Papers

Dynamic contrast enhanced breast MRI (DCE-MRI) is a powerful clinical tool for breast cancer detection, providing high resolution anatomical detail together with rich temporal contrast information. However, high dimensional 4D inputs, small lesions, and heterogeneous acquisition protocols across clinical sites hinder robust automated classification of healthy, benign, and malignant cases. To address these challenges, we propose a framework that dynamically analyzes latent representations to adapt to protocol-specific characteristics. Spatial variability is mitigated by reducing confounding background uptake and compensating for misalignment caused by deformable soft tissue. Additionally, relationships in the latent space across phases are leveraged to select the most informative temporal features, improving robustness to protocol-specific temporal variability. Finally, task specific discriminative features are promoted through large scale supervised lesion segmentation pretraining, which substantially enhances downstream finetuning. Evaluated under leave-one-center-out validation on the ODELIA dataset and the held-out AMBL cohort, the proposed framework substantially outperforms finetuned radiology foundation models and prior methods, improving mean AUROC by nearly 8 points and balanced accuracy by 4 points over the strongest baseline. Additionally, our method achieved first place in the MICCAI ODELIA Breast MRI Challenge 2025, further demonstrating its effectiveness for robust breast cancer classification. We publicly release our codebase under https://github.com/MIC-DKFZ/CURIAtor.

We propose a paradigm for quantum enhanced axion dark matter search, which does not rely on power measurements. We propose to measure directly the axion amplitude and phase in an interferometric protocol at the quantum limit, using a non-linear cavity. In addition, we introduce gyromagnetic modes as wide mass range transducers for axion signals compatible with standard haloscope designs. We expect this scheme to offer an improvement of at least 4 orders of magnitude in figure of merit and at least 2 orders of magnitude in mass window with respect to standard haloscopes. Owing to its generality, our proposed protocol has the potential to speed up axion search but also the search for dark photons or other cosmological objects, such as galactic masers.

Motivated by recent developments in generalized Kähler geometry, we study morphisms of double Lie groupoids from a simplicial viewpoint. We show that the codiagonal functor $\Wbar$ extends from objects to horizontal and vertical principal bibundles and to square-shaped $(1,1)$-morphisms. The latter determine 2-dimensional morphisms of square and globular shape between anafunctors of Lie 2-groupoids. We illustrate these constructions through two applications: First, we revisit the equivalence between two models of nonabelian gerbes: groupoid bundle gerbes and principal 2-bundles. We show that a principal 2-bundle and its associated groupoid bundle gerbe determine equivalent anafunctors from a manifold to the corresponding structure Lie 2-group. Second, we refine the integration of Manin triples in transitive Courant algebroids. We show that integrations associated with different choices of suitably transverse Manin triples in the same Courant algebroid are related by symplectic Morita equivalences. Consequently, the resulting integration of the background Courant algebroid is well defined up to symplectic Morita equivalence.

Convolutional Neural Networks (CNNs) are used as analytical tools to investigate the relationship between the progress variable field and the local chemical source term in lean premixed hydrogen flames. Rather than employing machine learning for modelling, CNNs are leveraged to analyse the physical information contained in spatially resolved fields from direct numerical simulations. CNNs are particularly well suited for this task because they exploit spatial correlations and multi-scale structures, allowing an assessment of how spatial organisation influences the source term. The analysis demonstrates that the progress variable based on water, $C_{\rm H_2O}$, contains all the information required to accurately parametrise the chemical source term whereas $C_{\rm H_2}$ alone does not. The inclusion of the mixture fraction $Z$ improves the accuracy of the latter but provides no significant additional information to the CNN when the spatial field of $C_{\rm H2O}$ is used as input. The same behaviour is observed in both a laminar thermodiffusively unstable flame and a turbulent slot-jet hydrogen flame, indicating that $C_{\rm H_2O}$ is a robust single variable for parametrisation. A complementary scale analysis in the laminar case shows that spatial features extending over at least two laminar flame thicknesses are required to reconstruct the source term accurately, thereby identifying the characteristic scale in the progress variable field that carries this information. These results demonstrate how machine learning can uncover physical dependencies that remain hidden to classical statistical analyses.

Text-conditioned human-object interaction (HOI) generation requires body motion, object trajectories & rotations, and hand articulation to remain coordinated. These components differ in scale and dynamics, but must agree on contact, relative pose, and timing. A shared representation may limit the distinct structure of each stream, while independent generation prevents each stream from responding to changes in the others. Latent supervision alone also does not directly constrain contact after decoding. We propose TRACE, a continuous latent framework that keeps stream states separate and couples their updates. TRACE encodes body, object, and hand motion into separate latents and predicts each stream velocity from the complete current interaction state. Geometric losses on decoded motion further constrain contact and object-relative motion over time. The same model supports completion of any single absent stream from the other two. Frozen flow features also serve as input to a language model for HOI understanding. Experiments on InterAct, OMOMO, and BEHAVE show that joint completion training improves generation and that frozen flow features improve understanding over raw-motion encoding. On InterAct, TRACE achieves the highest contact precision, recall, and F1 among the compared methods.

Understanding the safety risks of vision-language-action (VLA) models is essential for their deployment in the physical world. Existing safety research has mainly considered persistent perturbations that are applied continuously to observations throughout an episode. However, momentary observation corruption, in which observations are severely perturbed only briefly within an episode, remains an underexplored safety threat. To address this gap, this work investigates robustness to one-step perturbations applied at a single time step per episode. Our experiments reveal that these perturbations substantially degrade VLA performance and that their impact depends on the action chunk execution length. Based on them, we propose CARE, which dynamically selects the execution length based on consistency with the previously predicted action chunk. CARE improves robustness with low computational overhead while preserving clean performance by selecting shorter execution lengths only under perturbations.

The theory of polyptych lattices seeks to incorporate, into a single combinatorial framework, the piecewise-linear bijections (mutations) that appear in many theories that generalize toric geometry, such as Newton-Okounkov bodies, toric degenerations, cluster varieties, and other areas. The link to algebraic geometry comes from the accompanying notion of detropicalizations of polyptych lattices, and their associated compactifications. The main result of this manuscript gives a concrete construction of a strict dual pair $(\mathcal{E}, \mathcal{F})$ of polyptych lattices, which we call a Gorenstein PL cone extension. To define our construction, we introduce the notion of Gorenstein PL cones, which are a polyptych analogue of a Gorenstein cone in the classical setting. If the original polyptych lattices are detropicalizable, then the new polyptych lattices $\mathcal{E},\mathcal{F}$ are also detropicalizable, via a simple explicit formula. Our Gorenstein PL cone extensions give a rich source of examples of strict dual pairs of polyptych lattices. In particular, any pair of Gorenstein-Fano polytopes $\Delta,\Delta'$ of dimension $n,n'$ respectively, gives rise to a strict dual pair $(\mathcal{E}(\Delta,\Delta'), \mathcal{F}(\Delta,\Delta'))$ of detropicalizable polyptych lattices of rank $n+n'+1$. Other examples can be built from cluster data.

We propose an integrated communication and computing (ICC) architecture that repurposes the inactive antennas of a spatial index modulation (IM) transmitter to perform over-the-air computation (AirComp). While IM delivers excellent spectral efficiency by selectively activating a subset of transmit antennas, the unused antennas are conventionally completely deactivated, forgoing spatial degrees of freedom (DoF) that can be exploited without additional bandwidth. By instead transmitting a pre-equalized, low-power computing stream over these idle antennas, the proposed architecture achieves simultaneous data transmission and computation without requiring orthogonal frequency or time resources. Alignment is performed entirely at the transmitters, such that the pre-equalization occurs leveraging local channel state information (CSI), while at the receiver, a vector Gaussian belief propagation (VGaBP) detector recovers the data payload under the discrete IM codebook constraint before the target function is estimated from the residual. Numerical results against exact maximum likelihood (ML) baselines confirm near-optimal detection under the adopted statistical model, and reveal that the number of antennas assigned to data transmission governs a direct trade-off between modulation robustness and AirComp accuracy.

Large language models are commonly evaluated by generating a small number of stochastic responses for each prompt in a benchmark. Because inference budgets are limited, this shallow evaluation may fail to observe low-probability but operationally important failures. A prompt that produces no failures in a small sample may therefore appear reliable despite having a nonzero latent probability of failure under repeated inference. We formulate LLM reliability evaluation as a budget-constrained discovery problem in which each prompt is associated with an unknown per-generation failure probability. We propose a budgeted discovery framework that first performs shallow evaluation across the prompt set and then uses trial-level failure outcomes together with prompt-derived representations to learn a feature-based ranking of failure propensity. The resulting scores prioritize prompts with zero observed shallow failures for deeper evaluation, concentrating the deep-evaluation budget where hidden failures are more likely to be discovered. We evaluate this approach on AIRBench and StrongREJECT across multiple model and system-prompt conditions. The central empirical test asks whether models fit without access to deep-evaluation outcomes can rank prompts with zero observed shallow failures according to their likelihood of producing failures under deeper evaluation. On AIRBench, the highest-ranked 10% of unresolved prompts achieves 2.54x hidden-failure lift for Qwen 2.5 7B and 1.87x for Gemma 3n E4B, recovering 25.4% and 18.7% of subsequently observed hidden failures, respectively, compared with 10% expected under random allocation. Semantic-neighborhood and feature-ablation analyses further show that this predictive signal can be recovered from multiple representations of prompt content and relationships.

Large language models can generate several records within one autoregressive completion, making earlier answers available as context for later records. This paper shows that such shared-completion batching can distort relationships among variables in the resulting synthetic data, using controlled tests on synthetic survey respondents. In a matched experiment on 2,000 European Social Survey profiles, generating ten rather than one respondent per request increases mean absolute error in within-country correlations by 48-58% for Qwen3.8-27B and 114-127% for Llama-3.3-70B-Instruct across three seeds, holding profiles, examples, questions and decoding parameters fixed. The distortion primarily reflects exaggerated relationship strength, while retaining substantial agreement with the human ordering of correlations. Controlled interventions establish answer history as a causal channel: re-pairing the same preceding values, with profiles and marginal distributions fixed, changes correlations among subsequently generated responses. Hiding preceding answers reduces correlation error in the tested settings but worsens marginal accuracy. Exploratory corrections across social-attitude, health and economic data likewise show that lower correlation error can coexist with worse marginal distributions and regression estimates. Request construction is therefore part of the data-generating process, and synthetic-data validity must be evaluated against the analyses the generated data are intended to support.

In this paper, we prove the existence of an envy-free up to one item (EF1) division for a discrete chore. Our approach builds on the powerful framework of Simmons-Su, which leverages Sperner's lemma to guarantee the existence of a simplex corresponding to a sequence of similar fractional divisions, ensuring that each agent is satisfied with a different bundle. Bilò et al. [2022] introduced a rounding technique that converts the fractional divisions into a connected integral EF1 division for goods when there are at most four agents, and this method was later extended by Igarashi [2023] to accommodate any number of agents. However, these rounding techniques for goods do not directly apply to chores because the definitions of EF1 differ in the two settings. To overcome this asymmetry, we modify the existing rounding techniques and show that connected EF1 divisions exist for a discrete chore.

In this paper, we use Pokémon Battles as a case study to investigate how to improve the performance of LLM-based agents in tasks that require opponent-aware planning without additional fine-tuning. We propose Long/Short-Term Knowledge Augmented Generation (LSTKAG), a mechanism that enables LLM-based agents to leverage past states of the current task and retrieve experience summaries from similar previous task instances based on the current state. Based on LSTKAG, we design Porimon, an LLM-based agent structure for Pokémon Battles. For optimization, we introduce an external API for precise damage calculation and more detailed information about the game. We conduct tournament-like evaluation experiments comprising 15,000 battles for hyperparameter optimization, ablation studies, and performance evaluation. The results indicate that Porimon-based players with hyperparameter optimization significantly outperform players based on PokéLLMon, an LLM-based agent structure proposed in previous research, and the rule-based heuristic player. Furthermore, our ablation study shows that Porimon variants outperform the one without extension in game information retrieval, which shows the contribution of that extension. However, the current experiment results are inconclusive regarding the contribution of Long-Term KAG. These results suggest that introducing external resources, information from previous states of the current task, and experience summaries from similar previous task instances could elevate the performance of LLM-based agents designed for tasks requiring opponent-aware planning.

In the nonlinear source of the pressure Poisson equation the fluctuating enstrophy and strain nearly cancel. We ask where the cancellation is complete, which we call silent, and how the source escapes it. Pointwise the source depends only on the eigenvalues of the velocity gradient. Pure shear has none, so it is silent. It is the origin of the normalised invariant plane, whose second-invariant coordinate is the normalised source. Let $f$ be the shear fraction of the gradient and $m$ its distance from that origin. We prove that $m\le1-f\le3m$ at every point of every incompressible flow. Wherever the source nearly vanishes and the third invariant is small, the gradient is nearly pure shear. Otherwise the cancelling gradient is a balanced vortex, a swirl its own strain cancels at any strength. At a no-slip wall the silent object is exact: the wall shear-stress fluctuation, a rank-one sheet. Away from the wall we assume the same sheet, dressed with a small residual, and derive how it escapes. The escape passes through one element of the residual, the along-sheet variation of the sheet-normal velocity. Its amplitude is the geometric mean of sheet and residual, and its sign selects swirl or strain. A rank-two silent object has the same source through the same gates but balanced eigenvalues, so the assumption fixes the escape's topology, not its source. The shear fraction is thus a pointwise proxy for cancellation by shear, and a model of the source must be built on what escapes.

We prove a lifting theorem from two-party public-coin one-way communication to multiparty public-coin one-way number-on-forehead (NOF) communication. For every fixed $k\ge2$ and prime $q>2k$, there is a generalized inner product gadget $\GIP_{q,r}^k:(\F_q^r)^k\to\F_q$ with $r=O_k(q/\log q)$ such that, for every partial Boolean function $f:D\to\bits$, where $D\subseteq\F_q\times\F_q$, \[ R_{1/3}^1(f)-O(1) \le R_{1/6}^{1,\NOF}\bigl(f\circ\GIP_{q,r}^k\bigr) \le R_{1/6}^1(f). \] Thus, composition with the gadget preserves one-way randomized communication complexity up to an additive constant and a change in the error parameter. The lower bound holds in the general one-way NOF model, where the last player sees the entire gadget input. This extends the deterministic one-way NOF lifting theorem of Yang and Zhang to randomized protocols, and extends the randomized lifting result of Wang and Wu from the conservative model to the general one-way NOF model. Our proof introduces a one-way cylinder partition bound that lower bounds public-coin one-way NOF communication complexity. We show that, for the lifted function, this bound is at least half the one-way partition bound of the outer function. The main technical step transfers a dual solution between the two bounds, using Möbius inversion and a discrepancy estimate for generalized inner product to control the loss. Combining this transfer with the characterization of two-party one-way randomized communication complexity by the one-way partition bound yields the lifting theorem.

Space-time-varying electromagnetic structures unlock wave-control mechanisms inaccessible to static media, including frequency conversion, harmonic generation, and joint control over frequency and momentum. Here, we introduce an inverse-design framework for multilayer space-time-periodic structures in which the spatial permittivity distribution and temporal modulation profile are optimized jointly. The proposed platform comprises multiple pixelated layers that are periodic along one or both transverse directions and harmonically modulated in time, enabling the direct synthesis of complex multichannel frequency-momentum scattering responses. We demonstrate two representative functionalities: independent control over the amplitudes and phases of three transmitted spatiotemporal channels and the generation of a frequency-momentum comb. Across these examples, we consider both continuous and binary permittivity distributions. To facilitate reproducibility and further development, we make the numerical implementation of the optimization framework freely available. These results demonstrate that inverse-designed time-varying multilayers provide a compact and versatile route toward programmable frequency-momentum multiplexing and the synthesis of space-time wave packets. The optimization code is made openly available to facilitate reproducibility and further development of the proposed inverse-design framework.

Few-step autoregressive video diffusion generates a long video by splitting the video into temporal chunks and generating chunk-by-chunk, each through a short sequence of denoising stages. To memorize chunks that are already generated, previous methods reconstruct a clean or less-noisy key–value (KV) cache by additional forwards to build the cache without advancing an output latent. However, every denoising forward itself already computes the in-flight KV of the current chunk. We introduce FlashForward, which directly reuses this cache to avoid the heavy cache-update-only model forwards. After the current chunk completes one denoising stage, its stage-specific cache is already available for the next chunk. Assigning one GPU to each stage therefore lets different chunks occupy different stages concurrently. This early availability has a quality cost: the resulting stage-matched history is noisy, causing appearance and motion drift among chunks. To complement it, FlashForward produces sparse auxiliary clean anchor latents before the corresponding region is generated so the generation trajectories can be stabilized by this two-sided conditioning. The two memories operate at different temporal scales: sparse clean anchor KV supplies coarse, long-range two-sided structural guidance, while dense stage-matched history preserves fine, recent evolution. With up to four GPUs, FlashForward runs $1.16$–$1.69\times$ faster than HiAR and $1.42$–$2.92\times$ faster than Self-Forcing for 16 FPS videos of 20 seconds or longer across 1.3B and 14B backbone scales at 480p and 720p. On VBench, for the 1.3B model at 480p, it achieves higher scores and remains stable at longer durations, demonstrating that FlashForward generates high-quality and temporally consistent videos across durations of 20s, 35s and 65s at a much faster generation speed.

Deep learning models can achieve high accuracy for indoor localization, but their black-box nature limits interpretability and the reuse of learned information. We propose a hierarchical deep learning framework for WiFi fingerprint-based indoor localization that jointly predicts user location and learns an effective geometry of the surrounding access points (APs). Physics-informed decoders infer this geometry directly from RSSI measurements and labelled user positions, without requiring the true AP coordinates during training. The learned geometry is then used to rank and prune APs. On the UJIIndoorLoc dataset, the proposed chained model achieves a mean 3D localization error of 7.07 m, reducing error by 26% to 36% compared with baseline models. Previously published methods evaluated on the same official split report errors 10.6% to 31.0% higher. Pruning 35% or 50% of the APs causes only a small loss in localization accuracy. The inferred geometry also enables Fisher-information-based AP ranking even when fingerprint databases do not contain surveyed AP coordinates. Experiments on the Tampere/TUT and UTSIndoorLoc datasets show that geometry-guided AP selection performs comparably to selectors built directly from labelled data. These results show that physics-informed interpretability can improve indoor localization while also supporting effective feature selection.

This paper addresses fundamental questions about the existence and structure of isotropic and anisotropic spherical random fields and the properties of composite transformations of isotropic fields. It introduces a hierarchical structure comprising several classes of anisotropic random fields, using geometric and spectral approaches, and provides a geometric–spectral classification of second-order anisotropy on $\mathbb{S}^2$. We characterize feasible isotropic and anisotropic scenarios and identify those that, in principle, cannot occur, in contrast to the case of random fields in Euclidean space. Furthermore, we establish connections between geometric invariance properties and the structure of the spectral covariance matrix of spherical random fields. The obtained results lay the foundations for applications in statistics and data analysis, providing guidance on feasible models, their spectral representations, and appropriate sampling schemes for parameter estimation or anisotropy hypothesis testing for spherical data.

Spiking-Mamba hybrids reach strong accuracy on event-based vision, but existing designs often require tens of millions of parameters. Much of that cost comes from how the spiking front-end is connected to the state-space backbone rather than from the hybrid architecture itself. In a representative model, a single resolution-dependent projection accounts for 33.55M of 36.25M parameters. To that end, we introduce RIPE-MambaSpike (Resolution-Independent, Parameter-Efficient), which replaces that projection with a hierarchical multi-resolution bridge of fixed channel width. Its deployed footprint is 0.870M parameters, constant at fixed time steps and widths across a 43x range of input areas. Reparameterized spiking stages, temporal decoupled modulation, and a dynamic convex-hull-bounded dual-stream membrane-potential attention preserve accuracy under this compact design. Result-wise, RIPE-MambaSpike is pareto-optimal on CIFAR10-DVS, N-Caltech101, and DailyDVS-200. Notably, on the 200-class DailyDVS-200, a scaled 8.04M configuration achieves 45.7% top-1 accuracy, the best reported spiking result on that benchmark, and outperforms prior spiking methods with 3.0-15.1x fewer parameters than dense ANNs. Overall, our findings demonstrate that competitive event-based recognition does not require resolution-dependent parameter growth. Code is available at https://github.com/MuhiminOsim/RIPE-MambaSpike.

Audio language models can recognize spoken commands and invoke tools, but an agent must first decide whether the acoustic and conversational context warrants action. We introduce VGBench, a 1,018-item diagnostic benchmark for action-level addressedness across side-talk, self-talk, and speaker-switch scenarios. Each item uses a shared action space comprising silence, a tool call, and a natural-language answer. Speaker-switch pairs hold the specified words fixed while source, distance rendering, and a temporal boundary define a controlled wearer-to-bystander shift. Six raw Audio LLMs and three training-free adaptations often identify the target tool yet rarely withhold action under this shift; the highest raw switch mute rate is 14%. We then use VoxGate as a post-training case study. Supervised training mutes 91.3% of switched commands while choosing the correct tool for all nearby wearer commands and text-only controls. An exploratory GRPO stage has similar switch performance; side-talk accuracy rises from 68.4% to 70.9%, and self-talk muting from 52.0% to 60.0%. Factorized controls identify an independent source-change effect, while sensitivity to the far-field manipulation varies across acoustic renderings. The benchmark therefore measures multi-cue acoustic-context gating rather than isolated speaker identity.

In this note, we consider the advection-diffusion equation on the $d$-dimensional torus, where $d$ is arbitrary. First, for the associated transport equation, we provide an explicit example of a space-time smooth, divergence-free, random velocity field that is a universal exponential mixer. Second, we set the molecular diffusivity to a (small) positive value, and we show that the same velocity field is dissipation enhancing with optimal rate. Our construction is based on the Random Dynamical Systems viewpoint, and relies on the uniform ergodicity of the two-point Markov process associated with the two-dimensional Pierrehumbert model.

Depth is a natural way to increase the computational capacity in Transformers, yet the contribution of deeper layers can diminish as depth grows larger. Recent approaches enhance normalization (\text{e.g.}, LayerNorm Scaling) or residual connections (\text{e.g.}, mHC, AttnRes) to enable better information flow and depth utilization. However, it remains unclear whether they truly translate increased architectural depth into effective computational depth, and whether their reported gains stem from better access to information across depth, or unaccounted-for confounding factors. In this paper, we introduce DepthBench, a controlled benchmark for studying computational depth across various architectures. We systematically vary the width–depth aspect ratio ($d_{\text{model}}/n_{\text{layer}}$) from shallow–wide to deep–narrow shapes, while keeping the model size and pre-training recipe fixed. Across 10 representative architectures, we find that the benefit of allocating more capacity to depth is strongly architecture-dependent. Standard Pre-LN and most of its norm- and scaling-based variants provide little benefit and can even degrade performance as models become deeper and narrower, whereas HC and Full AttnRes improve consistently even at extreme deep shapes. These gains extend beyond pre-training loss and consistently translate into improved domain-specific performance and effective computation. Controlled layer-level analyses further show that the gains of HC and Full AttnRes are associated with more effective utilization of additional layers, revealing distinct mechanisms of computational depth across architectures. Overall, our results identify residual connection design as a key determinant of whether depth can serve as a meaningful scaling axis by enabling additional architectural depth to translate into effective computation.

Inserting advertisements (ads) into consumer-facing LLM output is emerging as a new business model, but there is little shared evidence on how such ad insertion should be evaluated or how it affects user preferences. We introduce LLMAdBench, a human-preference benchmark for studying advertising in LLM-generated content. The benchmark isolates a simple but practically important decision: given a user conversation, an LLM response, and a matched advertisement, where should the ad be placed? Our dataset compares pairs of responses that differ only in ad position while holding all other conditions fixed including the user query, base answer, advertisement, and disclosure condition. Human annotators evaluate each pair based on six criteria from both advertiser's and user's perspectives. The resulting benchmark contains more than 18000 human judgments across two disclosure conditions: explicitly labeling the ad as sponsored and merging it into the response without disclosure. We use LLMAdBench to evaluate eight frontier LLMs as preference judges and find that they are not reliable substitutes for human evaluation. Even the most stable models reverse roughly one quarter of their decisions when the presentation order is swapped, agreement across models is low, and their placement preferences differ systematically from those of human annotators. Moreover, LLMAdBench contains substantial learnable signal. In particular, a Qwen3-8B model fine-tuned on the human preferences improves substantially over its base model and outperforms all zero-shot frontier judges on the held-out prediction task. Beyond model evaluation, LLMAdBench provides quantitative evidence on the advertiser-user trade-off and shows that the sponsorship disclosure systematically changes users' preference over ad placement.

A generalization gap in reinforcement learning, return on training levels minus return on held-out levels, is usually reported without a reference point. We argue that it should be read against a measured random floor: the return of a uniform-random policy on the same levels under the same evaluation harness. On eight ProcGen environments with PPO at a compute-limited budget (8M steps, 16 parallel environments; three games extended to 25M), the floor changes what standard numbers mean. The test-time action rule decides which policy is measured: in miner, the sampled policy scores 5.1x the floor on held-out levels while its argmax scores below it in every run, and greedy evaluation places two environments significantly below the floor. Used as a convergence diagnostic, raw policy entropy flags six of eight environments, but 32-66% of that entropy lies on actions with identical effects; against the floor, five of eight sampled policies are clearly above it on held-out levels and heist's is not distinguishable from it. An audit of twelve ProcGen codebases finds that nine sample test-time actions with no explicit choice at the evaluation call site. We recommend that every reported gap state its action rule, seed its evaluation and specify its tests before analysis, and report the floor on both level sets.

Charged lepton flavor violation (CLFV) is forbidden in the Standard Model (SM) due to massless neutrinos and absence of right-handed neutrinos. The discovery of neutrino oscillations established that the neutrinos are massive and the lepton flavor is not an exact symmetry of nature. A minimal extension of the SM that incorporates right-handed neutrinos gives rise to CLFV through loop-level diagrams involving neutrinos. However, such processes are extremely rare, for instance, the branching fraction of $\mu \to e\gamma$ is expected to be at the level of $10^{-54}$, far beyond the sensitivity of current collider experiments. In contrast, several new physics models predict CLFV decay rates within the sensitivity of current experiments, such as the Belle and Belle II experiments, which have collected the world's largest asymmetric $e^+e^-$ collision data near the $\Upsilon(4S)$ resonance. The large data sample has utilized to explore the possibility of CLFV decays of $\tau$ leptons, $B$ mesons, and bottomonium states. The report reviews some latest results on these CLFV decays at Belle and Belle II.

Difference-in-means steering requires activations recorded while a model shows the desired behavior, which a sandbagging model withholds by deliberately underperforming. We introduce Activation Flow (ActFlow), which manufactures these activations from $k$ correct labels without fine-tuning. ActFlow sets target logits that rank each labeled item's correct answer first, and moves the logits toward them by adding one vector $x$ to all $k$ residual streams at one layer. ActFlow is a family of ordinary differential equations for $x$, one for each rule that maps the required logit change to the velocity of $x$. The smallest-norm rule lands exactly on the targets, while the others keep only the top singular directions of the Jacobian. We test ActFlow on three instruction-tuned models, each locked by a sandbagging prompt and by a password-locked LoRA. At $k=40$, ActFlow keeping five singular directions raises the mean held-out ARC-Easy accuracy over the six locked models from $0.05$ to $0.85$, against $0.88$ for rank-$16$ LoRA fine-tuning, which trains about $3{,}000$ times as many parameters, and $0.92$ for the honest models. Furthermore, it scores higher than the smallest-norm rule in 16 of the 18 combinations of locked model and $k$, and its steering direction is nearly orthogonal to the honest difference-in-means direction. It also unlocks two LoRA locks where the honest direction fails. Surprisingly, with one labeled item, the single-step smallest-norm rule raises the Qwen2.5-7B prompt lock from $0.04$ to $0.59$.

Open-prompt detectors allow users to specify targets with text, visual exemplars, or both. We argue that existing designs underuse the visual modality in two ways. First, multiple exemplars are commonly compressed into a single class-level embedding. This textualizes visual prompting: the resulting vector plays the role of another class name, is often aligned to or injected into the text pathway, and may be suboptimal when only a few heterogeneous exemplars are available. Second, existing methods interact primarily in prompt or representation space, before modality-specific detection states are formed. We address both issues from a set perspective: \setopd preserves modality-specific decoding states from a shared prompt-conditioned initialization and performs explicit multimodal collaboration at the candidate-state level. For the first issue, we introduce \br prompting, which reads every boxed exemplar in its full scene context and decomposes the pooled support evidence into a Base anchor and a learned Residual correction; the resulting prompt has fixed capacity regardless of the number of examples and drives its own visual detection pathway. For the second, we recast text–visual collaboration from representation-level fusion into a candidate-set modeling problem. Paired-Query Arbitration (\pqa) then performs explicit cross-modal state arbitration only after modality-specific candidate states have been formed. The two readers share query initialization so their candidates are paired by index; a learned gate arbitrates within each pair, followed by a permutation-equivariant module that reasons over the fused set.

Large language models (LLMs) have enabled data-science agents to automate multi-step analyses over heterogeneous files. However, incorrect choices regarding data sources, scope, or statistical definitions often lead to silent errors: computations execute successfully but produce plausible yet incorrect outputs that fail to answer the intended question. To mitigate this, we present RADAR, an auditable runtime that makes an agent's analytical choices inspectable and supports their revision through execution feedback. RADAR operates through three core mechanisms. First, an evidence-preserving exploration module retrieves task-relevant content while retaining source locations and observation coverage. Next, the runtime uses typed operators to record the agent's declared inputs, operation arguments, and resulting observations. Finally, runtime validation checks proposed operations against these observations. When a conflict is detected, the runtime rejects the operation or provides diagnostic feedback, allowing the agent to revise its choices before errors propagate. This design enables agents to fail loudly while leaving semantic interpretation to the LLM. On KramaBench, RADAR achieves overall scores of 0.723 with full source retrieval and 0.747 with gold sources supplied, corresponding to relative gains of 35.9% and 28.8% over the strongest baselines. Beyond KramaBench, RADAR achieves relative performance gains of 14.0% on DA-Code and 59.3% on DABStep, demonstrating its applicability across diverse agentic data-analysis workflows.

Predicting cellular responses to genetic perturbations helps prioritize experiments in single-cell genomics, where exhaustive measurement is infeasible. While computational models increasingly predict these responses, three evaluation deficiencies obscure what their scores demonstrate. First, absolute metrics cannot separate target-specific predictions from a shared background response. Second, common metrics remain high under gene shuffling, so gene-level accuracy is never verified. Third, a score at one training size says nothing about coverage, which depends on representation-space proximity and response-constraining power. We propose AmbiModBench, a specificity-aware, gene-resolved and coverage-aware benchmark. It pairs every score with a training-mean reference fitted on the same split, screens each readout by gene-coordinate permutation, and links embedding distance to response variation. Across K562, RPE1 and Norman, strong absolute scores largely reflect shared background rather than target-specific learning. Widely used readouts track response magnitude distributions rather than the affected genes. Detectable gain follows representation-space coverage rather than training-set size. Nonetheless, on RPE1 the protocol yields a reproducible target-specific gain across five additional splits and three gene selections, which absolute scores alone cannot distinguish from shared background.

Let $m\geq1$ be fixed. We consider the space $\mathcal{P}_m(\mathbb{R}^n)$ of real $m$-homogeneous polynomials on $\mathbb{R}^n$, endowed with the standard Gaussian measure $\gamma_{m,n}$ associated with the Bombieri norm. We study how well the norm of a typical polynomial on the unit ball of $\ell_\infty^n$, namely the cube $[-1,1]^n$, can be recovered from its values at the vertices. For $P\in\mathcal{P}_m(\mathbb{R}^n)$, set \[ M(P)=\max_{x\in[-1,1]^n}|P(x)|, \qquad V(P)=\max_{\varepsilon\in\{-1,1\}^n}|P(\varepsilon)|. \] If $P_n$ is chosen according to $\gamma_{m,n}$, we prove that the relative loss \[ 1-\frac{V(P_n)}{M(P_n)} \] is of order at most $n^{-1/2}$ in probability. Consequently, for every $0<\beta<1/2$, \[ \gamma_{m,n}\left\{ P\in\mathcal{P}_m(\mathbb{R}^n): V(P)\geq (1-n^{-\beta})M(P) \right\} \longrightarrow1 \] as $n\to\infty$. Thus, with respect to the Bombieri Gaussian measure, the vertices of the cube are asymptotically norming for homogeneous polynomials of fixed degree.