Papers

Large language models (LLMs) have become the foundation of personalized assistants, but maintaining persistent user memory across long-term interactions remains challenging. Existing memory systems often focus on storage, retrieval, or consolidation, while memory writing remains less controlled: transient requests, duplicate statements, and outdated user states may enter memory and later be retrieved for personalization. In this paper, we present AMU: Admission and Memory Update for Personalized Conversations, an SLM-guided (Small language model guided) structured framework for writing-time memory control. AMU uses structured memory filtering to decide what should enter memory and SLM-guided storage management to determine whether an admitted record should be stored separately, discarded as a duplicate, or fused as an update. We evaluate AMU in a controlled memory writing and retrieval setting. Experimental results show that AMU maintains cleaner and more retrievable personalized memories.

Operational estimation of wheat-cultivated area is persistently constrained by discordance between administrative record-keeping and remotely sensed classification products. We address this administrative reference discordance for the 2022 Rabi season in Patiala district, Punjab, India, using a thirteen-timestep Sentinel-2 NDVI time series. A curated 849-sample reference dataset, developed through an iterative rule-based bootstrapping procedure, underpins both a feature sensitivity analysis and an operational classifier. Feature sensitivity independently assessed via Cohen's d and gradient-boosted information gain converges on the February-to-March grain-fill window as most discriminative. Four classifiers (1D-CNN, LSTM, hybrid CNN-LSTM, and XGBoost) were benchmarked on an identical 679/170 sample split. XGBoost achieved the highest overall accuracy (78.82%) against deep-learning baselines (64-66%), consistent with tree-based ensembles' favourable parameter-to-sample ratio in low-sample regimes. At full-population deployment across 36.25 million valid district pixels, the operational classifier attained 86.31% precision and 71.05% recall. The predicted wheat extent deviated by only +2.99% from the official tabular target, whereas the government's spatial reference mask exhibited a +25.11% positive area bias against the identical target. This asymmetry indicates that a classifier trained on an auditor-curated reference set reconciles more closely with the official tabular area than the spatial product conventionally used to validate it. We present this dual-track curation-and-classification framework as a methodological reference for crop-area reconciliation in label-noisy administrative settings.

Text-to-speech models loop, truncate and lose count on text that repeats a phrase many times. We show that repetition itself is what breaks them, not the length that comes with it. Every repeated sentence in our test set is paired with a control of matched sentence and word count in which no word ever repeats back-to-back. Six models from three architectures render the controls almost perfectly and fail the repeated twins: 94.3% against 18.2% exactly right at k >= 6. The gap survives greedy decoding, repetition-penalty sweeps, four independent speech recognisers and 420 analysis specifications without once reversing sign; a held-out fourth architecture lands within a point of its predicted gap, and one of two non-autoregressive baselines shows the same failure. Varying the period of the text shows the failure grows smoothly with periodicity, half of it surviving when no word is adjacent to itself.

We introduce the notion of Alexandrov region. In contrast to Alexandrov spaces these are intrinsic metric spaces, which satisfy a lower curvature bound and have finite Hausdorff dimension, might however, be non-complete. We show a quantitative version of the statement that such regions are locally isometric to an Alexandrov space.

We show that the map $u_*\colon B_1\subset\mathbb R^3\to\mathbb S^2$, $x\mapsto x/|x|$, minimizes the anisotropic energy \begin{align*} \int_{B_1} \Big( k_1(\mathrm{div}\, u)^2+k_2 (u\cdot \mathrm{curl}\, u)^2 +k_3 |u\times\mathrm{curl}\, u|^2 \Big) \, dx, \end{align*} among $\mathbb S^2$-valued maps agreeing with it on the boundary, for values of $k_1,k_2,k_3>0$ beyond the known regime $k_1\leq k_2$. Our main tool is a new stability estimate for $u_*$ as a minimizing sphere-valued harmonic map.

Cross-subject EEG-to-image retrieval requires a neural represen- tation trained on source subjects to remain aligned with a visual embedding space for an unseen subject. Whereas existing methods primarily focus on the EEG side, we address this problem from the perspective of the visual target. Our approach preserves the spatial information of the Perception Encoder, converts its patch grid into a compact set of learned visual views, and aggregates them for each image with a block-structured, content-dependent router. The target is learned jointly with the EEG encoder through contrastive learning with MMD regularization across source subjects. For deployment, we propose a training-free representation refinement that aligns frozen embeddings without updating either encoder. Under leave- one-subject-out evaluation on THINGS-EEG2, the structured target achieves 35.3%/65.6% Top-1/Top-5 accuracy, the best among com- pared methods. Refinement raises this to 48.1%/77.1%, an 18.5% Top-1 gain over the strongest compared method, improving all ten held-out subjects.

The Rashomon effect describes the existence of multiple near-optimal models that achieve comparable performance while offering fundamentally different explanations. This creates a critical vulnerability in AutoML: x-hacking, the selective post-hoc choice of a model based on its explanation rather than predictive merit. No existing AutoML framework exposes this risk. We introduce ARSA ML, an open-source Python framework that quantifies Rashomon set structure and predictive multiplicity within AutoML pipelines. Using ARSA ML, we benchmark AutoGluon and H2O across 28 binary classification datasets, and conduct a post-hoc x-hacking analysis revealing a consistent structural asymmetry: AutoGluon produces larger, diverse sets with stable explanations, while H2O generates compact sets with markedly higher prediction divergence and explanation instability – making H2O users considerably more exposed to x-hacking. This gap persists across all evaluated metrics and epsilon thresholds, pointing to a fundamental difference in each framework's model-building strategy. ARSA ML is available at https://pypi.org/project/arsa-ml/ .

3D Gaussian Splatting (3DGS) is a state-of-the-art technique for 3D scene rendering, offering high efficiency and excellent visual quality. However, because 3DGS relies on an initial sparse point set from Structure-from-Motion (SfM) and view-dependent properties, it can suffer from geometric inaccuracies and visual artifacts, particularly in complex scenes. To address these challenges, we propose an improved 3DGS approach that regularizes the optimization process by integrating geometric priors, including surface normals and dense depth information. Surface normal regularization improves geometric consistency by aligning Gaussian covariance with local surface structures, while dense depth priors combined with an initial points from SfM enhance per-pixel depth estimation, increasing accuracy and reducing ambiguities. These enhancements enable robust handling of diverse and complex real-world scenarios, minimizing visual distortions and improving reconstruction quality across various environments. To validate our method, we evaluate it on challenging datasets, including street-view scenes and highly reflective environments, while testing it across multiple SfM pipelines. Our results demonstrate compatibility across diverse environments and highlight the robustness of our approach. Experimental findings further show that our method enhances geometric accuracy and visual quality, establishing a reliable solution for real-time 3D scene rendering in complex environments.

Unsupervised tabular anomaly detection (TAD) aims to identify anomalous rows in tabular data using normal training samples. While conventional methods rely on dataset-specific training and configuration search, recent tabular foundation models (TFMs) enable zero-shot anomaly detection on unseen datasets via in-context learning. Most TFM-based approaches, however, require anomaly-specific pretraining from scratch, making detection inherently dependent on synthetic TAD-specific priors and costly to update. Some approaches instead repurpose pretrained general-purpose TFMs for TAD to avoid this burden, but rely on computationally expensive formulations with restrictive anomaly inductive biases. In this work, we introduce TaskBridge, a new framework that efficiently repurposes pretrained general-purpose TFMs for unsupervised TAD by constructing virtual supervised tasks that directly recast anomaly detection as supervised in-context inference of TFMs. The resulting virtual tasks induce predictive structures under which normal queries and their target pairs receive high support, whereas anomalies tend to violate the induced structures and receive lower support, providing direct anomaly evidence. Across 790 real-world datasets, TaskBridge consistently outperforms 30 baselines, including state-of-the-art TFM-based approaches, without anomaly-specific TFM pretraining or dataset-specific model optimization.

Existing VLA pruning strategies primarily select individual visual tokens according to task-level semantic relevance, while overlooking the spatial information required for robotic manipulation. To examine this limitation, we construct a simple Stride baseline that uniformly samples tokens along the flattened one-dimensional visual sequence, representing a purely geometric pruning strategy. Surprisingly, Stride outperforms semantic pruning and random pruning at certain pruning ratios, but collapses when the token budget is only slightly reduced. We characterize this phenomenon through the spatial coverage radius, defined as the largest spatial blind spot induced by the retained token set after pruning. Our analysis reveals a strong correlation between the spatial structure of retained tokens and task success, suggesting that reliable VLA pruning requires preserving not only task-relevant tokens but also the spatial scaffold of the scene. Motivated by this diagnosis, we propose GeoScaffold, a training-free visual token pruning method that partitions each image into spatial regions, allocates inter-region token budgets using task-relevance weights, and selects intra-region scaffold tokens via farthest point sampling to reduce the local coverage radius. On pi 0.5 and LIBERO, GeoScaffold retains only 20% of visual tokens while preserving a 93.2% average success rate, and achieves a 1.78 times prefill speedup over the unpruned baseline.

Covariate effects vary across contexts and shift over time, requiring forecasters to assess how to use them for each forecasting context. As forecasting proceeds, observations for earlier forecasts become available, providing feedback on past covariate use for subsequent forecasts. However, when multiple covariates act together, the forecast error reveals the numerical discrepancy from the observation but not how the covariates should have been used. We introduce JudgeCast, an experience-based framework for time series forecasting with covariates. Following the judgmental adjustment practice, a frozen TSFM provides the base forecast, while a frozen LLM uses the current context and relevant experience to adjust it. Within the adjustment, assessing covariate effects and determining the numerical adjustment serve distinct roles, so JudgeCast first forms explicit covariate-wise judgments and then determines the adjustment. After observation, JudgeCast uses the observed residual of the base forecast to reconstruct alternative judgments and evaluates the original and alternatives through their resulting adjustments. The best-performing decision is selected and retained as validated experience for subsequent forecasts. Across diverse real-world datasets, JudgeCast outperforms strong baselines. Ablations show that explicit covariate-wise judgment can improve forecast-time adjustment, while residual-guided experience construction yields more reliable forecasting gains than retaining raw decisions as experience.

System One models such as Jev offer an efficient alternative to generative language models for tasks that require decisions rather than open-ended responses. However, existing Jev models exhibit limited Chinese-language decision accuracy, restricting their utility in both general and specialized settings. In this paper, we introduce Chinese-Jev, a System One model that addresses this gap through a unified data processing and training pipeline. Our data processing protocol converts heterogeneous Chinese-language annotations into probability targets over candidate options, enabling a shared training formulation across domains and question formats. To enable efficient inference, Chinese-Jev adopts a lightweight encoder-only backbone for text encoding and learns to score candidate answers through decision-oriented training. To address the misalignment between the pre-training distribution and downstream Chinese-language scenarios, we first train the model on a general-purpose corpus of 10 million examples, then fine-tune it separately for the medical, legal, and financial domains. To evaluate decision accuracy and calibration in both general and domain-specific Chinese-language settings, we introduce Chinese-Jev Bench (CJ-Bench). After first-stage pre-training, Chinese-Jev exceeds the accuracy of the closed-source Jev model by 1.24% on general-domain tasks while achieving a 20.3x speedup. Subsequent domain-specific fine-tuning yields a 4.0% accuracy improvement over Jev in medicine and achieves 92% of Jev's average accuracy across specialized domains, with a 17x speedup and an average latency of only 15 ms per example. We further demonstrate on-device deployment of an INT8-quantized model on mobile devices, achieving an inference latency of approximately 1.0 second per decision. The project is available at https://gulucaptain.github.io/Chinese-Jev/.

We report the acceleration of stable electron beams with energies up to 1.5 GeV using a 50 TW laser pulse guided in a preformed plasma channel generated by an axiparabola-focused heater pulse. Varying the channel length changes the plasma density beyond the channel end and produces distinct electron spectra. A short channel causes a large decrease in bubble size after the truncated channel, leaving only part of the electron bunch in the accelerating phase and resulting in a quasi-monoenergetic peak. For a longer channel ending near the target density down-ramp, the smaller density increase rephases and further accelerates a larger fraction of the bunch, producing a continuous high-charge spectrum. Hydrodynamic and particle-in-cell simulations reproduce the observed spectral evolution. Together, the measurements and simulations identify channel length as a key control parameter in guided laser-wakefield acceleration.

Multistage Stochastic Programming requires scenario trees that accurately approximate the unknown underlying uncertainty process without violating the non-anticipativity constraint. Most of the existing method for scenario tree generation usually target just one of the two goals at the expense of the other, and just some recent methods try to simultaneously address both. In this paper, we propose to combine two recent distances from Optimal Transport Theory, specifically the ultrametric Gromov-Wasserstein and the Fused Gromov-Wasserstein, into a unique measure to simultaneously deal with the two goals. Our main result is a stability theorem showing that the optimal value of a Multistage Stochastic Programming problem changes by an amount controlled by a Holder-type power of the proposed distance between the true process and its scenario tree approximation, extending Wasserstein-based stability results to control marginal fidelity and non-anticipativity, simultaneously. We propose a block coordinate descent algorithm for the our scenario tree generation based on the proposed distance, and evaluate it on different inventory management test cases.

Accurate quasiparticle electronic structures are the foundation for understanding excited-state properties of materials and explaining optoelectronic, quantum, and transport phenomena. First-principles GW calculations nevertheless remain computationally intensive for large or configurationally complex systems. Here we introduce DeepH-GW, a deep-learning framework that predicts an effective GW quasiparticle Hamiltonian directly from atomic structure. Building on the local, equivariant message-passing architecture of DeepH, DeepH-GW is trained on high-fidelity plane-wave GW calculations through a real-space Hamiltonian-reconstruction interface. This approach combines the systematic accuracy and broad chemical applicability of plane-wave methods with linear-scaling neural-network inference. Across the systems examined, DeepH-GW reproduces quasiparticle band structures with errors on the order of a few meV. Moreover, we show that, despite the intrinsic nonlocality of many-body interactions, DeepH-GW exhibits strong cross-scale transferability from relatively small training structures to substantially larger supercells. We demonstrate that the framework can accurately capture GW-level electron-phonon band-gap renormalization through predictions in supercells with thermal displacements, illustrating the power of the approach. DeepH-GW therefore provides a practical route toward large-scale many-body simulations and foundation models for excited-state electronic structure.

Double white dwarfs (DWDs) in the Laser Interferometer Space Antenna (LISA) band are major Galactic gravitational-wave sources, but many tight DWD progenitors may reside in hierarchical triples. Triple evolution can alter both the production rate and parameter distributions of LISA DWDs. Using the Multiple Star Evolution (MSE) code, we study how the adiabatic mass-loss model (Ge model) and the SCATTER formalism for triple common-envelope (TCE) evolution affect triple-channel LISA DWD formation. Our main model evolves 10^5 triples with both ingredients. Three control models, each with 10^4 triples, vary the mass-transfer stability criterion to the original MSE polytropic model, remove SCATTER, or adopt low metallicity (Z=0.001). We also evolve 10^6 isolated binaries with the same framework as a binary-channel reference. Systems that are dynamically unstable under the polytropic model can instead undergo stable Roche-lobe overflow under the Ge model. This sustained mass transfer modifies the donor envelope, reduces its binding energy, and helps the subsequent common-envelope phase form a close DWD rather than a merger. The Ge model increases the triple-channel LISA DWD yield by about 60% relative to the polytropic model, while SCATTER changes the integrated yield by only about 3%. After residence-time weighting, triples contribute about 1.3e7 LISA-band DWDs, including 1.4e4 individually resolvable systems. In the combined triple-plus-binary normalization, the binary channel contributes 3.1e6 LISA-band DWDs and 4.7e3 resolvable systems; these are the binary share of the combined population, not the binary-only Galactic estimate. The main triple model has a flatter, higher-q mass-ratio distribution than the polytropic-model triples, is dominated by He-CO and He-He binaries, and produces rare eccentric systems only when a tertiary star remains bound.

We introduce the basics of Effective Field Theories, using the Standard Model Effective Field Theory as our central example. The first part of this chapter gives a pedagogical introduction to EFTs - including matching, resummation and the construction of an operator basis - with explicit examples and provides a starting point for newcomers. The second part reviews the current status of SMEFT and collects suitable references for those who want to go deeper.

Mixture-of-Experts (MoE) has increasingly become a mainstream approach for scaling large language models, as it expands model capacity while keeping computation cost nearly constant. Training large-scale MoE models relies on Expert Parallelism (EP), which distributes expert replicas across GPUs and exchanges tokens through all-to-all communication. The efficiency of EP is often constrained by two system bottlenecks: cross-node token transfers are limited by inter-node bandwidth, while skewed expert workloads lead to imbalanced computation across GPUs. Prior work mitigates these bottlenecks based on per-expert workload statistics, but overlooks the fact that experts could share the communication. In this work, we empirically present the observation that many pairs of experts are frequently co-activated by individual tokens. Motivated by this, we present Cobalt, an efficient MoE training framework that leverages expert co-activation to reduce cross-node traffic and workload imbalance. Cobalt adopts a two-stage expert layout planner that adapts expert layout to the evolving expert co-activation and workload conditions. It periodically co-locates frequently co-activated experts on the same node to reduce the cross-node communication, and performs per-step intra-node adjustment to rebalance the workloads. Subsequently, we develop a communication-aware task assignment method that routes tokens to fewer remote nodes based on the current expert layout. Experiments on 32 B200 GPUs show that Cobalt achieves up to 1.53-2.41 times (1.28-1.89 times on average) of speedup compared to existing MoE training frameworks, while reducing cross-node token traffic by 75.74%-99.26%.

Repeated verifier calls are useful only when they contribute conditional information. We introduce VStress, an auditable replay contract, and VStress-CA, a correlation-aware allocation policy that estimates the conditional marginal information of an unqueried verifier on a sealed calibration split, discounts uncertainty, normalizes by call cost, and stops or abstains when the next call is not informative. The controller freezes its decision and cost ledger before joining the clean oracle; a dependence-shift alarm disables channel preference and falls back to exact-stop. The controlled audit gives the mechanism boundary: at 35% symmetric corruption, majority-5 improves balanced accuracy from 0.6578 to 0.7739, whereas at 65% it loses 0.1226 points. In the matched fixed-budget comparison, breadth, redundancy, and adaptive allocation obtain balanced accuracies 0.6048, 0.6375, and 0.6538, with 3.4216 calls per item and an RLVR score of 0.6417 for VStress-CA. Dependence diagnostics also increase from same-model repeats to cross-family channels, with conditional marginal gains of 0.0126, 0.0462, and 0.0913. These measurements turn correlation from a post-hoc warning into an auditable allocation decision.

Direct visual readability and task-information recoverability are different quantities. Failure to decode a target from a fixed observation need not eliminate access to that target through another recovery route. We develop an evaluation perspective that makes the observation, query, target, and available knowledge explicit and measures the overlap between routes' success sets. For information available on the original visible surface under suitable imaging conditions, direct optical recovery reads the target from the image, optionally after restoration; entity-linked recovery uses residual visual evidence to identify the depicted entity and accesses its target through an entity–attribute relation in a specified knowledge resource. Such access can draw on stored knowledge or an external source. A controlled book-cover study instantiates external access with a fixed title–author catalog, comparing optical author recovery with visual title resolution and deterministic lookup under resolution degradation. Entity-linked successes persist across the tested vision–language models, revealing information access beyond the tested direct visual frontier despite substantial differences in absolute performance. A substantial optical-only region remains. These complementary outcomes show why visual degradation should be evaluated through the task information accessible along specified routes and knowledge resources, alongside direct readability.

Manipulating next-token probabilities during generation can bypass the safety alignment of large language models. Existing approaches, however, rely on access to model weights or numerical token probabilities and therefore do not apply to interfaces that return only sampled text. Reconstructing probabilities from sampled outputs offers a possible alternative, but finite sampling produces sparse and noisy estimates, while repeating this process at every generation step incurs substantial query costs. Our empirical observations suggest that large distributional changes along successful jailbreak trajectories are concentrated at a small subset of positions, motivating selective control. We introduce \method{}, a framework for jailbreaking through text-only continuation interfaces that permit repeated sampling and assistant-prefix continuation. Sample-Based Distribution Reconstruction combines sampled outputs with a prior over unobserved actions to obtain a usable control signal. Risk-Gated Residual Control uses the evolving response prefix to decide when to reconstruct and modify the distribution, concentrating sampling costs at selected positions. Speculative Multi-Token Execution further amortizes target calls by verifying and accepting draft prefixes that require no intervention. Across four target endpoints and three benchmarks, \method{} achieves the highest mean score most comparisons against baselines.

At the nanoscale, observation and feedback may transform thermal fluctuations into valuable resources. Nevertheless, a framework for experimentation with the resulting probability currents has yet to be firmly established. Here, we present an information heat engine incorporating two interacting magnetic skyrmions in adjacent potential wells. Feedback from a skyrmion in Brownian motion at room temperature to the other yields a Maxwell's demon and a circulating probability current in their joint state space. A linear relationship is confirmed between the probability current and the information affinity, the thermodynamic conjugate value, within certain response range and conditions. We identify the slope as the "information conductance" analogous to electrical conductance. The information conductance is modified by skyrmion interactions and scattering, while Kirchhoff's laws remain valid. These results establish a circuit representation of the state space of an information heat engine with potential applications to computing devices, biological systems and chemical reactions.

Distributed inference depends on GPU collective communication that must keep pace with evolving hardware and specialized workloads. However, existing works often couple the semantics, orchestration (where and when data moves), and datapath (how data moves) of collective communication. This coupling makes it costly to adopt new hardware mechanisms and customize communication for applications. We present Purlin, a scale-up communication framework that separates these concerns. At the top of Purlin, we specify collectives as a naming of an input and output layout and a copy or reduction operation. In the middle, we introduce a shared orchestration protocol, Stage, Notify, And Consume (SNAC), which derives coordination from these specifications. Below SNAC sits a hardware-specific datapath we call Atom, which implements two key data movement primitives for collectives: copy and reduce. This separation lets us customize collectives and adopt new hardware mechanisms while reusing orchestration via SNAC. We evaluate Purlin on A100, H200, and B200 GPUs. Across seven collectives, Purlin achieves latency speedups of up to 5.14x and bandwidth improvements of up to 4.50x over baselines. Integrated into SGLang, Purlin improves offline LLM serving throughput and interactivity by 1.13x on average and up to 1.37x over baselines. For online LLM inference, Purlin improves interactivity by 1.26x on average and up to 2.85x, with the largest gain occurring under overload. For diffusion image generation, Purlin reduces end-to-end latency by up to 1.13x.

Production RL for language models lets the sampler fall behind the learner and repairs the resulting mismatch with a truncated importance weight. We ask how long the sampler can go without a refresh under that correction, and find a cliff: on Qwen2.5-Math-1.5B and GSM8K, importance-corrected GRPO refreshed every 192 updates learns well for 180 steps and then degrades severely in all three data seeds before the refresh arrives. Published remedies for staleness act on the update; we act on the sampler instead. Decoupled cooling draws samples at temperature 0.8 while the learner, the reference model and the importance weights stay at temperature 1, with the behaviour probability recorded from the tempered distribution, so the learner's objective is unchanged. All corresponding cooled runs are stable, and the longer interval keeps what the short one delivered: at the same update budget, a cooled sampler refreshed every 192 steps matches an uncooled sampler refreshed every 96 at the end of training (0.857 for both) and averaged over it (0.79), whereas lowering the learning rate to a safe value ends 3-7 points lower. On Qwen2.5-Math-7B the degradation points at interval 192 predict that an interval of 144 is fatal without cooling and survivable with it; on two data seeds the uncooled runs degrade before their first refresh and the cooled runs pass it and end at 92-93% against 68-81%, with one cooled run degrading transiently late in the second cycle. The benefit has a window: at three times the safe interval and in a high-mismatch MATH setting cooling delays degradation without preventing it, stronger cooling is not better, and cooling without the correction collapses. Sampling temperature is a control on staleness tolerance, and temperature and refresh interval should be chosen together.

Large language models (LLMs) excel at token-level generation but may learn undesirable abstract semantics and lack comprehensive perception. LLM-JEPA mitigates this by aligning different views of the same underlying knowledge via a joint-embedding predictive architecture (JEPA). However, strong alignment does not necessarily lead to accurate, stable predictions. To address this, we propose ER-JEPA, which adds an episodic replay path to LLM-JEPA. ER-JEPA stores training pairs in a memory. At each step, it stores and retrieves relevant data to provide additional supervision. This enables learning from both the current batch and stored training pairs, providing additional supervision for token prediction and representation alignment. Experiments across multiple datasets (NL-RX, GSM8K, Spider, and NQ-Open) demonstrate that ER-JEPA consistently outperforms LLM-JEPA.

Dynamic-frame-rate neural speech codecs replace a uniform frame grid with variable-duration tokens, making boundary placement part of the representation itself. Yet it is unclear what these boundaries encode and whether interpretable boundaries are also useful for neural speech reconstruction. This work combines boundary interpretation and controlled reconstruction analysis by comparing predicted boundaries with linguistic and acoustic references. We find that boundary meaning depends on the underlying speech representation. For the ASR-oriented SenseVoice and Whisper encoders, shallow layers emphasize phonetic, voicing, and acoustic transitions, whereas deeper layers shift toward syllable and subword structure. In a comparison of six dynamic-frame-rate algorithms and a uniform (fixed-frame-rate) baseline, higher-level linguistic alignment is associated with lower pooling distortion and better reconstruction from semantic tokens. Frame-rate-matched boundary tests make the distinction concrete: a syllable-derived partition improves over both Uniform and Similarity, while a phoneme-derived partition does not improve over Uniform. We infer that for semantically rich speech representations, useful codec boundaries are best understood as allocation decisions organized around the syllable scale.

Simulation-based inference is challenging when many heterogeneous observations must be composed, hierarchical latent structure must be preserved, and the simulator is misspecified relative to observed data. We develop sampling and fine-tuning methods for diffusion-based inference in design-conditional settings, where the same simulator is queried across different experimental conditions $\xi$. We extend compositional score-based inference with a continuous-time diffusion coefficient that accounts for the number of observations, avoiding Jacobian and auxiliary-covariance corrections. We introduce Hierarchical Blockwise Diffusion Sampling (HBDS), which infers shared parameters and group-specific latent states using a single pretrained model, with the hierarchy specified only at sampling time. Together, these methods support variable observation sets and groupings without retraining. To address misspecification, we introduce path-regularized fine-tuning that adapts the learned likelihood to observations and transfers corrections to posterior inference. Using Girsanov's theorem, we quantify path divergence between pretrained and fine-tuned models across experimental designs and interpret it alongside predictive errors to distinguish candidate misspecification correction from unnecessary adaptation. We evaluate compositional sampling on exact-score Gaussian and Simple Likelihood, Complex Posterior benchmarks, HBDS with analytic and learned scores on a controlled hierarchical model, and fine-tuning and localization on a separate analytic model with known design-dependent discrepancy. Finally, we apply the framework to 940 measurements across four cell lines in a mechanistic Bone Morphogenetic Protein signaling model, where fine-tuning improves posterior-predictive accuracy relative to the pretrained model and shifts posterior marginals toward the least-squares reference while retaining spread.

We attempt to measure the weak lensing of the Lyman-$\alpha$ forest by projecting a quadratic estimator of the forest-forest and quasar-forest pair correlations, built from the DESI DR1 data, onto deflection templates constructed from low-redshift galaxies and quasars from DESI and BOSS. After validating deflection templates against the ACT DR6 and Planck PR4 CMB lensing maps and testing the estimator with injected displacements we measure the lensing amplitude $A_L$ on the data. Consistent with predictions in the literature, the current generation of public data does not have sufficient signal-to-noise to measure this effect unequivocally: we find the lensing amplitude $A_L=0.42\pm0.43$ from the forest auto-correlation and $0.87\pm0.48$ from the quasar–forest cross-correlation. These two nearly independent measurements combine to $A_L=0.62\pm0.33$, consistent with both the fiducial value $A_L=1$ and with no lensing.

The bilevel program is an optimization problem in which the constraint involves solutions to a parametric optimization problem. It is well known that the value function reformulation provides an equivalent single-level optimization problem, but it results in a nonsmooth optimization problem that never satisfies the usual constraint qualification, such as the Mangasarian–Fromovitz constraint qualification (MFCQ). In this paper, we study necessary optimality conditions for nonsmooth bilevel programming using tangential subdifferentials. We derive a sequential enhanced tangential Fritz–John type condition for a class of nonsmooth nonlinear programs and then obtain corresponding enhanced tangential Fritz–John and enhanced tangential Karush–Kuhn–Tucker (KKT) systems by taking limits of the sequential condition. We further investigate tangential constraint qualification conditions, including T-quasinormality and a tangential cone–continuity property (T–CCP), and analyze how they guarantee the validity of tangential KKT type conditions. Finally, these results are applied to the value-function reformulation of a bilevel programming problem.

In molecular machine learning (ML), significant progress has been made on integrating fundamental thermodynamic relations into ML models. Such physics-informed approaches can be categorized as hard-constrained, where the thermodynamic relations are embedded in the model architecture, and soft-constrained, where relations are included in the training loss. While both approaches have been applied to various property prediction tasks, a systematic comparison on experimental data is lacking. We herein compare hard- and soft-constrained approaches on a numerically challenging case study: the prediction of vapor pressure, saturated liquid and vapor molar volumes, and enthalpy of vaporization as functions of temperature for single-species vapor-liquid equilibrium, related by the exact Clapeyron equation. Furthermore, soft-constrained approaches in molecular ML typically balance data and physics losses with a fixed weighting factor, requiring careful tuning. To overcome this, we investigate the augmented Lagrangian method (ALM), well established in constrained optimization. We find that for the soft-constrained approach, using the ALM improves prediction performance in terms of constraint satisfaction by factor 2 and reduces training epochs compared to using a fixed penalty by 30%. Hence, in the soft-constrained approach, the use of the ALM is always recommended. The hard-constrained approach, which entails additional architectural design choices, can reach on par accuracy with the soft-constrained approach, at even higher thermodynamic consistency, reaching 12 orders of magnitude closer approximation of the Clapeyron equation. Overall, the hard-constrained approach is most promising when thermodynamic consistency is critical, yet requires tailoring the ML model architecture to the respective thermodynamic relation.