Papers

German is a morphologically rich language whose syllable structure is exceptionally well-predicted by the Knuth–Liang hyphenation algorithm. We ask whether phonologically informed tokenization can serve as a competitive target for end-to-end speech recognition. We compare three tokenizer families on the Omnilingual ASR wav2vec 2.0 backbone fine-tuned with CTC: the pretrained multilingual character inventory, a data-driven Byte-Pair Encoding (BPE) over orthography, and phonologically informed units from Pyphen syllabification and grapheme-to-phoneme conversion. Across 40 fine-tunes, we evaluate on three German test sets spanning orthogonal shifts: in-domain read speech, dialectal spontaneous speech, and standard-German spontaneous speech. In-domain, all phonologically informed tokenizers match BPE and the multilingual character baseline on both WER and CER. Under domain shift the picture splits along vocabulary size rather than the linguistic axis of variation: at small vocabularies, syllable-aware tokenization improves on dialectal speech, where phonetic surface forms vary but syllable structure is preserved, and stays ahead on spontaneous speech, where new word-forms violate vocabulary closure. A phoneme-level confusion analysis further shows that all tokenizers commit the same canonical function-word errors, indicating that the acoustic encoder, not the tokenizer, dominates the error topology. Our findings suggest that tokenizer choice may depend on the vocabulary budget as much as on the distribution shift expected at deployment rather than reducing to a single universal optimum.

As orbital populations grow, conjunction screening must search more object pairs for potential collision risks. Existing pipelines use early-stage filters, such as an apsis filter, to eliminate pairs whose orbital altitude ranges cannot produce a close approach. However, these filters apply the same fixed threshold to every object, making the screening unnecessarily conservative for well-constrained objects while potentially insufficient for objects with greater orbit-prediction uncertainty. In this work, we present an uncertainty-aware conjunction screening method that replaces a single fixed apsis threshold with object-specific thresholds derived from propagated radial uncertainty. The resulting thresholds vary across both objects and prediction time, becoming narrower when propagated uncertainty is small and wider when uncertainty grows. We pair this adaptive screening rule with a binary-search altitude sweep that makes resulting heterogeneous intervals efficient to evaluate at catalog scale with log-linear search complexity. We evaluate the method on 33,791 objects from the Space-Track General Perturbations (GP) catalog using the preceding eight public ephemeris snapshots and a seven-day prediction horizon. The adaptive sieve retains 162.8 million candidate pairs, reducing the candidate set by 28.78% relative to a fixed 50 km apsis threshold and by 71.48% relative to brute-force all-pairs screening. It retains all close approaches identified in two independently propagated reference sets. At the largest tested catalog size, the binary-search sweep is 27 times faster than exhaustive interval evaluation while returning the same candidate set. These results demonstrate that uncertainty-aware screening can substantially reduce the candidate burden passed to downstream conjunction analysis, providing a scalable path toward managing increasingly large and heterogeneous orbital populations.

Quantum optical sensing has evolved from isolated solid state spin defect platforms to increasingly bio-integrated systems designed for biomedical applications. This review presents a perspective on this transition, spanning nitrogen vacancy centers in diamond to emerging fluorescent protein based spin qubits. Advances in diamond fabrication have enabled photonic cavities, nanomechanical resonators, and microstructured devices that enhance spin photon interactions and room temperature sensitivity. Exploration of spin defects in alternative materials further expands the sensing landscape. Parallel progress toward biological compatibility includes chip based architectures for in vitro and in vivo studies, microfluidic integration, and minimally invasive nanodiamond probes capable of intracellular nanothermometry and nanorheometry. Improvements in surface functionalization have enhanced coherence and sensing reliability in complex environments. Collectively, these advances highlight the progress and remaining challenges toward application oriented quantum sensing for the life sciences.

The electronic properties of amorphous silicon nitride vary with processing, which often changes stoichiometry and density together. Separating their contributions is essential to connect film structure with electron trapping. Across 499 stoichiometric configurations spanning 2.86-3.19g cm$^{-3}$, porosity controls the trapping mixture. Lower density increases accessible void fraction from 8.4 to 12.6%, expands internal surface and more than triples the supply of free three-coordinate Si precursors. Capture at pre-existing Si dangling bonds (K centres) rises from 9 to 29%, while induced-K and polaronic trapping decline. A neutral precursor fingerprint predicts the trapping mixture in held-out configurations and reproduces its density dependence. Local compensation controls K-centre activation, electronic competition selects capture, and bond strain predicts polaronic relaxation or induced-K formation. Routes involve distinct network volumes but retain their characteristic depths and relaxation energies across the density range. At fixed stoichiometry, density controls the trapping mixture through porosity and precursor availability.

Erdős and Sós conjectured that every $n$-vertex $3$-uniform hypergraph whose link graphs are all bipartite has at most $(1/4+o(1))\binom n3$ edges. We disprove this conjecture by constructing counterexamples with edge density at least $0.250000356>1/4$ for all sufficiently large $n$.

Handcrafted minutiae detection algorithms remain fundamental to biometric science and forensic practice due to their full auditability, adherence to international standards, and operational independence from training datasets or GPU hardware. However, current open-source traditional baselines are severely outdated, relying almost exclusively on legacy C/C++ codebases that lack seamless integration with modern scientific software ecosystems. To bridge this gap, the present work introduces SBMEX (Skeleton-Based Minutiae EXtraction), a fast and deterministic minutiae detection method integrated into the open source pyfing package. SBMEX achieves high computational throughput by employing a dual Look-Up Table architecture that replaces runtime neighborhood scanning during Crossing Number computation and skeleton tracking. Additionally, it incorporates a continuous quality scoring framework driven by tracking path length, dual ridge-valley skeleton fusion, and spatial density decay. Rigorous evaluation on NIST SD302 datasets demonstrates that SBMEX delivers feature extraction accuracy comparable to or outperforming traditional open-source baselines without fine-tuning, while achieving a drastic reduction in minutiae detection latency relative to classical Crossing Number Python implementations.

Botnets represent the backbone of most modern cybersecurity attacks through which botmasters deploy distributed denial of service and advanced persistent threats against legitimate computer networks. In this paper we introduce the Host Attack Graph model and propose two botnet propagation strategies which we study across network topologies, target selection, and attack strategies together with their effectiveness in time. While strategy and botnet size are important for propagation, our study and simulations show that network topology and target selection should not be easily discarded. Implementation and benchmarks are available at https://codeberg.org/AndreiN/HAGP.

In this paper, we study $r$-neutralized Pfister–Sullivan entropy, defined by measuring the complexity of orbit segments whose empirical measures lie near a prescribed invariant measure. We mainly obtain the following three results: (i) We disprove a variational conjecture posed by C.~Dong and Q.~Qiao [On $r$-neutralized entropy: entropy formula and existence of measures attaining the supremum, Comm. Math. Phys. 406 (2025), Article No. 74.] (ii)] We obtain variational formulas over probability measures for both upper and lower $r$-neutralized topological entropies. We also introduce new measure-theoretic quantities through PS localization and establish some variational principles for upper $r$-neutralized topological entropy (iii) We obtain upper bounds for $r$-neutralized PS entropy under different smoothness assumptions and lower bounds for $C^{1+\alpha}$ diffeomorphisms, in terms of entropy, dimension and Lyapunov exponents. For $C^\infty$ diffeomorphisms and $C^{1}$ Anosov diffeomorphisms, the $r$-neutralized PS entropies equal $h_\mu(f)+r\dim M$ for every ergodic measure $\mu$. We also establish exact formulas on mixing surface hyperbolic basic sets and show that the whole-manifold measure formula can fail for $C^2$ surface diffeomorphisms.

Evaluating user experience (UX) with automated computational methods has gained increasing attention, supported by empirical evidence from UXBench. However, binary preference prediction provides limited insight, while relying on a single user-agnostic reward model overlooks the inherent heterogeneity of users, whose expectations can differ substantially. In this paper, we present UXBench Pro, comprising 1{,}000 test instances derived from real user interactions across 12 task scenarios and 82 domains. Each instance is paired with a FACTORS user profile that characterizes the user through seven interpretable behavioral facets, differentiating user groups. To provide richer evaluation insights, we introduce a dual-perspective paradigm that combines a personalized User Reward Model (URM) for third-person judgment with Sim4Eval, a user simulator that enables multi-turn interactions and provides first-person evaluation across four cognitive state dimensions. To assess the reliability of these based evaluators, we further introduce two meta-benchmarks, URMBench and USimBench, that evaluate how faithfully they reproduce real human preferences and behaviors. Extensive experiments reveal seven key findings that highlight the importance of user modeling and multi-perspective evaluation, offering a fresh perspective on user-centric benchmarking and motivating personalized model optimization.

Gravitational microlensing is a unique planet-detection method that does not rely on detecting light from the host star or planet. Instead, planets are discovered when their gravitational field perturbs the light coming from a distant source. The mechanics of this method offer several advantages. It has the ability to detect wide-orbit planets of various masses around host stars of different types, at and beyond the system's snow line. Thus, microlensing complements the other detecting techniques that are more sensitive to close-orbit or high-mass planets. In this Chapter, we review the microlensing method and discuss how the physical properties, such as mass and distance, can be extracted from the microlensing light curve or from additional high-angular resolution observations. Microlensing is essential for the broader context of exoplanet demographics, as it can place constraints on planets, in wide orbits, beyond the snow line. Studying this parameter space can help us unravel the details of planet formation and how our own Solar System formed. Lastly, we provide an example of how to model a light curve and obtain accurate physical parameters for the host and planet.

We propose a symmetric Nitsche method for the biharmonic Dirichlet problem on a smooth domain whose boundary may cut arbitrarily through the computational mesh. Bulk least-squares stabilization in a boundary strip and tangential boundary penalties control the consistency terms through an explicit lifting of the two Dirichlet traces. This gives coercivity on $H^4(\Omega)$ without inverse inequalities and permits a fixed penalty independent of the cut configuration for sufficiently thin strips. Under suitable regularity assumptions, splines of degree $p\geq4$ give energy and $L^2$ error bounds of orders $h^{p-1}$ and $h^{p+1}$. We also analyze a face-stabilized extension to $C^1$ splines of degree $p=2$ and $C^2$ splines of degree $p=3$. Exact-ellipse experiments examine convergence, cut stability, and parameter sensitivity.

We study dielectric properties of holographic insulators from the quantum geometry of the ground state. We realize the insulating state by the pseudoscalar interaction of holographic fermions and calculate the fermionic Green function to extract the effective hamiltonian. We then derive the quantum metric and Berry curvature to compute the transports and finally get the dielectric constant from the imaginary part of conductivity.

Large language models (LLMs) perform remarkably well on complex tasks, yet remain highly vulnerable to prompt injection attacks, where malicious instructions embedded in external data can override user intent. Existing defenses remain limited by model fine-tuning requirements, vulnerability to adaptive attacks, or reliance on brittle handcrafted prompts. We argue that a fundamental source of this vulnerability is the lack of an explicit representation of trust provenance. To address this, we introduce Learnable Trust-Boundary Delimiters (LTBD), a lightweight defense that explicitly encodes trust boundaries in the input while keeping the LLM parameters unchanged. LTBD uses a small number of learnable delimiters to distinguish trusted user instructions from untrusted external data, enabling the model to better respect the intended trust hierarchy. Experimental results show that LTBD substantially outperforms inference-time defenses and performs competitively with training-based approaches, while preserving benign-task utility and introducing negligible inference overhead. In particular, LTBD achieves 0.00% ASR on AlpacaFarm and only 0.11-0.19% ASR on TaskTracker. LTBD also remains effective under adaptive attacks, where adversaries have full knowledge of the defense and explicitly attempt to bypass it.

Orbital angular momentum (OAM) characterization is central to understanding quantum materials such as valley Hall and Rashba systems and, through its connection to Berry curvature, provides key insight into band topology. Conventional OAM detection via photoemission dichroism, however, is susceptible to interference artifacts and is largely restricted to occupied states. Here, we demonstrate a real-space approach to map the $L_z$ OAM character using scanning tunneling microscopy in triangular atomic monolayers. Unlike photoemission, this method exploits interference effects arising from the $L_z$ orbital phase and the Bloch phase associated with neighboring atoms. This shifts the local density of states (LDOS) maxima to distinct Wyckoff positions between atoms, encoding $L_z$ information into characteristic real-space LDOS patterns. We exemplify this approach in 2D quantum materials with contrasting $L_z$ sequences, namely Tl/Si(111) and In/SiC(0001) monolayers, identifying $L_z$-dependent LDOS accumulation associated with distinct topological phases in both systems. Our results establish $L_z$-dependent charge localization as a proxy to atomic obstruction for energetically isolated states in triangular lattices, including transition-metal dichalcogenides.

As electric vehicle volume increase, the interest for reusing their retired batteries grow. Giving a second life to batteries reduce their environmental impact and provide access to more affordable energy storage systems. This lifespan extension is of interest in a wide range of fields as the reuse offer a solution for end-of-life batteries management and the development of renewable energies. Although several review articles have described opportunities and barriers related to reuse, to our knowledge none have synthesized insights from studies that conducted experiments on retired batteries. This article aims to address this gap by providing a detailed analysis of more than 150 articles. This article's main findings are the proposition of a new vocabulary for batteries end-of-life, an evaluation of state of health and dispersion in retired batteries as well as the synthesis of aging and safety studies on these batteries. In a nutshell, this review show that the performance of retired batteries can be high, their dispersion should be managed and their degradation can be slow and safe. Limitations of current battery research ideas for future experiments and ideas for future studies are also presented.

Deploying intelligent robotic systems that interact with humans through gestures requires neural networks capable of recognizing diverse temporal patterns. We present a systematic benchmark of ten abstract sequential tasks–five permutation-invariant (set) and five order-dependent (sequence) problems–evaluated across eighteen neural network architectures spanning recurrent, convolutional, attention-based, and set-function families. Beyond the core architecture-task grid, we explore numerous preprocessing and target-variable transformations, yielding more than 250 distinct experimental configurations. All variants are trained and tested under strictly identical conditions (fixed random seeds, shared hyperparameters, shared data splits) to ensure fair and reproducible comparison. Ranking across all ten tasks reveals four consistently top-performing architectures–BiGRU, TCN, Conv1D, and GRUReLU–all compact enough for real-time deployment (under 2,000 parameters in the benchmark setting). Based on this ranking, we apply three architecturally diverse top models (BiGRU, TCN, and GRUReLU) to a practical robotics problem: estimating the execution speed of dynamic arm gestures from skeletal keypoint sequences. Three speed interpretations (peak count, period time, and mean spike spacing) are evaluated on a custom dataset of eight traffic-related gesture classes comprising 256,710 frames recorded via OpenPose. The best configuration achieves a mean absolute error of 0.198 on the peak-count interpretation, corresponding to roughly 5% relative error, while the period-time interpretation reaches approximately 4% relative error, and the mean spike spacing interpretation approximately 8% relative error. These results demonstrate that neural networks can reliably estimate gesture speed from skeletal data, opening a path toward speed-aware gesture-controlled robotic systems.

We have employed Bayesian inference to extract the centroid energies and widths from the isovector giant dipole resonance (IVGDR) data for the IVGDR built on the ground states of nuclei.These IVGDR properties were modeled by explicitly incorporating the structural effects like isospin asymmetry, nuclear deformation and shell correction. The resulting parametrizations reproduce the IVGDR systematics with remarkable precision for both spherical and axially deformed nuclei. Subsequently, the extracted IVGDR parameters were used to constrain the coefficients of the nuclear symmetry energy, which are found to be in good agreement with recent measurements of nuclear dipole polarizability. Furthermore, these symmetry-energy coefficients were employed to predict the dipole polarizability within different self-consistent mean-field models and to investigate its correlation with the neutron-skin thickness. The present findings provide valuable input for large-scale calculations of radiative capture processes relevant to nuclear astrophysics and for improving our understanding of astrophysical scenarios under extreme conditions.

We investigate spin-outcome correlations and memory-assisted entropic uncertainty in tree-level Bhabha scattering, where an incident electron $A$ scatters with a positron $B$ initially entangled with a spectator electron $C$. For the fixed-angle post-selected final state, we compare the total memory-assisted uncertainty $U_A$ for three alternative Pauli measurements on $A$ with the entanglement entropy $S(\rho_A)$ across the $A|BC$ partition. We resolve $U_A$ into the separate uncertainties $U_B$ and $U_C$ and evaluate the corresponding outcome-memory mutual-information sums $G_B$ and $G_C$. This separates changes in the randomness of the outcomes on $A$ from changes in their correlations with each memory. In the parameter regions studied, enhanced $A|BC$ entanglement is accompanied by lower $U_A$, most prominently near high-energy backscattering. An analytic treatment of this limit shows that $B$ factorizes from $AC$: $G_B$ vanishes, whereas $G_C$ increases with the initial $B$-$C$ entanglement. An entropic complete complementarity analysis characterizes the accompanying loss of local helicity predictability in $\rho_A$ and changes in the coherence, predictability, and mixedness of the interacting pair $\rho_{AB}$. These results show that initial $B$-$C$ entanglement can lower the total memory-assisted uncertainty in high-energy backscattering by strengthening spin-outcome correlations with the spectator $C$, thereby identifying the distinct roles of the interacting particle and the spectator as quantum memories.

Machine unlearning updates a trained model after records are deleted, aiming to match exact retraining without repeating the full training procedure. We develop Gaussian mechanisms for Newton updates under sequential deletion requests. Using Gaussian differential privacy (GDP) and its adaptive composition rule, we show that the full sequence of released models is statistically difficult to distinguish from matched exact retraining. To calibrate these mechanisms for empirical risk minimization, we derive upper bounds on the error of the Newton approximation relative to exact retraining and on how this error changes after each deletion batch. Independent Gaussian noise is calibrated using bounds on the full residual at each release, whereas Gaussian random walk noise uses smaller bounds on residual increments. These bounds yield allocations minimizing the worst-case maximum noise variance across releases under the resulting GDP certification constraints. With count-based bounds, the random walk asymptotically matches the worst-case variance of a single release at deletion cap $M$, while independent noise incurs an additional factor of order $M$. Set-based bounds can reduce the noise variances by using gradients and Hessians of the deleted records. For singleton deletion, we further show that count-based independent noise, count-based random walk noise, and set-based independent noise are minimax among fixed Gaussian covariances under their respective residual or increment bounds. With set-based bounds, allowing variances to adapt to deleted records can improve on every fixed covariance by a factor of order $(\log M)^2$ on some data sequences. The residual and noise bounds also yield parameter and predictive consistency relative to exact retraining, uniformly over deletion policies. Simulations and a credit default data analysis evaluate bounds, noise variances, and estimation errors.

Repairing scientific models from functional observations differs fundamentally from supervised prediction: feedback may certify a solution without revealing which structural correction is responsible. We study this setting for genome-scale metabolic model (GEM) repair, where multiple reaction edits can explain the same phenotypes and many apparently distinct edits correspond to the same biological mechanism. This many-to-one structure creates a hidden failure mode for conventional exploration: diversity in the output space need not translate into diversity of scientific hypotheses. We introduce QuotientPO, which collapses equivalent repairs into canonical mechanisms and optimizes exploration directly over the resulting quotient space. To make quotient exploration informative under finite rollouts, we derive a kernelized Rényi estimator that resolves graded crowding among distinct repair cores beyond coarse exact-match counts. On 2,212 held-out GEMs, QuotientPO improves Success@32 from 17.93% to 20.10% (+12.1% relative) while consistently increasing distinct successful-core discovery under the same sampling budget. These results establish quotient-space exploration as a principled approach to mechanism-level discovery under verifier-induced equivalence.

We study FLRW-compatible tensor fields on Friedmann–Lemaître–Robertson–Walker spacetimes in $(D+1)$ dimensions equipped with Levi-Civita connection. We prove that any covariant rank-$N$ tensor field invariant under the spatial homogeneity and isotropy group, with $D>N$, admits a finite decomposition into monomials constructed from the metric tensor $g_{\mu\nu}$ and the unit timelike vector $u_\mu$, with scalar coefficients depending only on cosmic time. As a direct consequence, any rank-two correction term appearing in an Einstein-like rearrangement of the FLRW background field equations has the effective perfect-fluid form for $D>2$. Thus, the result provides an arbitrary-rank tensorial classification underlying the perfect-fluid structure of modified gravity equations on FLRW backgrounds, complementing existing rank-two universality results. Applications to generalized Friedmann equations and to the geodesic deviation equation in modified gravity are also discussed.

A watermark detector can flag a text even when only a small fraction of its tokens are watermarked. We study sequential auditing to identify local passages whose watermark proportions exceed a prespecified threshold. Each alarm fixes a candidate passage's endpoint, while its start is chosen from earlier tokens. Reusing the same evidence to locate and test a passage can invalidate inference. We propose FissionGuard, which uses data fission to construct an alarm value and a validation value from each token-level $p$-value. The alarm values guide alarms and localization. The validation values yield passage-level $p$-values through partial conjunction, excluding as many of the strongest signals as the proportion null allows. These $p$-values are used to test the corresponding hypotheses at levels assigned by an online multiple-testing procedure. Under the stated watermark generation model, allowing dependence among token-level $p$-values at watermarked positions, we prove finite-sample false discovery rate (FDR) control among reported passages at every fixed finite horizon. We also derive a finite-sample lower bound on power for testing nonnull candidates. Experiments on synthetic streams and Mistral and Qwen generations across four marking layouts show gains in watermarked-token coverage over sample splitting, with empirical FDR below the target.

Accurate and computationally efficient evaluation of non-linear interference (NLI) remains challenging in multi-band WDM systems comprising heterogeneous spans and nonidentical channels. In such systems, channels may have distinct spatial power profiles and multi-subcarrier structures. Spans may use different fiber types with span-dependent dispersion and loss profiles. These systems may further employ general amplification schemes, including forward and backward Raman amplification. Inter-channel stimulated Raman scattering and multiple lumped gain or loss elements may also be present. In this encompassing framework, we provide an efficient decomposition of the GN-model NLI contributions. We show that the NLI power spectral density at any frequency across the WDM comb is fundamentally driven by the spatial auto- and cross-correlations of the channels' spatial power profiles or, equivalently, by the corresponding energy spectra. For this reason we call this formalism the `correlation GN-model', or cGN. Once these correlations or spectra have been computed, the remaining calculations reduces to a fast, well-behaved one-dimensional numerical integral. Moreover, cGN fully accounts for coherent interference of NLI in every pair of spans. Such coherence is often neglected or approximated in existing analytical and semi-analytical GN-model formulations, although it can increase the NLI by several dB. In short, the cGN framework applies to any general systems over any band and any bandwidth, with very high computational efficiency. It is capable of providing the frequency spectrum of NLI at multiple frequencies across each channel or subcarrier. Its accuracy is very high because virtually no approximations are invoked. The cGN framework is available as freely downloadable software from the European Union Zenodo website.

We propose an analogue of the no particle production condition in two-dimensional anti-de Sitter space, and demonstrate how this criterion uniquely reconstructs the curved-space Liouville and Toda theories via general-multiplicity holographic correlators.

Traversability is essential for visual navigation but varies with robot capabilities and user preferences. Conventional pipelines often rely on explicit costmaps or segmentation masks with predefined criteria, requiring hand-crafted rules and careful tuning. Moreover, viewpoint-dependent segmentation masks complicate asynchronous planning under perception latency. We present LaTraNav, a framework that learns language-conditioned traversability representations for adaptive visual navigation. Its asynchronous architecture combines a slow vision-language model that produces latent representations of traversability and navigation goals, with a fast flow-matching planner conditioned on these representations. To train the system, we develop a simulation-based data generation pipeline with controllable trajectories, producing observations paired with language instructions, traversability maps, goal locations, and diverse trajectories. Photorealistic image translation further enhances visual realism. Evaluations on datasets from multiple sources demonstrate effective language-guided traversability segmentation and goal localization by the slow VLM, alongside adaptive pixel-space path planning by the fast planner. Latent conditioning improves planning performance over explicit segmentation masks, while asynchronous scheduling increases the path-update rate by $6.05\times$ at the same semantic-update rate.

Consensus can make pseudo-label learning more reliable, but only when its predictors contribute genuinely different evidence. Multiple models that repeat the same boundary provide additional votes without additional information. We address this problem by con structing decision sources through controlled changes to within-class structure. Starting from a shared graph representation, we vary center granularity and neighborhood mixing, reproduce each resulting source to test its stability, and select a complementary subset using node pair coassignment. Unanimous predictions from the selected sources are then ranked for student training. On the public fixed splits of Cora, CiteSeer, and PubMed, evaluated with five random seeds, the constructed sources improve fixed-budget training pseudo-label precision by 1.19 to 4.39 percentage points over three conventionally initialized GCN sources. Under matched structural filters, three-source consensus is more precise than each constituent source in all 45 dataset slot eed comparisons. The gains are strongest in pseudo-label quality: downstream accuracy remains competitive but does not lead on every dataset. These results identify source construction rather than model count alone as an important design problem for consensus-based pseudo-label learning.

Parameter-efficient fine-tuning (PEFT) enables large language models (LLMs) to adapt to specialized tasks, but often at the cost of degrading general capabilities acquired during pretraining. Existing approaches primarily mitigate this trade-off through data replay or regularization, relying on additional data or explicit optimization constraints. We instead focus on a different question: where should adaptation be applied? We find that fine-tuning different Transformer layers produces different target-task gains and degrees of capability degradation, suggesting that not all layers are equally suitable for adaptation. To characterize this difference, we use layer-wise empirical Fisher information to measure target-task sensitivity. However, computing Fisher scores requires backward computation and becomes increasingly expensive for large models. We therefore introduce input–output cosine similarity as a lightweight, forward-only proxy for ranking layer sensitivity. Across models and tasks, layers with lower input–output similarity consistently exhibit higher empirical Fisher scores. Building on this observation, we propose Layer-Selective LoRA (LS-LoRA), which places trainable LoRA adapters only in layers with low input–output similarity. Experiments on mathematical reasoning and code generation show that LS-LoRA improves average target-task performance while retaining substantially more commonsense reasoning capability than standard all-layer LoRA, demonstrating that carefully choosing where to adapt can provide a simple and effective way to balance target-task adaptation and general capability retention.

We develop a model-free policy gradient method for discrete-time mean-field control (MFC). In MFC, the policy affects the objective both through the controlled dynamics and through the population distribution. Standard REINFORCE estimators capture the first effect but not the second. We introduce Transport REINFORCE, a transport map-based approach that perturbs a suitable transformation of the population distribution to estimate this missing mean-field contribution. The method applies to both finite and continuous state spaces. In finite state spaces, we perturb the population distribution directly on the probability simplex through a convex combination of the current population weights and random weights. In continuous state spaces, we project the population distribution onto the manifold of Gaussian mixtures, and then randomize it via a transport map that ensures the perturbed law remains within this manifold. We prove consistency of the perturbed objective and gradient as the perturbation vanishes, and derive bias and mean-square error bounds for the resulting sample-based gradient estimator. Numerical experiments on several MFC benchmarks show that Transport REINFORCE improves over standard REINFORCE.

As large language models increasingly move toward repository-level software engineering, existing code-generation benchmarks remain fragmented across language coverage, task granularity, and evaluation protocols, impeding systematic comparison. To address this gap, we present PolyCodeEval, a unified multilingual and multi-granularity benchmark for code generation. It comprises 2,590 code generation tasks spanning functions to repositories, derived from 58 real, executable open-source repositories in five programming languages. All tasks are evaluated under a unified execution-based protocol with integration procedures tailored to their generation targets. Building on this benchmark, we evaluate frontier large language models, state-of-the-art specialized methods, and general coding agents. Our results show that existing approaches still struggle to correctly generate complete code fragments across granularities and languages. Specifically, the studied methods generate at most 71.7%, 76.7%, and 31.0% correct functions, files, and repositories, respectively, with performance varying widely across languages. Paired experiments further show that implementation context from related functions in the same file improves the executable correctness of function generation. Method rankings also vary across task granularities and programming languages, highlighting the importance of multilingual, multi-granularity evaluation for comprehensively assessing code generation capabilities.

Knowledge distillation enables efficient spatiotemporal prediction by transferring knowledge from an accurate teacher to a compact student. However, matching outputs or features independently for each sample leaves cross-sample predictive structure underused. Exploiting this structure requires representations and historical references that reflect the dynamics of each task. We propose TAM, a Task-Aware Memory Distillation framework that organizes a frozen teacher's knowledge into a bounded, retrievable history. Memory entries encode latent features, forecast changes, or flow residuals, while task-specific selection rules identify relevant historical references. The student either matches the teacher's similarity distribution over shared references or regresses observation-conditioned residual prototypes. These objectives complement supervised prediction and conventional distillation. The teacher, memory, and auxiliary adapters are used only during training, leaving student inference unchanged. We evaluate TAM on video prediction, weather forecasting, and traffic flow prediction across multiple teacher-student configurations. Averaged over four paired runs, adding TAM improves SSIM on all six video datasets and reduces MSE on five relative to the corresponding KD baselines. Mean paired MSE reductions reach 1.86% on KittiCaltech, 1.93% on WeatherBench with a gSTA teacher, and 1.01% on TaxiBJ. These results demonstrate the utility of historical teacher supervision across distinct forecasting tasks without additional student inference cost.