisation offer a more scalable and efficient framework for massively parallel high-order compressible-flow simulations. UK Ministry of Defence (copyright) Crown owned copyright 2026 AWE.

We study the route-set allocation problem between end nodes in capacity-constrained flow networks with transfer costs. We introduce an origin–destination (OD) bidding language, in which a participant bids for bounded conic combinations of inlet–outlet–quantity licences and the mechanism performs the routing through a network-flow optimising winner-determination problem. We then study payment rules for the OD mechanism. Under a restricted model with openly and uniformly priced sources and bidders who are indifferent to the source of the gas, the Vickrey–Clarke–Groves (VCG) payment is core-stable; the key structural fact is that in this regime some optimal aggregate flows never use a physical pipe in both directions, so the winner-determination problem reduces to a single-commodity network flow whose value is submodular in the set of active bidders. If source access is allowed to depend on the individual consumer – a realistic feature, e.g. when gas deals are embedded in bilateral political bargaining – the reduction fails and VCG payments may lie strictly outside the core. For this general model we propose a dual (nodal) pricing rule derived from the shadow prices of the flow-conservation constraints, and prove it is always core-stable, at the cost of dominant-strategy incentive compatibility. Illustrative examples accompany each result.

Human mesh recovery (HMR) provides a geometric foundation for understanding human motion, but accurately reconstructing both body motion and fine hand articulation from videos remains challenging, especially in real-world scenes where occlusion and partial visibility are common. Video generation pretraining offers rich spatial and temporal priors for this task, but its focus on visual synthesis leaves a gap between these representations and explicit human geometry. We present DreamHMR, a general framework that adapts video generation priors for complete body-and-hand reconstruction. It follows a global-to-local strategy, first grounding video representations in human geometry and then refining body and hand detail with reliable local evidence. Global Body Representation Transfer trains the video backbone to predict complete human renderings, turning its native prediction task into a means of learning surface correspondence and posed geometry. Local Joint Feature Enhancement combines contextual estimates with reliable local evidence to recover fine body and hand detail, while limiting interference from occluders and background. Experiments on five datasets with four video backbones demonstrate strong reconstruction performance across backbone families. Under a unified evaluation protocol, our best model outperforms the best existing methods on all reported metrics of 3DPW, EMDB-1, and RICH, reducing body MPJPE by 4.2%, 4.7%, and 6.1%, respectively. In more challenging occlusion settings on held-out BEDLAM and EgoBody sequences, DreamHMR reduces overall body and hand MPJPE by 7.0% and 10.0%, respectively. Our code will be made publicly available.

Large language models can reduce vocabulary mismatch in sparse retrieval by generating alternative views of a query. Once these views are searched independently, however, their ranked lists may be complementary, redundant, or affected by query drift, making fixed fusion suboptimal. We present XViewRank, a lightweight retrieval-side method for conditionally fusing BM25 results from an original query, a hypothesis view, and a step-back view. For each pooled candidate, XViewRank retains three normalized BM25 scores and three query-level overlaps between the retrieved result sets. A small nonlinear scorer uses the overlaps as context that changes how the view-specific scores are interpreted, without jointly encoding query and passage text. Under a matched-input protocol, XViewRank improves nDCG@10 and MAP numerically over reciprocal rank fusion on both TREC DL 2019 and 2020, with a statistically significant MAP gain on DL19. Additional analyses show that the clearest benefit occurs when the view rankings have low agreement, while the evidence on DL20 is weaker. The results support overlap-conditioned score fusion as a compact retrieval-side alternative to fixed fusion, while also delimiting where its advantage is empirically established.

Rapid progress in AI agents and robotics makes it a matter of the near future that humans are served by a team of robots. Such a team will receive several goals at once, each specified in whatever form is convenient for the human, such as a category name, a photo, or a spoken description. Existing benchmarks study multi-agent, multi-modal, and multi-object goal navigation in isolation, and a benchmark that places all three in one episode is missing. To fill this gap, we introduce HM3D-M3. A team of K agents starts at distributed locations in a photorealistic indoor scene and searches for an unordered set of M object goals, each given as an open-vocabulary category, an instance image, or an instance description. HM3D-M3 provides 1,000 validation episodes over 3,926 goals from 379 open-vocabulary categories, 100K observation-assignment training samples, and two team-level metrics, goal-level SR and makespan-normalized SPL. We further contribute two baselines. HiGA is a trained hierarchical goal assigner on top of a frozen open-vocabulary navigation policy, and LaViRA-M3 is a training-free team of VLM agents with shared mapping and progress analysis. Adapted multi-robot and zero-shot methods reach 16–27% SR, while our baselines reach 38–41% with complementary strengths and distinct failure modes. Furthermore, we deploy LaViRA-M3 on a heterogeneous three-robot team. Project page: https://robo-lavira.github.io/hm3d-m3/

We present several advances on neural operators by viewing the action of operator layers as the minimizers of Bregman regularized optimization problems over Banach function spaces. The proposed framework allows interpreting the activation operators as Bregman proximity operators from dual to primal space. This novel viewpoint is general enough to recover classical neural operators as well as a new variant, coined Bregman neural operators, which includes the inverse activation operator and features the same expressivity of standard neural operators. Numerical experiments support the added benefits of the Bregman variant of Fourier neural operators for training deeper and more accurate models.

Multi-agent path finding (MAPF) solvers are evaluated on the schedules they produce, not on how well those schedules can be executed. Specifically, continuous-time optimal solvers can achieve even lower cost than their discrete counterparts by scheduling agents with near-zero temporal/spatial separation, and by exploiting graphs with higher connectedness; however, such schedules may prove hard to execute under a realistic controller. We integrate AOC-CBS, an anytime-optimal continuous-time MAPF solver, into a fleet coordination framework in which each agent tracks its timed path using model predictive control, and we evaluate the resulting system on congested MovingAI benchmark instances with 21 to 40 agents. Without intervention, only a fraction of the instances complete execution. We then apply T-robustness by adding a time-margin, i.e., a minimum temporal gap between agents at shared locations, as well as spatial inflation of the agent footprint. Increasing either margin raises execution success, for a relatively small increase in the schedule's sum-of-costs; it also improves how closely agents adhere to the schedule: when tracking each agent's position throughout execution and comparing it against the position it should occupy according to the schedule, the mean discrepancy falls significantly as the margin increases. In general, the two margins have different effects; temporal margin decreases deviation, giving the controller room to stay on schedule. Spatial margin affects deviation less and instead increases the clearance between agents, so that the same deviation no longer leads to a blocking stall. They are nonetheless substitutes rather than complements: spatial inflation is almost free in schedule cost, but its effect on execution success is small and does not compound with a temporal margin already in place.

KANLab is a browser-based visualization tool designed to interactively teach the core concepts of Kolmogorov-Arnold Networks (KAN), forked from the widely successful TensorFlow Playground. To maximize educational clarity, KANLab omits residual connections and learnable scaling parameters found in modern KAN implementations, ensuring that displayed B-spline activation curves directly correspond to their underlying trainable control points. Users can inspect and modify activation functions, initialization schemes and observe gradient descent correcting manual disruptions. Furthermore, user can explore phenomena such as activation variance, interactive pruning, and re\-/training within a compact 2D machine learning setting. KANLab is open-source and available at https://kanlab.app/

Long coding agent sessions outgrow the model's context window, so agents compact: older history is replaced by something shorter, usually a model-written summary. A summary selects and rewords, and the agent cannot see what was lost. Existing work has explored a range of compaction techniques, but comparatively little attention has been paid to their effectiveness with lightweight models, including those deployed locally or used for high-volume workloads. Here, we present transparent-compact, a simple deterministic method for the open-source Pi coding agent, designed specifically for use with such models and with the purpose of maximising the transparency of the agent's view of its own compacted context. It guarantees that user and assistant prose in its context is unchanged by a model or algorithm, and explicitly flags where compaction has stripped information. We compare it with Pi's default model-written summary compaction and with pi-vcc, a popular deterministic method with similar methods but without verbatim retention of prose. Our benchmarks run on four models of varying performance levels, with emphasis on Qwen3.8-27B: a lightweight model at the top end of what consumer hardware can serve locally. On sets probing commitments (rules the user set which exist only in the conversation) and recall of facts, transparent-compact matched or outperformed both alternatives. In particular, it showed an advantage in avoiding confident blunders, i.e. assertion of false information as opposed to recognizing that something was not recalled. Even with a frontier model, its advantage was present but smaller. We argue that transparent-compact spares lightweight models from reasoning or searching their way around what compaction lost, and results in improved recall and application of information as a result, without the need for task or context-specific structuring of the session history.

Speech Emotion Recognition (SER) models and Audio LLMs fail when vocal tone contradicts textual semantics. Extensive training on congruent data engenders a rigid semantic prior, causing "Semantic Dominance", a severe text bias during cross-modal conflicts. We introduce the ASPIRE benchmark of adversarial audio-text pairs across four conflict types, alongside two metrics: Semantic Overconfidence Penalty (SOP) and Latent Decoupling Degree (LDD). Unlike fusion methods that merely re-weight collapsed features, our Acoustic Conflict Resolution Network (ACR-Net) uses Cross-Modal Attention and Contrastive Decoupling Loss to disentangle contradictory representations into orthogonal latent spaces. Experiments show existing models fail under polarity conflicts due to extreme semantic overconfidence. Conversely, ACR-Net preserves acoustic fidelity, maintaining high LDD to drive SOP near zero, achieving superior Acoustic Accuracy (ACC) in both adversarial and standard scenarios.

Prior work on implicit bias shows that gradient descent selects the Frobenius max-margin direction on linearly separable problems, whereas Fan et al. (2025) established spectral max-margin behavior for spectral-norm steepest descent methods, including idealized Muon with exact orthogonalization. In practice, however, Muon approximates orthogonalization using finite Newton-Schulz iterations and a positive stabilization constant. We prove that these implementation choices cause Muon to select the Frobenius rather than the spectral max-margin direction in linear classification with separable data. We also establish this behavior for both exponential moving average and Nesterov-style momentum. The key technical ingredient is a summable perturbation argument showing that the resulting parameter trajectories remain at bounded distance from those of gradient descent. We then disentangle the roles of Newton-Schulz iteration count, the stabilization constant, and the Newton-Schulz polynomial, revealing distinct asymptotic behaviors, including directions that achieve neither Frobenius nor spectral max-margin. Finally, we show that Muon can recover its spectral max-margin bias with Newton-Schulz iterations growing logarithmically over training.

Distilling pretrained video diffusion models into few-step autoregressive (AR) generators typically combines student initialization with distribution matching distillation (DMD). An intermediate AR teacher can provide causally compatible targets for ODE initialization. However, causally compatible targets alone do not ensure high-quality generation under a short sampling schedule. We analyze the posterior-mean predictions favored by direct latent regression and empirically find that causal ODE initialization alleviates blur but still leaves fine detail unresolved. To address this limitation, we propose Representation Forcing, which learns few-step causal initialization through representation-space supervision rather than intermediate-teacher trajectory regression. The student is trained with teacher forcing by comparing clean predictions and training targets through normalized features of a frozen video diffusion transformer. This phase reuses the synthetic videos generated by a bidirectional model for AR teacher training and replaces both teacher training and causal ODE distillation, without generating additional causal ODE trajectories. The subsequent asymmetric DMD procedure remains unchanged. Empirical results show that, at four sampling steps, RepForcing better preserves generation quality than intermediate-teacher-based initialization before DMD. After DMD, our four-step chunk-wise model improves VBench Total from 84.04 to 84.66 (+0.62 points) and VisionReward from 6.326 to 10.074 (+59.25%) over the reported causal-ODE-based baseline.

Data movement between memory and processing units remains a major bottleneck in data-centric systems. Memristive Stateful Logic (SL) addresses this problem by performing Boolean operations directly on stored resistive states. However, arithmetic efficiency depends jointly on the device-supported logic primitive set, the dependency structure of the resulting operation sequence, and its mapping to datapath architectures. While prior designs typically optimize device, logic, and architecture layers separately, we jointly consider them to co-design a stateful adder. We experimentally demonstrate a mixed MAGIC and FELIX logic primitive set on printed Electrochemical Metallization (ECM) memristors and derive a full adder sequence optimized for inter-bit carry propagation. Mapping this sequence to bit-serial, bit-parallel, and bit-pipelined datapaths reduces the latter two to 2n+6 cycles for an n-bit addition. For 32-bit addition, this requires 60-88% fewer cycles than prior SL adders. Across four representative workloads, we achieve 3.7x-16.4x speedup over prior. These results show how cross-layer co-design can translate into improved workload-level performance.

Neural symbolic regression (SR) models require diverse datasets of mathematical expressions for robust training. However, existing datasets rely predominantly on stochastic sampling methods, which suffer from high structural redundancy and incomplete functional coverage. To address this, we introduce an exhaustive dataset of univariate canonical skeletons generated via a deterministic constrained context-free grammar. A canonical skeleton acts as a generic structural representation that subsumes more specific functional forms, eliminating redundant expressions and avoiding repeated computational efforts. Although our dataset consists of univariate skeletons, it is designed to support the decomposable SR paradigm to solve multivariate SR problems without assuming linear separability. Furthermore, we pair this dataset with a sampling strategy that generates clean numerical data without triggering numerical instabilities or relying on artificial protected operators. Canonicalizing two existing neural SR datasets reveals that our dataset is substantially more compact, eliminates structural repetitions, and achieves a uniform operator distribution free of simplicity bias. Finally, we establish a benchmark of 150 SR problems across three tiers, on which state-of-the-art SR methods are shown to struggle significantly.

Web agents built on vision language models decide every action by generating text such as click(12), although most of that text encodes a choice among a few action types and a few dozen marked elements. We read these choices directly from the hidden states that the generator computes while it reads the goal and the screenshot. Plug-in decision heads on a frozen generator predict the action type, pool the visual tokens inside each numbered box to pick the target, and point at a listed text option or at a WRITE slot that leaves the writing of new text to the generator. A confidence cascade executes the heads' action when they are sure and otherwise lets the generator decode from the KV cache of the same forward pass. With Qwen3-VL 4B and 8B generators trained on 213k web steps, a strict threshold keeps the cascade between 0.3 points below and 1.4 points above the generator on seven offline test sets while the heads decide 12 to 64% of the steps. A looser threshold raises step success on unseen Mind2Web websites by 3.1 and 1.8 points, costs up to 2 points on GUIAct and MiniWoB++, and lifts WebArena task success from 12.5 to 13.8% (4B) and from 13.6 to 14.6% (8B) with less than half of the generated tokens. Training the heads jointly with the generator lowers the generator's own step success by up to 4.9 points; heads on a frozen generator avoid this cost.

We approximate the solution operator $(a,f)\mapsto u$ of $-\operatorname{div}(a\nabla u)=f$ by operator networks whose only primitives are heat-semigroup functions of a fixed reference operator and pointwise products. The main result is an emulation theorem on strongly local Dirichlet spaces with a spectral gap and a Bakry–Émery bound $\mathrm{BE}(-K,\infty)$: an explicit scheme reaches $\dot H^1$-error $\varepsilon$ uniformly over coefficients with bounded carré du champ and bounded generator image and over $L^2$ data, with parameters depending only on the coefficient bounds, the accuracy and $K$; on the torus this gives a $C^2$-type coefficient class with constants independent of the dimension. The torus version refines this to Wiener classes of fractional regularity $A^{1+\theta}$, with an explicit rate attained by the scheme's product-defect block. The scheme unrolls the Laplacian-preconditioned Richardson iteration, controls each inexact step in the operator norm of $\dot H^1$ without Galerkin truncation, and synthesises the coupling $\nabla a\cdot\nabla u$ from the product defect of one heat layer. All parameters are explicit; the results are approximation theorems for continuum operators.

Autoregressive transformers trained on limit order book (LOB) message streams are increasingly proposed as foundation models for market microstructure. Because the generative process of real markets is unknown, these models can be judged only by the realism of their output, leaving open what they actually learn. We make the question answerable by inverting the setup: we train GPT-style models on order flow from simulators whose latent parameters are known, so that the Bayes-optimal predictor, the generating state, and the exact data entropy serve as references. Measured against them, next-token pretraining on order flow implements amortized Bayesian inference over market regimes. The learned predictive matches the exact Bayes optimum to within a millinat per token and approaches the oracle floor even on a nonstationary family whose latent state drifts within an episode, a regime that requires filtering. The latent regime is linearly decodable from frozen activations, well above baselines, so a frozen-model read-out calibrates simulator parameters more accurately than windowed maximum likelihood at a fraction of the cost. In our experiments these capabilities are not produced by scale: capacity and token volume saturate early, and it is the diversity of training regimes, not the token count, that determines generalization to unseen markets. On real NASDAQ data, zero-shot transfer is set by how well the simulator prior covers real flow: an enriched mixture nearly closes the gap to an in-domain baseline, and pretraining beats from-scratch training at every budget across held-out tickers. Known ground truth turns average next-token loss, which is uninformative near the entropy floor, into an exact comparison against the Bayes-optimal and oracle predictors.

We investigate two aspects of multilingual pre-training with limited data, namely how language exposure is distributed across English, Dutch, and Chinese, and how that exposure is ordered during training. In particular, we compare balanced multilingual exposure to language-dominant exposure, where one language is exposed with more training data than the others. We also compare mixed training, where multilingual training data is used throughout training, with curriculum training, where monolingual exposure of the dominant language is followed by balanced multilingual exposure. Our results show that multilingual BabyLM performance varies with both the amount and order of language exposure. Mixed training outperforms curriculum learning across all 15 matched configurations, indicating that simultaneous multilingual exposure is more effective than staged exposure in our setting. Several language-dominant mixed training configurations achieve higher scores than the balanced configuration, suggesting that a balanced distribution is not necessarily the highest-performing configuration in our setting. Interestingly, the Dutch-primary configurations achieve the best overall performance.

Graphical user interface (GUI) agents now execute workflows across real applications, but when execution drifts, users must determine what changed, what should have remained untouched, and what can still be recovered. Plan-centric interfaces leave users supervising procedure, but users live with changes to application state. We introduce task-state governance, an interaction paradigm that represents a task as an editable contract linking desired outcomes, facts that must be preserved, and observed application state. We instantiate this paradigm in TaskVM, which projects live state into editable task controls, executes user edits as genuine interface actions, verifies outcomes through independent observation, and marks each action as reversible, compensatable, or irreversible. Across an automated contract benchmark, a controlled user study, and an open live-model exploration, the paradigm strengthened contract compliance and users' recovery control and trust, while execution outcomes were held constant. Task contracts open a design space for governable agent interfaces.

Large language model (LLM) agents are increasingly used for long-horizon research and engineering, yet evaluating these capabilities remains expensive, time-consuming, and heavily dependent on human expertise. This motivates a natural question: can LLM agents themselves construct efficient benchmarks that preserve the conclusions of costly reference evaluations? We study this capability by formulating automatic benchmark development as a meta-evaluation problem, where an agent is given a task and an expensive reference benchmark and requested to construct a lower-cost proxy benchmark that faithfully distinguishes among candidate models and solutions. Based on this formulation, we introduce EvalBuilder, a framework for evaluating such benchmark-generation capabilities along three key dimensions: fidelity, efficiency, and robustness. Specifically, we assess whether generated benchmarks preserve model rankings, generalize to held-out models and freshly collected responses, and provide well-calibrated and discriminative scores while substantially reducing evaluation cost. We further provide an iterative environment in which agents can refine their benchmarks under a fixed time budget. Our framework provides a systematic testbed and benchmark for understanding when current LLM agents can develop useful evaluators and how benchmark construction itself can be increasingly automated.

Many complex systems involve group interactions, which can readily be represented by higher-order structures such as hypergraphs. However, when observing a system's dynamical evolution, establishing whether that evolution actually results from higher-order interactions, or whether pairwise interactions suffice to explain the data, remains a challenge. Even in the simple case of a temporal sequence of nodes visited during a diffusion process, such a task is intricate. Indeed, group interactions leave their traces as memory on the observed nodes, while path dependence in a purely pairwise process can mimic that same footprint without any group structure. To resolve this, we propose an inference pipeline based on a hidden Markov model (HMM) that tests, directly from an observed sequence of nodes carrying no group labels, whether higher-order interactions are required to describe the dynamics. The HMM latent states represent higher-order interactions, seen as hyperedges of an unobserved hypergraph, so that their detection becomes a model selection problem between the HMM class and a class of null models without higher-order interactions. We first validate this approach on synthetic node sequences generated from random walks on graphs and hypergraphs, showing that it recovers both the number and the composition of the underlying hyperedges, while correctly rejecting them when the generating process is pairwise. As a second step, we consider a validation based on empirical data, namely empirical trajectories of ball possession in football matches. This application shows that the pipeline goes beyond providing statistical evidence of the presence of higher-order interactions, and yields interpretable underlying structures: the recovered hyperedges map directly onto recognizable tactical team formations.

Near-field beamfocusing confines radiated power into a localized three-dimensional focal spot, making the analytical characterization of radial and angular resolution critical for sixth-generation (6G) multi-user systems, yet existing models treat the focal resolution as a fixed geometric outcome rather than a tunable design variable. Based on the Angular Representation (ASR) approach, this paper presents a resolution framework to design contiguous focal regions across both the radial and angular domain for steered multi-user systems. By extending the definition of a tunable inter-beam overlap $\rho$, we derive closed-form expressions for the radial and angular focal boundaries and extents, alongside the steering-dependent minimum focal distance $d_{0,\text{min}}$ and maximum steering angle $\theta_{\text{max}}$. Using these boundaries, iterative synthesis algorithms are developed to construct uniform and non-uniform radial and angular beam arrangements. Finally, by analyzing both isolated and joint radial-angular configurations, we quantify the fundamental trade-offs between $\rho$, achievable user capacity, inter-user interference, and system sum-rate across varying steering angles.

Video scene graph generation (VidSGG) aims to capture objects and their evolving relationships in videos. This requires accounting for both the persistence of object identities and abrupt changes in relationships. Existing methods aggregate information across frames to exploit temporal context, but an emphasis on consistency can obscure these changes. As a result, a model may retain its previous prediction even after the underlying relation has changed. We refer to this failure mode as relational semantic inertia. To address this issue, we propose Relation-Transition-Aware Transformer (RTAformer), a one-stage framework that separately models entity persistence and relation changes. It injects calibrated differences between adjacent-frame relation queries to guide predicate updates, while cross-frame instance alignment during training helps reduce cross-frame representational drift. Experiments on Action Genome show improvements in recall and mean recall over existing methods, with ablation studies supporting the contributions of the proposed components and comparisons using matched backbones showing larger gains on relation-transition frames than on non-transition frames.

Pairwise preference feedback has become a widely adopted supervision signal for reinforcement learning in video generation models, as pairwise comparisons provide a more reliable alternative than pointwise scalar ratings for evaluating the quality of generated videos. Specifically, we define a preference graph for each GRPO group under a pairwise evaluator, where nodes represent sampled rollouts and edges encode pairwise preferences. Auditing 6,976 on-policy comparison graphs, we find that cyclic preferences are prevalent and persist throughout training. Naive pairwise reward aggregation based on average win rates discards the underlying comparison structures and treats each rollout independently, which may introduce unreliable advantage estimates for certain rollouts and consequently affect reinforcement learning optimization. Therefore, we propose the Reliability-Anchored Preference Walk (\ours{}), a graph-aware advantage aggregation method that strengthens the global preference signal from win rates with reliable relational evidence from cyclic preference graphs through a reliability-guided preference walk. Under matched evaluator, comparison protocol, group size, prompt schedule, and query budget, \ours{} requires no additional evaluator queries beyond the observed comparisons within each rollout group and improves over Pairwise-GRPO across controlled comparisons, transfer settings, and blind human evaluations.

We introduce binary cyclic gradient coding (BCGC), an exact gradient coding scheme for $n$ workers and up to $s$ stragglers. Its encoding matrix is binary, circulant when $(s+1)\mid n$ and nearly circulant otherwise, and the master decodes by a single summation. Like the fractional repetition coding scheme (FRC) and its generalized version (GFRC), BCGC splits the workers into $s+1$ classes that each cover the whole dataset, but it assigns the data through cyclic supports. Its encoding matrix has rank $n-s$, and $n-s-1$ workers keep the balanced load $s+1$, which is the largest possible number. We prove that encoding matrix of BCGC is not permutation-equivalent to those of FRC and GFRC schemes. Moreover, We also prove that a binary coded scheme with compression is equivalent to an uncoded compressed scheme on a coarser partition of the data. This explains why binary schemes remain stable under compression, whereas the large decoding coefficients of MDS codes amplify the compression error and lead to divergence in our experiments. For every contractive compressor, the standard convergence guarantee of compressed gradient descent holds on each coarse partition. FRC, GFRC and BCGC therefore share the same convergence guarantee and differ only in their partitions, with one partition for FRC, two for GFRC and $s+1$ distinct partitions for BCGC. Choosing among these partitions can recover the exact gradient in cases where FRC and GFRC cannot. Experiments on four datasets confirm the theory. BCGC matches GFRC, keeps its accuracy under top $\kappa$ sparsification while the MDS codes diverge, and on heterogeneous data its partition diversity reduces the compression error and improves the accuracy.

Synthetic data is increasingly used in language-model pretraining. Existing knowledge-corpus synthesis pipelines, including WRAP and Nemotron-CC, usually take a source passage or document segment as input and generate rewritten or QA-style samples from that text. This can improve surface form and prompt diversity, but it does not explicitly require fine-grained source facts – especially secondary entities, list/table entries, and precise temporal attributes – to appear in the synthesized corpus. We propose AFCS (Atomic-Fact-Conditioned Synthesis), a pipeline that first extracts source-supported atomic facts and then uses each retained fact to drive generation. AFCS uses structure-aware extraction over the document tree, source-grounded verification, local context expansion, and multi-format generation. In continued-pretraining experiments with OLMo-3-7B and a 32B base on the same FineWiki source corpus, AFCS outperforms WRAP and Nemotron-CC on SimpleQA, SimpleQA Verified, PopQA, and HotpotQA. Granularity ablations show that fact-level synthesis improves over chunk-level, paragraph-level, and entity-level synthesis; and a 100B-token mid-training integration improves factual benchmarks and raises the 12-task general benchmark average by +1.09.