Papers

We study a sequential pricing problem in which a seller jointly posts a price and an advertisement generated by a large language model (LLM). The seller aims to maximize revenue under an unknown product demand that depends on both decisions, while observing only whether each offer leads to a purchase. We propose an online actor-critic algorithm that combines low-rank adaptation (LoRA) of a pretrained LLM with a demand model fitted to available data. At each round, the actor generates an advertisement, and the critic estimates purchase probabilities to guide price selection. Then, the resulting feedback is used to update both the actor and the critic, with the critic's revenue estimates providing a baseline for policy gradient updates of the actor. To evaluate our approach, we develop an evaluation framework with three synthetic demand models and a demand simulator built from real-world marketplace data. Finally, we compare our algorithm with benchmarks that do not jointly optimize price selection and advertisement generation, achieving expected revenue gains over the reference policy of 5.69%, 5.18% and 55.96% under the three synthetic demand models and 5.81% under the marketplace simulator.

While binary classification is one of the most extensively studied problems in machine learning, the regime in which the goal is to learn a classifier with an almost zero false negative rate remains largely unexplored. In this paper, we introduce the Extreme Binary Classification problem, where the objective is to learn a classifier whose false negative rate $\alpha$ is constrained by $\epsilon_{N_1}=o_{N_1\to\infty}(1/N_1)$, with $N_1$ denoting the number of positive examples in the training set. To address this problem, we propose a threshold adaptation method theoretically grounded in guarantees derived from Extreme Value Theory, together with a feature selection procedure based on a permutation test applied to sample maxima. Experimental results on four real-world datasets of varying sizes demonstrate that our approach compares favorably with state-of-the-art methods. In addition, we illustrate its interpretability through an application to a cancer screening dataset.

Retrograde tracer injections in marmoset monkeys revealed a striking feature of neural circuitry in primate visual system: among neurons projecting to the primary visual cortex (striate cortex, V1), the proportion of extrastriate feedback neurons relative to thalamic feedforward neurons decreases markedly with visual field eccentricity. This supports the Central-Peripheral Dichotomy theory, which proposes fundamentally distinct computational specializations — visual recognition and attentional selection, respectively — in the central and peripheral visual fields.

We give a combinatorial criterion for detecting roots of Bernstein-Sato polynomials of ideals. As applications, we prove the strong monodromy conjecture for determinantal and Pfaffian ideals.

LLM routers pick a cheap or expensive model per request by its content, and many gateways and some cloud platforms can log that choice with content logging off. We measure this privacy channel beyond token counts, accounting for noisy labels and repeated prompts. We run pre-registered studies on 1.7 million real requests (WildChat-1M, LMSYS-Chat-1M) with two cost/quality routers and a domain router, survey eleven systems' logging, and test post-processing defenses. At matched length, the shift's direction depends on category and router. For RouteLLM at the 50% operating point, harassment and self-harm requests reach the strong model 19 points less often than comparable ones on prompts unseen in exploration, medical requests (exploratory: LLM labels failed their gate) 31 points less often on distinct prompts (both post hoc), and sexual requests 10 points more often (secondary); the other router's four are negative. Twenty RouteLLM decisions separate frequent medical askers with AUC 0.71, exploratory and below the pre-registered primary endpoint's 0.75 (domain router: 0.92, an upper estimate). Per-category length-matched parity with accurate labels removes the gap on real traffic, costing at most 0.2 accuracy points on RouterBench (post hoc), where routers' gaps on sensitive subjects (13-42 points, pre-registered) exceed those of an oracle routing by realized accuracy gain (1-11, post hoc). Per-conversation stickiness, per-user budget bands, and pooled parity fail, the last as categories' shifts differ in size or sign. A post hoc exact per-user rate hides only even-prefix strong counts and forfeits most self-assessed routing value; it preserves odd-position decisions, from which a post hoc log attack reaches AUC 0.73 after 20 RouteLLM requests (exploratory).

Modern artificial intelligence is predominantly executed on computing architectures optimized for dense linear algebra. While this has enabled the success of contemporary neural networks and motivated compute-in-memory (CIM) architectures, a growing class of artificial intelligence (AI) workloads depends on associative retrieval, identifying stored information by content or similarity rather than by explicit memory addresses. Such operations are central to transformer attention, tree-based inference, genomic search, and other retrieval-intensive applications, yet remain inefficiently supported by conventional memory systems. In this Perspective, we argue that content-addressable memories (CAMs) provide a complementary hardware primitive for associative processing in AI. We review their ability to perform massively parallel in-memory matching, discuss how emerging memory technologies can improve density and energy efficiency, and identify hierarchical search, application-specific architectures, hardware-aware learning, and heterogeneous integration with CIM as key directions for scalable associative computing. Together, these developments suggest that associative retrieval should complement linear algebra as a foundational computing primitive for future AI systems.

Building on the recent publication of a fresh catalog of blue straggler stars in Galactic open clusters using Gaia DR3, we conduct a study focusing on the blue straggler stars with short-period photometric variability found in the open cluster Collinder 261: nine eclipsing or ellipsoidal binaries and five SX Phe pulsator candidates. Our investigation delves into the potential of mass transfer in binary systems to account for the presence and properties of these stars. We first measure their photometric periods and classify the variables into eclipsing/ellipsoidal binaries, SX Phe pulsator candidates, and instrumental aliases or blends. Subsequently, we perform a set of simulations based on comprehensive calculations of the structure of both stars within the binary system, as well as their orbital evolution and mass transfer rates. To compare with observations, we solved the atmospheric structure of each stellar component and computed filter-weighted integrated fluxes in the Gaia DR3 passbands, obtaining a combined evolutionary sequence that accounts for the unresolved flux of both stars as seen photometrically. Our investigation includes the study of several parameters, such as the initial mass of both stars, the initial orbital period, and the efficiency of mass transfer. The aim is to identify potential progenitor candidates matching color-magnitude diagram observations at the same age as that determined for Collinder 261. We find that binary evolution serves as a viable mechanism for producing these eclipsing blue stragglers and that the efficiency of mass transfer plays a key role in accurately matching the fainter objects.

Bacterial co-infection, particularly with Streptococcus pneumoniae, is a major driver of morbidity during influenza epidemics, and this burden falls disproportionately on sub-Saharan Africa, where both pathogens circulate year-round at high levels and pneumococcal carriage remains common and persistent across the region. Yet existing mathematical models rarely track a host's bacterial carriage status jointly with their full influenza history through recovery. We develop a twelve-compartment ODE model that classifies each host by viral status (susceptible, infected, recovered) and bacterial status (susceptible, carrier, actively infected, recovered), allowing carrier-to-active progression to depend on prior influenza exposure. We establish positivity, boundedness, and a positively invariant region, then derive the basic reproduction number $\mathcal{R}_0$ and quasi-endemic equilibria of a single disease, together with invasion reproduction numbers $\mathbf{Inv}^{\mathbf{B}}$ and $\mathbf{Inv}^{\mathbf{V}}$ quantifying each pathogen's ability to invade a population where the other is already endemic. Center-manifold analysis shows that the bacterial subsystem and the full coexistence equilibrium undergo a backward bifurcation precisely when carrier transmission exceeds the carrier's total exit rate, $\beta_c>\kappa_S+\theta+\mu$, so that reducing $\mathcal{R}_0$ below one is not sufficient for elimination, while the viral subsystem bifurcates forward. A global sensitivity analysis (Latin Hypercube Sampling with partial rank correlation coefficients) shows that the viral and bacterial thresholds are governed by largely separate parameters, while the bacterial invasion potential is strongly shaped by viral transmission and recovery, giving a quantitative signature of cross-pathogen synergy.

Animal evidence shows that precise voluntary movements arise from rotational neural population dynamics in motor cortex, but their physical effects remain unknown. We developed a robotic analog of biological motor systems with artificial muscles, multimodal sensors, and a neural network controller trained via reinforcement learning. The robotic analog exhibited accurate movements, robustness to damage, and neural population dynamics akin to animals. This task-driven, embodied model illuminates the causal link between neural population dynamics and motor outcomes. We discovered that neural rotations generate oscillatory maneuvers orthogonal to the reaching direction, optimizing trajectory adjustments, which is confirmed by primate neural data. The model also revealed counterintuitive neural energy principles under sensor and motor redundancies, and striking Eureka moments during motor learning, bridging biological and artificial systems. These findings provide new perspectives on how neural dynamics contribute to accurate and flexible movement, inspiring future intelligent robots with animal-like mobility.

AI-generated text (AIGT) detection can be sensitive to the decoding choices of the source large language model (LLM). We observe that perturbing next-token logits or adjusting sampling temperature can reduce detection performance, providing a clear signal of detector vulnerability to decoding-time distribution changes. Building on this observation, we propose EASE (Entropy-Adaptive Distribution Shaping for Evasion), a training-free and detector-agnostic framework for evading AIGT detectors. EASE computes predictive entropy directly from the source LLM's next-token distribution and uses it to adapt both logit perturbation and sampling temperature, without detector feedback or model fine-tuning. Experiments across three source LLMs and multiple detectors demonstrate consistent reductions in detection performance, with negligible degradation in text quality and negligible inference overhead.

Prunescu, Sauras-Altuzarra, and Shunia asked whether addition, totalized integer division, and base-two exponentiation form a substitution basis for the Kalmár elementary functions. We answer this problem negatively, even after adjoining multiplication. On independent powers of two, every fixed term built from addition, multiplication, and integer division takes only polylogarithmically many distinct values below a given bound, and the same restriction survives arbitrary nesting of $2^x$. No such term recovers $n$ from $2^n$, so neither $\lfloor\log_2 x\rfloor$ nor integer remainder is representable. The obstruction is not special to the base two. For every fixed integer $b\geq 2$, the same sparsity bound holds with constants independent of $b$, and $\langle x+y,\lfloor x/y\rfloor,b^x\rangle$ remains a proper subclass of the elementary functions. Variable exponentiation reverses the conclusion: addition, integer division, and $x^y$ do form a substitution basis, $$ \langle x+y,\ \lfloor x/y\rfloor,\ x^y\rangle=\mathcal{E} .$$ Once the base may depend on the input, integer remainder and truncated subtraction become representable. The incomplete base-two class is nevertheless complete for truth values. Every elementary characteristic function belongs to it, as does every elementary function with a fixed finite range.

Electrically neutral spin-1/2 fermions can interact via Yukawa or gauge couplings with electrically charged particles, leading to quantum-induced couplings between the neutral fermion and the photon. In this paper, we present general expressions for the diagonal and transition electromagnetic moments of Dirac and Majorana spin-1/2 fermions generated at the one-loop level. We also briefly discuss the implications of these results for several neutral spin-1/2 fermions of phenomenological interest, such as neutral components of $SU(2)_L$ multiplets (which are potential candidates for dark matter) or Standard Model neutrinos.

When the harmfulness of an LLM agent's output can be quantified, a natural jailbreaking objective is to maximize expected harmfulness over admissible input modifications. An alternative approach constructs or selects harmful target outputs and modifies the input to increase their likelihood. We establish a precise connection between these two approaches through a probabilistic reformulation. Specifically, we show that the gradient of the logarithm of expected harmfulness with respect to the input equals the expected input gradient of the model's log-likelihood under a harmfulness reweighted output distribution. This identity provides a unified interpretation of expected harmfulness and target likelihood optimization. Building on this connection, we propose OPUR, a sampling distribution designed to generate highly harmful target outputs and use the resulting samples to guide likelihood-based input optimization. Experiments demonstrate the effectiveness of the resulting method in jailbreaking LLM agents.

We give a deterministic polynomial-time algorithm for the Bilu–Linial signing problem on bipartite graphs. For every finite simple bipartite graph of maximum degree at most an integer $\Delta\ge3$, the algorithm assigns signs to its edges so that the signed adjacency matrix has operator norm strictly less than $2\sqrt{\Delta-1}$. Our algorithm builds on the randomized recursive repair framework of Jadbabaie, Saberi, and Sra~\cite{JSS26}, with deterministic rules for sign selection and vertex deletion.

Dissipation is generally regarded as detrimental to nonclassical-state generation, yet whether it can determine which quantum resource is emitted from an open hybrid quantum system remains largely unexplored. Here, we show that the symmetry of bosonic dissipation channels can select and switch quantum-emission pathways in a tripartite Mollow system, where a strongly driven two-level emitter simultaneously couples to cavity photons and motional phonons. In the Mollow regime, higher-order tripartite interactions generate resonant photon-phonon pair excitations, whose emitted quantum states are subsequently selected by dissipation. Under symmetric dissipation, the system produces strongly correlated photon-phonon pairs, enabling both single-pair emission and correlated multipair bundles with pronounced nonclassical correlations. Breaking the dissipation symmetry progressively suppresses the hybrid correlations and dynamically decouples the photonic and phononic degrees of freedom. In the strongly asymmetric regime, long-lived phonons accumulate in highly excited Fock states, while rapidly leaking cavity photons are converted into high-purity single-photon and two-photon bundles. The crossover between these distinct emission regimes is characterized through time-resolved correlation functions, Fock-state distributions, and Wigner quasiprobability distributions. Our results establish dissipation symmetry as a resource for steering quantum emission between hybrid and purely photonic nonclassical states, providing a route toward programmable hybrid quantum emitters and dissipative quantum-state engineering.

Affine frequency division multiplexing (AFDM) has emerged as a promising waveform for high-mobility communications owing to its ability to exploit multipath diversity. However, channel state information (CSI) acquisition remains challenging in channels with fractional delay and Doppler and an unknown number of propagation paths. In this paper, we investigate the channel estimation for AFDM systems. We first analyze the AFDM response in the presence of fractional-delay-induced frequency wrapping. The analysis reveals that fractional delay may displace the dominant extremum and generate informative secondary extrema. Then, we develop a model-order-adaptive channel estimator within an improved space-alternating generalized expectation-maximization (SAGE) framework. A dual-pilot reference symbol is employed for path and delay initialization, while pilot observations across multiple AFDM symbols provide temporal information for Doppler estimation. New paths are identified through statistically controlled residual tests, whereas unsupported or redundant paths are removed by conditional support pruning. The delay, Doppler frequency, and complex gain of each retained path are subsequently estimated through a coarse-to-fine procedure and refined using the exact AFDM likelihood. Simulation results demonstrate that the proposed method achieves lower normalized mean squared error (NMSE) and bit error rate (BER) than representative SAGE, sparse-recovery, and Bayesian benchmarks. It also provides reliable path detection and model-order estimation and maintains robust performance at terminal velocities of up to 600~km/h.

Post-training quantization (PTQ) enables efficient deployment, yet transformer architectures remain challenging to quantize due to nonlinear layers. While existing methods attribute accuracy loss to insufficient numerical precision, often necessitating floating-point fallbacks, we demonstrate that degradation is actually driven by specific structural error sources. We find that learned scale parameters in normalization layers and compounded approximations in GELU are the primary error contributors, whereas SoftMax remains inherently robust to aggressive quantization. To address these bottlenecks, we introduce TR-PTQ, a unified integer-only formulation using shared Taylor Region (TR) exponential and logarithm primitives. This approach allows computationally expensive operations, including division and square roots, to be performed entirely in the log-domain via standard integer arithmetic. Combined with a calibration-free, outlier-aware optimization for LayerNorm parameters, our method eliminates the need for floating-point hardware units for nonlinearities, achieving less than 1.5% absolute accuracy degradation across vision and language benchmarks.

A perfect Italian dominating function assigns a label from $\{0,1,2\}$ to each vertex so that the labels in the neighborhood of every zero-labeled vertex sum to exactly two. Although the associated decision problem (PID) is NP-complete on chordal graphs, the complexity status on split graphs remains open. We give an $O(n^4)$ time algorithm for split graphs on $n$ vertices. We note that this is in contrast to Roman, Italian, and perfect Roman domination, which are NP-complete on split graphs. We further show that this tractability boundary is tight: PID is NP-hard on graphs at deletion distance one from split graphs. We also provide dichotomy results for PID on restricted extensions of split graphs formed by attaching pendant stars, and establish inapproximability results for the same.

Nano unmanned aerial vehicles (nano-UAVs) can navigate confined spaces that larger robots cannot, but their payload capacity severely limits the sensors and compute available for self-localization in global navigation satellite system (GNSS)-denied environments. We present a vision-based system that localizes a nano-UAV from a quadruped robot with an arm-mounted camera. The quadruped tracks the drone, estimates its position in its own coordinate system using segmentation masks and depth, and transmits that position over a real-time radio link. To ensure continuous tracking, we developed a perception-aware nonlinear Model Predictive Controller (NMPC) that dynamically adjusts the quadruped's body and arm to maximize the drone's visibility, treating observation reliability as a primary control objective. Relying solely on this external estimate, the nano-UAV executed predefined trajectories from takeoff to landing with a median 3D localization error of 59 mm. The system demonstrates robustness against short visual occlusions by smoothly transitioning to onboard inertial flight when line-of-sight is temporarily lost. Ultimately, this framework allows a quadruped to offload the localization burden of a nano-UAV, enabling inspection of complex spaces that neither robot could navigate alone.

Universal maps (maps whose renormalized iterations approximate every map in a given class) are locally generic in several spaces of diffeomorphisms [Bonatti–Díaz 2003, Turaev 2015]. In fluid dynamics, steady Euler flows whose Poincaré maps are universal are also known to be locally dense [Berger–Florio–PeraltaSalas, 2023]. Motivated by Arnold's vision [Arnold, 1966] that the complexity of orbits in celestial mechanics and that in fluids should be comparable, in the present paper we address the question of the existence of universal maps in celestial mechanics. We give a partial answer by proving a weak, finite-dimensional form of universal dynamics. More concretely, we show that every symplectic embedding of the disk into $\mathbb{R}^2$ can be approximated, with arbitrary precision, by a renormalization of a Poincaré map of a restricted planar circular $(n+1)$-body problem, for suitable $n$ and appropriate choice of the masses of the primaries. The proof relies on Gonchenko–Shilnikov–Turaev theory [Turaev 2003, Gonchenko–Turaev–Shilnikov 2007] and requires control of the dynamics near homoclinic tangencies of arbitrarily high order; the techniques developed in [Garrido–Martín–Paradela, 2025] are essential for such control.

We report the discovery of PALTA-1 b, a sub-Jupiter mass exoplanet candidate orbiting the nearby early-L dwarf DENIS-P J1253108-570924 (DE1253-57) located at a distance of ~17 pc. The exoplanet was discovered by monitoring the L dwarf's astrometric motion with FORS2 at the Very Large Telescope over more than eight years and characterising the deviations from the expected single-source motion. The estimated planet mass is between 0.54 MJup and 0.77 MJup, depending on the uncertain mass of the ultracool dwarf host. The system's mass ratio is about 0.010, and the relative separation is estimated at 1.6-3.4 au. If confirmed, PALTA-1 b would be one of the very few known examples of a giant planet orbiting a nearby ultracool dwarf. This discovery finalises a larger astrometric planet search survey that targeted 20 nearby ultracool dwarfs and previously discovered three new ultracool binary systems.

Large language models (LLMs) can be tuned to influence human attitudes, yet the respective contributions of successive post-training stages remain un-clear. This study examines how three successive training stages affect LLM persuasiveness: (1) misalignment through supervised fine-tuning (SFT) on conspiracy data, (2) additional persuasive SFT on argumentative data, and (3) Identity Preference Optimization (IPO), a preference-optimization method. A total of 835 participants recruited on Prolific were randomly assigned to five between-subject conditions (neutral text, conspiracy-trained model, persuasion-trained model, preference-optimized model, and GPT-4) and were exposed to texts on 10 divisive political issues, personalized from their individual profiles in all model conditions. Attitude change was measured as the difference between pre- and post-exposure positions on continuous Likert scales and analyzed with an analysis of covariance (ANCOVA). A significant condition x baseline-attitude interaction, F (4, 825) = 5.33, p < .001, indicated that training effects depended on participants' initial attitudes. Persuasive SFT produced greater attitude change than conspiracy training alone, d = 0.30, whereas IPO provided no additional benefit, d = 0.03, and GPT-4 did not differ from neutral text, d = –0.01. These results show that targeted supervised training on persuasive data increases LLM persuasiveness, whereas preference optimization yields no significant gains beyond it.

Generative speech enhancement (SE) is prone to linguistic hallucination when semantic constraint is unreliable under severe noise and reverberation. Moreover, approaches that generate discrete codec tokens are bounded by the quantization error of the codec decoder, regardless of token-prediction accuracy. We propose LIFT-SE, a two-stage generative framework that decouples linguistic inference from acoustic synthesis within QRes-Codec, which exposes a quantized latent and its residual-completed continuous form. The first stage predicts clean codec tokens autoregressively, conditioned on frame-aligned features from a self-supervised front-end distilled toward clean speech. The second stage applies conditional flow matching to transport Gaussian noise to the continuous latent conditioned on the predicted tokens, and the frozen decoder reconstructs the enhanced waveform. Discrete generation provides naturalness, while continuous refinement restores signal fidelity. Experiments on the DNS1 and URGENT benchmarks show that LIFT-SE attains favorable linguistic consistency under reverberant conditions together with competitive perceptual quality, and systematic ablations verify the necessity of both stages. Code will be released in the future.

Using the recent proof of the Cassels–Swinnerton-Dyer conjecture for cubic surfaces, we finish the classification of $G$-unirational complex cubic threefolds. In particular, we prove that the Klein cubic threefold is $\mathsf{PSL}_2(\mathbb{F}_{11})$-unirational, establishing that the essential dimension of $\mathsf{PSL}_2(\mathbb{F}_{11})$ is $3$. This disproves a conjecture of Dolgachev that the essential dimension of a group is at least its Cremona dimension.

The electrostatic potential of a point charge in a geometry with an arbitrary number of parallel layers of different permittivity, bounded by two grounded planes, is derived. Different boundary conditions in the transverse direction are discussed. For the case where a grounded plane is segmented into strips and pads, the solution is used to derive the weighting fields for these electrodes.

We develop a geometric characterization of electromagnetic memory in terms of the optical data of charged-particle congruences. We derive the worldline deviation of charged particles near null infinity for both massless and massive particles and construct the corresponding optical data. We show that the asymptotic shear of the charged-particle congruence provides an equivalent encoding of electromagnetic memory and its associated large gauge transformation. This establishes a geometric interpretation of electromagnetic memory and large gauge transformations in terms of the asymptotic evolution of charged-particle congruences.

Microwave shielding has recently emerged as a powerful tool for engineering interactions in ultracold polar molecules, yet its potential for controlling cavity quantum electrodynamics remains largely unexplored. Here, we investigate a molecular cavity quantum electrodynamics platform in which two microwave-shielded polar molecules are coupled to a single optical cavity mode and demonstrate that shielding-induced interactions provide a unified mechanism for both photon blockade and Bell-state protection. The anisotropic interaction reshapes the few-excitation spectrum by enhancing its anharmonicity, thereby suppressing multiphoton transitions and improving the single-photon purity by more than three orders of magnitude. The enhanced blockade is accompanied by the emergence of negative longitudinal spin correlations, revealing the interaction-induced suppression of simultaneous molecular excitations. We further show that the photon statistics are highly sensitive to the relative molecular configuration, with positional variations on the scale of the relative zero-point fluctuation substantially modifying the blockade performance. Beyond few-photon nonlinear optics, the same interaction protects an initially prepared molecular Bell state by dispersively decoupling molecular excitations from the lossy cavity mode, thereby suppressing cavity-mediated dissipation and slowing the fidelity decay. Our results establish microwave-shielded interactions as a unified interaction resource for engineering few-photon nonlinearities and protecting quantum states in molecular cavity-QED systems.

Training can drive a transformer into a rank collapse: all token representations point in one direction, and learning stops. In a related collapse of attention, a loss term with a bounded corrective force repairs the network during the run. We ask whether such a term repairs rank collapse. We collapse small transformers by weakening their skip connection and treat copies of the collapsed network. No added loss term repaired the collapse, although the stronger kind pushed with about a tenth of the task gradient. The reason was the path, not the strength. The task gradient no longer reached the query and key weights, which decide where attention looks, and the added term's gradient faded before the blocks where the collapse forms. Restoring the skip connection, which changes no weight, reopened this path at once. The rank then recovered, but only far above the scale of collapse. After a burst of high learning rate the path stayed open and the rank recovered untreated. Registered predictions from the path ranked recovery times but did not transfer to this cause. In every case the loss stayed above that of a healthy network after the rank recovered. Whether a collapsed network can be repaired depends on whether the gradient still reaches the weights that must change, not on how strongly a loss term pushes.

Mask-conditioned diffusion models have become standard for generating labeled training data in biomedical imaging. But the tools used to evaluate them were built for a different problem. Standard metrics like FID and KID score images through ImageNet-trained features. Those features do not transfer to microscopy, leaving the metrics poorly calibrated for specialized domains. Even with a domain-appropriate feature space, a single blended number can still hide a real failure: a model can produce visually plausible images that still transfer poorly to downstream tasks. We identify a specific failure mode: Late in training, realism keeps improving while texture diversity collapses. The model settles onto a fixed color palette, converging to a high-similarity, low-diversity state that standard metrics do not penalize. Therefore, we propose a diversity-aware diagnostic in a domain-validated feature space. It combines a realism axis based on inter-similarity to real images with a diversity axis based on intra-similarity among generated samples. This diagnostic selects a checkpoint that domain experts cannot reliably distinguish from real recordings in a forced-choice study. It also yields useful downstream segmentation: a segmenter trained only on synthetically labeled data reaches a median skeletonised IoU comparable to one trained on real data, at lower cross-fold variance, and clearly ahead of the best available public alternative in this domain, a parametric renderer. We also reproduce a data-efficient hyperparameter-transfer experiment from prior work, tuning several foundation segmenters on a small labeled subset and evaluating on real images: transfer is stronger with our data. We release DiffuMT on HuggingFace, including the triplet dataset, code to reproduce the downstream-utility validation, and a standalone diagnostic tool for evaluating mask-conditioned diffusion models.

Probing the hypothesis space after seeing the data remains valid under possibilistic inferential models (IMs), provided the significance level stays fixed. The price is computation, as each plausibility is a supremum of the possibility contour over the hypothesis, and the contour itself is approximated at each queried parameter value. We propose a possibilistic radial transport, which hides the contour value of a parameter in the radius of its source point. When a transport that maximizes within-shell entropy is picked, sampling parameters covering a confidence cut becomes a matter of truncating the radius. We provide a deep learning algorithm that enforces the contour depth condition while maximizing the entropy within each shell. Our amortization makes coverage and power assessments of the learned approximation practical as well as predictive check of new datasets. We also use the sampler to construct a Bel-Pl spectrum for comparing and selecting interpretable hypotheses that satisfy a prescribed Bel-Pl decision criterion. In simulations the learned contours match or improve on ellipsoidal approximations to the cuts, while the coverage and power track the exact reference. Finally, we probe hypotheses about ovarian aging using synthetic AMH records, asking for each woman how many more years her median AMH level will remain above a specified reference value.