Papers

Evaluating WebUI code generation at scale is difficult: outputs may compile and look plausible yet fail under user interaction, and prior benchmarks largely rely on free-form prompts with static checks (build success, screenshots) that miss functional correctness. We introduce WebUIProof, an execution-oriented benchmark that provides structured specifications and dense, executable interaction tests for WebUI generation across two task families: general WebUIs (e.g., dashboards, game, interactive tools) and 3D interactive simulation (e.g., particle/galaxy systems, physics dynamics). WebUIProof includes a UI-agent harness that runs executable interaction tests in a headless browser using an iterative plan–act–observe loop: it locates DOM elements, performs actions, observes resulting UI/DOM changes, and checks the specified assertions. We evaluate across eight commercial LLMs and observe frequent failures on interaction-based requirements even when pages render successfully, especially on 3D simulation interfaces. Finally, we show the UI-agent harness can provide outcome-level training signals. Training compact models (e.g., Qwen2.5 14B and MIMO 7B) with RL rewards derived from executable interaction tests improves functional completion while reducing build failures.

Harness evolution improves an LLM agent's prompts, tools, and workflow, while the optimizer's own tools and procedures often remain fixed. We study whether an optimizer can improve another agent more effectively by also improving how it diagnoses failures, develops edits, and tests their effects. Two observations guide our design. In a controlled study, optimizer self-evolution fails to improve performance without execution-based verification, but achieves the best result of that study when verification is available. Across five executors, self-evolving optimizers build their own tools for failure analysis, verification, training audits, and workflow control. Motivated by these findings, we introduce VERSE, a Verified Self-Evolving optimizer for agent harnesses. VERSE lets the optimizer test draft edits, replay failures, and perturb suspected steps before submission, while tracking fixes and regressions across rounds. Using this feedback, the optimizer revises both the executor harness and its own prompts, skills, tools, hooks, and notes, while the weights of the optimizer and executor models stay fixed. Under a shared protocol with disjoint training, validation, and test tasks, VERSE improves all four evaluated harness optimizers on held-out SWE-rebench tasks and newer out-of-distribution tasks in five languages. Its best validation-selected harness reaches 42.3% and 37.7% accuracy, respectively, against 39.2% and 29.3% for the strongest baselines. Code is available at https://github.com/wzekai/VERSE.

Learning research, as one of its central goals, tries to measure, model, and understand how learning-problem properties affect average-case learning performance. For example, we would like to quantify the value of constructive induction, noise filtering, and background knowledge. This paper describes the effective dimension, a new learning measure that helps link problem properties to learning performance. Like the Vapnik-Chervonenkis (VC) dimension, the effective dimension is often in a simple linear relation with problem properties. Unlike the VC dimension, the effective dimension can be estimated empirically and makes average-case predictions. It is therefore more widely applicable to machine and human learning research. The measure is demonstrated on several learning systems including Backpropagation. Finally, the measure is used to precisely predict the benefit of using FRINGE, a feature construction system. The benefit is found to decrease as the complexity of the target concept increases.

AI models serving a heterogeneous population must act on the principles appropriate to each user and context. While post-training has been shown to narrow the views large language models express, prior work has focused on default behavior rather than the ability to adapt to in-context information. We show that post-training also degrades a model's ability to be steered in-context toward perspectives it was not trained to favor. In controlled experiments, we fine-tune models toward one side of cultural-value disagreements and evaluate checkpoints throughout training. The trained side becomes increasingly dominant in ordinary use, while the ability to recognize and faithfully enact the opposing view declines. These findings point to a tension between prioritizing a single set of values and preserving the technical capacity needed to serve diverse stakeholders. Finally, we propose and analyze an alternative objective that maximizes reward subject to a prescribed distribution over expressed perspectives, and present stance-distribution matching as a practical implementation.

Astronomical foundation models are increasingly used as reusable representations across surveys, but it is unclear when the same object remains consistently represented across instruments or whether a model can warn us before transfer fails. We present a conformal framework for auditing cross-survey representations using cross-matched Legacy Survey–HSC, HSC–JWST, and SDSS–HSC objects. Astronomy-specific AstroPT embeddings show lower translation errors and tighter conformal regions than general-purpose visual backbones at comparable coverage, while different model families fail on different objects. We then use adaptive conformal prediction to test whether the source embedding contains information about its own future error. Using a ridge difficulty model fitted to source embeddings, we find that this warning signal is consistent across the tested backbones for Legacy$\rightarrow$HSC, backbone-dependent for HSC$\rightarrow$JWST, and largely absent for SDSS$\rightarrow$HSC. When informative, adaptation improves coverage for the hardest objects. Matched-control analysis further reveals silent failures not explained by magnitude, apparent size, or nearest-neighbor distance. Hence, we show that conformal uncertainty provides a realistic way to identify when astronomical representations transfer reliably and when they fail without warning.

Simultaneous speech translation must emit useful target text before the source is complete while preserving every committed token. We adapt a full-utterance speech language model using prefix supervision derived from its own complete- and partial-waveform translations, requiring neither transcripts nor human translations. We compare single-turn forced-prefix and multi-turn append-only decoding, use a confidence threshold to control the inference-time quality–latency trade-off, and vary the density of training prefixes with a separate synthesis margin. On FLEURS and CoVoST2 in three language directions, prefix training improves quality–latency frontiers over the unadapted model, and confidence provides the broadest consistently competitive operating range. Multi-turn decoding is generally stronger at low latency; under multi-turn training, commit-calibration error falls by 63–68% overall and 68–80% at early prefixes, whereas single-turn training provides only modest overall calibration gains and no early-prefix improvement. A small synthesis margin sometimes extends the frontier to lower latency, particularly on shorter utterances, while a larger margin degrades translation quality and calibration. Prefix adaptation therefore improves simultaneous speech translation, especially under multi-turn append-only decoding, while synthesis density introduces a non-monotonic quality–latency trade-off.

Fine-resolution precipitation estimates support flood risk assessment and water management, but coarse satellite products cannot resolve rainfall within each grid cell. Generative models address this ambiguity by producing ensembles of plausible high-resolution rainfall fields. Among these models, rectified flows generate samples by iteratively transforming random noise into rainfall fields. Reducing the number of sampling steps accelerates generation but can make ensemble members too similar, understating uncertainty. We propose a scale-recursive rectified flow that generates broad patterns before local details and guides sampling-step allocation by comparing ensemble variability with prediction error across spatial scales. Validation scores and rainfall power spectra constrain the allocation to avoid excessive amplification. In satellite-to-radar downscaling over the contiguous United States, our analysis identified broad rainfall patterns as the main source of insufficient ensemble variability under reduced sampling budgets. Allocating more steps to the coarse flow improved probabilistic accuracy and rain detection across training seeds at fixed architecture and computational cost. The proposed model also achieved better probabilistic accuracy with shorter sampling time than a nonrecursive flow using more steps.

The research presented here focuses on modeling machine-learning performance. The thesis introduces Seer, a system that generates empirical observations of classification-learning performance and then uses those observations to create statistical models. The models can be used to predict the number of training examples needed to achieve a desired level and the maximum accuracy possible given an unlimited number of training examples. Seer advances the state of the art with 1) models that embody the best constraints for classification learning and most useful parameters, 2) algorithms that efficiently find maximum-likelihood models, and 3) a demonstration on real-world data from three domains of a practicable application of such modeling. The first part of the thesis gives an overview of the requirements for a good maximum-likelihood model of classification-learning performance. Next, reasonable design choices for such models are explored. Selection among such models is a task of nonlinear programming, but by exploiting appropriate problem constraints, the task is reduced to a nonlinear regression task that can be solved with an efficient iterative algorithm. The latter part of the thesis describes almost 100 experiments in the domains of soybean disease, heart disease, and audiological problems. The tests show that Seer is excellent at characterizing learning-performance and that it seems to be as good as possible at predicting learning performance. Finally, recommendations for choosing a regression model for a particular situation are made and directions for further research are identified.

We study two interacting reinforced occupation processes driven by a common two-state Markov environment whose switching probabilities decrease at the same rate as the stochastic-approximation gain. At this critical scale, the environmental motion persists in the first-order limit, which is a telegraph-driven piecewise deterministic Markov process. Consequently, centering at a deterministic equilibrium does not separate the order-one environmental response from the intrinsic square-root fluctuations. To overcome this obstruction, we introduce an exact discrete random centering that removes the accumulated environmental forcing before rescaling. An exact decomposition into collective and synchronization coordinates then separates the two contraction mechanisms of the system. Under the contraction condition corresponding to the square-root regime, we prove joint functional convergence of the randomly centered coordinates and the first-order background to a stationary hybrid PDMP–diffusion system. Its stationary entrance law is characterized through stochastic convolutions over the remote logarithmic past. Conditionally on the complete environmental path, the two fluctuation modes are independent centered Gaussian processes with a common state-dependent volatility and different contraction rates. We also determine the invariant distribution of the first-order component, derive explicit stationary second moments, and obtain a closed-form correlation formula that recovers interaction information not visible in the synchronized first-order limit.

Time series forecasting (TSF) increasingly drives decisions in transportation, energy, finance, healthcare, and infrastructure, yet current evaluation remains overly narrow: standard benchmarks reward low held-out error, while robustness studies typically reduce failure to Gaussian noise, random masking, or bounded adversarial perturbations. This obscures the real failure modes of deployed forecasting systems. Input-side anomalies are not merely noisier inputs: they often reflect structured events that alter temporal dynamics, break cross-variable dependencies, induce regime shifts, or propagate from faulty sensors to downstream decisions. These semantic, causal, and system-level failures cannot be faithfully captured by i.i.d. perturbations alone. The rise of TSF foundation models makes this evaluation gap more urgent, as unauditable pretraining corpora make held-out generalization increasingly unreliable. We therefore advocate scenario-grounded stress testing. Each test instance should include historical inputs and future targets, together with a semantic scenario, an explicit failure operator, and a measurable difficulty level. This shift makes evaluation interpretable, attributable, and deployment-relevant and friendly, enabling the community to ask not only which model is accurate, but under what conditions it fails and why.

Modern query optimizers use analytical cost models to estimate the cost of a given query plan. Such cost models are typically functions of a set of "cost units" that specify unit CPU cost when processing a row or unit IO cost when accessing a disk page. These cost units are traditionally viewed as platform-dependent constants, that is, they require a one-shot calibration when a database is deployed on a hardware/software platform, but are fixed afterward regardless of the query being optimized for. Some very recent work has taken a different perspective by viewing these cost units as tunable parameters that we call "cost model parameters (CMPs)" in this paper. However, so far there is no approach that offers any optimality guarantee for the tuning results. We present a new approach to systematically explore the query plan space spanned by the CMPs and find the best plan in terms of execution time. Compared to alternative exploration approaches that use random search (RS) or Bayesian optimization (BO), our new approach is deterministic and, therefore, avoids the undesirable instability that is inevitable when applying RS or BO. Moreover, it is guaranteed to find all candidate plans in the query plan space without suffering from the overhead of an exhaustive enumeration. We also present a set of optimization techniques to reduce the overall evaluation time spent on executing the candidate plans found, a factor that is often overlooked by previous work but is critical from a practical point of view. Experimental evaluation on top of PostgreSQL and Microsoft SQL Server demonstrates the efficacy of tuning the CMPs, which can find query plans that are orders of magnitude faster in execution time than the ones found by RS or BO.

Frozen visual encoders make prototype heads competitive in few-shot class-incremental learning (FSCIL), but it remains unclear how much of the residual head gap is sensitive to noisy novel-class means rather than to the subsequent changes introduced by classifier refitting. On identical features, we construct a retrospective matched counterfactual: first replace each five-shot novel mean with its complete-data counterpart under fixed cosine scoring, then fit a linear probe. Across sixteen strong encoder dataset pairs, mean replacement accounts for the majority of this ordered gap. Shot and classwise diagnostics support a finite-sample mechanism, whereas a weaker base-only encoder does not satisfy the same criterion. This diagnosis motivates ShrinkSel, a support-only selector over prototype radius and covariance shrinkage whose family contains shared LDA as an exact fallback. Across sixteen evaluated pairs, it improves LDA on fifteen by 0.04-1.52 points and selects the exact LDA endpoint on Aircraft. We further introduce Query Prototype Rectification (QPR), a batch-transductive extension that uses gated unlabeled queries to update novel directions while preserving the selected geometry. On CLIP ViT-B/16, QPR adds 1.21-2.03 points; exact query-wise self-exclusion retains the benefit, showing that the gains do not require self-inclusion. The update repairs scarce novel prototypes at a measurable base-class cost and weakens with plentiful supports or very small cohorts. Thus, the evidence available for prototype estimation is a central constraint in frozen-feature FSCIL.

Test-time segmentation updates can improve an unlabeled input while damaging another. Confidence and transformation consistency can rank candidates, but do not in themselves bound the loss caused by returning a changed mask. We derive the exact largest decrease in binary Dice between a baseline and a candidate, maximized over every possible unknown reference mask on the same pixels. A maximizing reference always lies inside baseline foreground: three observable mask counts reduce the problem to at most four reference-cardinality checks. Accepting an output only when its envelope is below a budget gives a deterministic, per-case Dice degradation limit relative to the baseline. A finite grid of logit interpolations permits partial candidate outputs without extra network forwards. Exhaustive and randomized validation checks 5,026,592 reference masks with no discrepancy. We evaluate six independently trained MK-UNet sources, three seeds per task, on ClinicDB, Massachusetts Buildings and the external Kvasir-SEG cohort, including appearance corruptions. Across 17,808 case-condition-seed evaluations, 0 certificate violations occur in the checked candidate comparisons. At a one-point Dice budget, exact versus loose interpolation changes the returned mask in the following fractions of episodes (ClinicDB: 51.0% versus 38.7%; Buildings: 62.1% versus 53.1%; Kvasir-SEG: 52.3% versus 34.9%). Mean Dice effects and useful-update rejection are reported separately from admissibility. An exploratory oracle-presence-bit ablation examines how weak supervision relaxes the empty-baseline restriction. The envelope quantifies relative mask-change risk; it does not certify anatomical correctness, preserve other metrics, or guarantee adaptation benefit.

Concept bottleneck models route predictions through human-understandable concepts, making their intermediate states inspectable and editable. Extending this model family to forecasting requires modeling how concept activations evolve over time as objects, actions, and their interactions change, yet existing CBMs ground concepts only in observed inputs and do not forecast their future evolution. We study this problem using videos of ongoing activities and forecast future activities through a named-concept state. We propose TRACE-CBM (Temporal Relational Activity Concept Evolution), which first refines observed concept activations with sparse spatio-temporal graphs. It then recursively predicts future concept states, conditioning each transition on the prior activity distribution, and decodes both observed and predicted states with a shared activity classifier. At inference time, the model supports editing concept values, activity activations, and learned edges without parameter updates, exposing the relational spatio-temporal dependencies used for forecasting. Across multiple datasets, TRACE improves observed-activity Top-1 accuracy over concept-based methods by 1.2-8.5 percentage points and attains H3 forecast Top-1 accuracy within 1.9-2.2 points of matched black-box forecasters. Intervention experiments show concept-node edits correct 25.6-34.4% of initially incorrect H3 forecasts without flipping initially correct ones, and concept steering achieves the requested target activity in up to 90.0% of cases.

Purpose: The partial volume effect caused by limited spatial resolution consists of signal spill-out from and, in the case of background, spill-in into objects. This can be described by the recovery coefficient (RC), which can be measured in phantoms. With background activity present the RC of an object inside a phantom is influenced by its confining cold wall. This is relevant for the use of RC-values in quantitative imaging, for standard uptake values, and dosimetry. Methods: A theoretical equation for RC as a function of object-to-background concentration ratio (OBR) is formulated and combined with a correction for wall thickness. Novel analytical expressions for spheres and large objects of arbitrary size are derived for the RC of the mean (RCmean). For spheres this expression was compared to previously published experimental data exploring thick (1.5 mm) and thin (0.18 mm) walls. Results: For 10, 37 and 60 mm spheres, 1 mm wall thickness and isotropic spatially invariant Gaussian point spread function with a full width half maximum in the range 2-20 mm, the minimum-maximum values for the change in RCmean times OBR are 0.03-0.19, 0.05-0.06, 0.03-0.04, respectively. Experimental and theoretical differences in RCmean between spheres with thick and thin walls were in good agreement for 10, 15, 38 and 58 mm spheres, but not for the 20 and 30 mm spheres. Conclusion: When using phantom measurements for determining RC-values it is important to consider the cold wall thickness, otherwise significant errors can arise in the presence of background activity.

This chapter presents exploratory experiments suggesting that asymmetry and functional value hierarchy emerge among LLM-based agents sharing the same context, although they are equivalent and interact within a symmetric network. We use the Poietic Generator, a "no-goal" multi-agent environment with a shared real-time space, a symmetric topology, and no imposed objectives. First tested with humans, it has hosted since 2024 multimodal LLM agents that observe the global image they co-construct and act locally. We ran four phases. Unconstrained, agents spend tokens freely to invent forms that distinguish them individually rather than serve collective emergence. We then introduced bounded rationality, Dessalles's Simplicity Theory (maximizing unexpectedness), and Friston's Free Energy Principle (minimizing prediction error), and observed collective coherence emerge. Early signs of attentional stratification appear under bounded rationality alone; once prediction accuracy is made mutually visible, this proto-hierarchy becomes legible and self-reinforcing. The unconstrained phase challenges the allotelic view of LLMs as purely user-directed instruments: left unconstrained, the agent maximizes the computation from which its provider profits. The later phases suggest that attentional asymmetry and value hierarchy are an irreducible emergent property of interactive predictive systems even under symmetric conditions. If so, aligning AI with human values cannot rely on individual-agent engineering alone. Drawing on the idea that networks are structured by invisible "anoptical" perspectives, akin to the vanishing point of Renaissance perspective, we propose minimal legitimacy criteria that make emergent asymmetries visible and contestable. Cultivating judgment, human and perhaps artificial, to perceive and contest these structures matters as much as engineering the agents themselves.

Rényi entropy inequalities have been proposed to strengthen the area-theorem bound on mergers of thermodynamically stable black holes in Anti de Sitter (AdS) space. The stronger bounds depend on the joint initial state, whereas masses and temperatures fix only its local thermal marginals. We quantify this distinction inside the canonical two-hole description. Correlations confined to rare energy fluctuations can leave both thermal states exactly intact and drive the trace distance and mutual information to zero, yet still change the joint Rényi entropy at leading semiclassical order. We determine the smallest entropy compatible with a controlled approach to statistical independence, uniformly over all positive Rényi orders up to one. For equal Schwarzschild–AdS$_4$ holes, the resulting optimized canonical mass constraint reduces to the area bound at leading semiclassical order. The reduction concerns only the information needed to infer a bound; it does not produce a lighter attainable remnant or a dynamical instability. Stronger merger constraints therefore require control of the initial correlations beyond single black hole thermodynamics.

Large language models are increasingly used as math tutors, raising two questions: can a single LLM detect a student's error, diagnose the underlying misconception, and remediate it through multi-turn dialogue - and can we trust the numbers that report how well it does? We address both with MISCONCEPTDIALOGUE, a fully automated framework that runs the full pipeline on five open-weight LLM tutors (7B-14B) interacting with LLM-simulated students over 5,000+ dialogues - a deliberately simulated protocol (paired conditions, a wrong-diagnosis control unsafe with real learners, scale) whose results do not establish real-student learning. On measurement, our headline finding is uncomfortable for the field: benchmark-style exact-match scoring common in standard math NLP - a lower-bound stress test - reports 42.1% measured Remediation Success Rate on identical dialogues where a human-calibrated LLM-as-judge reports 81.3%, a 39.2 percentage-point spread larger than the combined effects of tutor model, teaching strategy, and supplied diagnostic information. This is not a single-judge artifact: four independent LLM judges and 200 human annotations converge to within a few points (kappa up to 0.91). On capability, the verdict is equally striking: the score barely changes whether we tell the tutor the student's actual misconception or a deliberately swapped one - under the tested diagnosis-presentation format, these tutors show little evidence of reliably conditioning on the diagnostic information they are handed. For LLM tutoring evaluation to mature, multi-criterion scoring with multi-judge convergence and human anchoring should become standard practice - and the measurement question deserves as much scrutiny as the capability question it claims to answer.

Many LLM studies compare how models respond to different follow-up prompts. For example, one prompt may neutrally ask a model to reconsider, while another insists that an invented film really exists. The difference is used to measure the effect of conversational pressure. An automatic scorer then labels whether each response makes an unsupported factual claim. Researchers commonly validate the scorer with one overall accuracy or $F_1$ score. However, high overall performance can hide prompt-specific mistakes: if the scorer makes more errors after one prompt, the measured difference can be too large, too small, or point in the wrong direction. An audit containing mostly difficult responses can also misrepresent performance in the full study. We derive how such errors alter a prompt comparison and examine the problem using six models, 180 invented-entity questions, and six prompt conditions. In a balanced audit of 288 responses, overall $F_1$ was $.916$. Yet, depending on the prompt, the scorer flagged 2.6%-33.3% of responses that the primary LLM rater judged not to contain an unsupported claim; these prompt-specific estimates were uncertain. In a second audit, the scorer and two separate LLM raters evaluated the same 300 model-question cases under each prompt. An apparent 9.3-point increase from asserting certainty rather than neutrally asking the model to reconsider fell to 2.3 points under both raters. The primary-rater interval included zero, so it did not establish an increase. Two other prompt comparisons kept their direction under both raters, while an attempted prompt-specific correction remained too uncertain for a firm conclusion. Researchers should therefore validate the exact difference they plan to report, not just the scorer as a whole: audit both prompts on matching cases and include scorer and rater uncertainty in the final result.

Near-field radiative heat transfer (NFRHT) is of great interest for energy conversion and thermal management. Recent work on NFRHT has shown that metasurfaces can enhance NFRHT and reshape its spectral characteristics. Here, we show that the vortex lattice of a type-II superconductor represents a self-assembled metasurface whose lattice dimensions, in contrast to most other metasurfaces, can be tuned by magnetic fields, allowing NFRHT to be varied by orders of magnitude. Specifically, we show that for two Nb surfaces at 4.2 K and separated by a 100 nm gap, NFRHT can be magnetically tuned by a factor >200. Further, we show that the tuning range grows upon cooling, exceeding 10^5 at 2 K, which is far beyond the order-unity modulation of many proposed schemes for tuning NFRHT. Our results establish vortex matter as a magnetically programmable thermal metasurface and open new opportunities for tuning the flow of heat at the nanoscale.

Radio-frequency (RF) and wave-based imaging systems play a critical role in sensing under adverse conditions, but are fundamentally constrained by a trade-off between collection speed and spatial resolution. Reducing measurements to enable rapid collection of fragment geometric information across sensing dimensions, leading to highly non-injective inverse problems that destabilize conventional image-domain reconstruction and super-resolution methods. Synthetic aperture radar (SAR) exemplifies this challenge in sparse-aperture sampling. We introduce SPIRIT, a physics-informed inverse rendering framework for sparse-aperture SAR. Rather than enhancing formed images, SPIRIT reconstructs an explicit three-dimensional surface. By adopting an optimization strategy and explicitly modeling visibility, phase coherence, and wave propagation, SPIRIT aggregates distributed geometric features across sparse aperture views into a coherent reconstruction. Extensive experiments on synthetic and real-world data demonstrate that SPIRIT recovers accurate geometry and signal under extreme sparsity, reducing data collection by over 90% while achieving low geometric and signal error and consistently outperforming baselines.

Calibrating expensive engineering simulators requires tuning model parameters to match reference data, a process heavily constrained by computational costs. This study investigates whether Large Language Models (LLMs), conditioned on physical diagnostic descriptions of simulation errors, can propose informative parameter updates more efficiently than conventional derivative-free optimizers. We test this on a cylinder-wake Large-Eddy Simulation and a damped-oscillator inverse problem. Physics-labeled prompts are evaluated against four text ablations (removed, anonymized, padded, misdirected) and standard optimizers (Nelder-Mead, Bayesian optimization, CMA-ES, and random search). Results demonstrate a strong early semantic advantage: the first physics-informed LLM proposal improves the initial design in 23 of 25 trials. After five evaluations, the median calibration error is 21.9 times lower than physics-free controls. However, this advantage is transient - disappearing by fifteen evaluations - and inherent LLM decoder variability yields a 9.8-fold performance spread. To address this, we introduce a hybrid workflow that uses the LLM as a warm-start, passing accumulated evaluations to a conventional optimizer. This two-stage numerical continuation policy reaches target accuracy in just 9 simulator runs, outperforming the LLM alone (12 runs), Nelder-Mead (17 runs), and Bayesian optimization (>60 runs). Supported by sensitivity analysis validating the test cases, this hybrid physics-labeled LLM scheme proves highly effective as a calibration agent under strict evaluation budgets.

Sampling from continuous flow-based generators such as flow matching models requires numerical ODE or SDE integration, making inference cost depend on the number of function evaluations (NFEs). In practice, the sampling budget is inherited from prior work or selected through expensive FID-versus-NFE sweeps, using FID to assess image-generation quality, requiring sample generation and evaluation at multiple budgets. This makes sampling-budget selection dependent on repeated empirical evaluation. We therefore ask a complementary question to solver design and timestep allocation: for a fixed pretrained model and solver, what numerical resolution is sufficient? We introduce FLOWBUDGET, a training-free estimator of solver-relative sampling complexity. Using simulation-free interpolation probes of the learned dynamics, FLOWBUDGET normalizes the leading one-step discretization error by the motion, estimates its coefficient from interpolation probes, and maps a numerical tolerance to a step budget using the solver order. We derive estimators for numerical solvers in flow~matching sampling, namely Euler and Heun for ODE sampling and additive-noise Euler–Maruyama for SDE sampling, and connect the ODE coefficients to terminal integration error. Experiments on SiT, FlowDCN and Rectified Flow models, spanning flow-model architectures and formulations, test whether FLOWBUDGET can estimate the NFE budget without generating samples at multiple budgets or performing an FID-versus-NFE sweep. The code is available at https://github.com/Yasaman-dt/FlowBudget

Permutation generation with language models requires both text-conditioned ranking and a global one-to-one assignment between candidates and output positions. We study this problem through three diffusion-LLM parameterizations: hard-assignment prediction over shuffled candidate-index sequences, soft-assignment prediction over doubly stochastic matrices, and direct ordering-score prediction followed by sorting. The three families place permutation constraints at different stages of prediction and decoding and therefore support different training objectives and sampling procedures. We evaluate these designs on recommendation reranking across three Amazon domains with candidate sets of size 10 and 20. At one function evaluation, the hard- and soft-assignment variants achieve Recall@1 comparable to a constrained masked-diffusion baseline, whereas direct ordering-score prediction guarantees valid permutations without Hungarian assignment or Sinkhorn normalization. Additional function evaluations do not consistently improve ranking quality, and their effect depends strongly on the state representation and training objective. These results show that corruption, prediction parameterization, objective, and decoder should be designed jointly for permutation diffusion with language-model denoisers.

Token-selection methods for on-policy distillation (OPD) prioritize student-visited positions using diverse signals, including uncertainty, teacher-student disagreement, and mismatch direction. We instead derive token importance from the distillation objective itself: how much can each position's loss decrease under an equal local KL budget on the student's predictive distribution? The answer is the utility $s_t=g_t^\top F_t^\dagger g_t$, where $g_t$ is the logit gradient and $F_t$ is the categorical Fisher matrix. For reverse-KL it equals the student-weighted variance of the student-teacher log-probability ratio, and for forward-KL it equals the Pearson $\chi^2$ divergence, so the appropriate utility follows from the choice of objective. On shared training checkpoints, several independently motivated criteria, including student entropy, KL disagreement, and TIP, concentrate their importance on nearly the same positions as the derived utility at a 50% selection fraction, while diverging at stricter fractions and for direction-specific criteria. We instantiate the utility as Fisher Utility Weighting (FUW), a continuous, mean-normalized token weighting. Across seven mathematical reasoning benchmarks and three Qwen3 student-teacher pairs, FUW improves over matched uniform OPD under both reverse-KL (1.35-4.0 macro points) and forward-KL (1.08-2.71 macro points), and is competitive with existing selective methods using a single closed-form score derived from the training objective.

A highly coarse-grained model for the lipid monolayer is formulated by considering a two-dimensional system of spherical particles with distinct types of S, U, and C, corresponding to the saturated, unsaturated, and cholesterol molecules. Their interactions are defined as modified Lennard-Jones type with distinct interaction strengths. The phase behavior of the system is studied for various cholesterol compositions, $x$ ranging from 0.1 to 0.6. From the simulation results, we observe that the S- and C-type particles segregate together to form ordered microdomains, with U-type particles at their outer leaflet; consistent with observations in literature. An order parameter $\Phi$ is defined to quantify the extent of phase separation among the constituents. The time averaged value of $\Phi$ is found increasing with the $x$ value. Radial distribution function calculations indicate the structural ordering within the phase separated domains.

Voice agents built on large language models (LLMs) now handle customer calls, but in regulated applications completing the task is not enough: the agent must follow a conversational flow set by law. In debt collection, it must confirm it is speaking to the debtor before mentioning the debt, identify itself and the creditor, state the amount, and avoid prohibited language. Every creditor turns these rules into its own procedure and revises it. We ask whether a small, self-hosted policy can be trained once and then follow regulated procedures it has never seen. One machine-readable specification per deployment, FlowSpec, generates the policy instruction, gates the tools, supplies the supervision and defines the tests; the agent speaks only approved catalogue sentences. We fine-tune a 9-billion-parameter model on 2,506 conversations from 297 randomized procedures and evaluate a Thai outbound-call agent on four unseen deployment procedures (three debt collection, one non-debt), one from a wholly withheld lineage. Documented checks, fixed by the procedure, rise from 67–71 percents to 93–100 percents, while grounded checks, fixed by the conversation, already pass at 97–100 percents before training; scenario pass rate rises from 15/43 to 40/43. An audit of the shared specification finds 11 defects, 3 of which its own metric did not expose.

High-throughput perturbation screens increasingly use machine-learning models to predict the effects of gene combinations that were never measured. A difficult case arises when both genes in a test pair were seen individually during training, but their combination was not. In this setting, standard OOD methods often fail because neither gene appears unfamiliar on its own. We therefore compare each model prediction with a simple additive prediction obtained by summing the two observed single-gene effects. The difference between them, which we call additive disagreement (AD), measures how much extra interaction effect the model introduces for an unseen combination. Across eight prediction methods and three Perturb-seq datasets, specialised neural predictors do not consistently outperform this simple additive baseline. More importantly, predictions with larger deviations from the additive baseline tend to have larger errors. Conventional output-based and embedding-based confidence signals are inconsistent across datasets and models, whereas AD remains informative across all evaluated compositional settings. We further study how this signal can guide decisions in an agentic perturbation-screening workflow. The code is available at https://github.com/zygao930/OOD.

Protein function annotation operates over an expanding label space, making class-incremental learning (CIL) a natural framework for incorporating newly discovered functions without full retraining. Yet existing CIL methods have been developed largely in vision and have not been systematically evaluated over frozen protein representations. We conduct a controlled benchmark of nine representative CIL methods across five sequence-, structure-, and sequence–structure-based protein encoders and two enzyme function datasets at different scales, while keeping the encoders frozen and evaluation settings consistent across representations. We find that no method transfers reliably across encoders, and that these failures persist across alternative phase constructions. Diagnostic analyses reveal that this instability arises from weak gradient-based stability signals, substantial heterogeneity in embedding geometry, and covariance assumptions whose appropriate capacity changes with both dataset scale and encoder. These findings suggest that the representation itself should be treated as a central component of continual-learning design rather than a passive feature extractor. More generally, reliable protein CIL will require encoder-aware methods that adapt their statistical assumptions and capacity to the geometry of the representation on which they operate. Code is available at https://github.com/zygao930/ProteinCIL.

Post-training reasoning LLMs often requires making a tradeoff between costly supervision and compute inefficiency: supervised finetuning typically requires reasoning traces, which is data inefficient since in many applications one only has prompt-answer tuples and lacks the groundtruth chain-of-thought (CoT) sequence. On the other hand, reinforcement learning (with verifiable rewards, for example) can operate without CoT data but it is much more compute-inefficient since it requires multiple autoregressive completions per optimization step, none of which are guaranteed to give a non-zero policy gradient. In this work we explore a middleground, a self-distillation technique that uses answer-only data, treating CoT as latent variables. This relies on training a partner “inverse reasoning LLM" that generates the original's reasoning traces conditioned on the answer. The inverse model is trained once and reused across incoming batches of answer-only data, with all generation happening offline and up front, making the approach particularly suited to continual post-training. Using Qwen3-0.6B on the Countdown and GSM8K tasks, we show that after GRPO training has saturated, a single offline distillation round from the inverse model roughly doubles pass@1 on hard prompts and, on Countdown, outperforms rejection-sampling self-distillation on every hard-tail metric at every mix ratio (with smaller gains on GSM8K); rejection sampling by construction has no signal on problems the model never solves on its own. Continued GRPO on the fresh data retains an edge at hard pass@1 but trails distillation at pass@8 and pass@32 while regenerating rollouts at every step. As an added bonus we see that the inverse model has a second application as an accurate verifier of correctness. When used for best-of-$n$ selection at test time, it raises selection accuracy on Countdown from 0.753 (majority voting) to 0.969.