Papers

Mixture-of-Experts (MoE) inference is increasingly deployed in local and on-premise environments, where expert parameters often exceed GPU memory capacity. In latency-sensitive, low-concurrency settings, repeatedly staging routed-expert weights from CPU memory to the GPU can be prohibitive, leaving routed-expert feed-forward networks (FFNs) on the critical path of multi-socket CPUs. Existing CPU accelerations often rely on intrusive, hardware- or topology-specific requirements, such as AMX-specific weight layouts or manual NUMA-aware placement. These requirements reduce portability and complicate integration with standard CPU-GPU offloading pipelines. We present HiNa-MoE, a high-performance, non-intrusive operator library for MoE inference on CPUs with Intel AMX. HiNa-MoE (1) exploits AMX with an optimized micro-kernel that keeps expert weights in standard layouts and instead fuses lightweight layout transforms into token gathering and stores; (2) applies NUMA-aware task partitioning under a simple page-interleaved policy without modifying the framework allocator; and (3) converts decode-phase memory matrix-vector operations into small matrix-matrix execution to utilize AMX. Across multiple MoE models, HiNa-MoE achieves up to 3.37x speedup for FFN kernels and up to 2.09x end-to-end inference speedup over state-of-the-art baselines, while remaining plug-and-play with existing frameworks and deployment workflows.

In recent years, Unmanned Aerial Vehicles (UAVs) or drones have been increasingly adopted in urban environments for applications such as parcel delivery, infrastructure inspection, emergency response, and drone light shows. A non-negligible issue is how to manage the increasing number of drones operated by different entities, to enable them to cooperatively avoid potential collisions and determine conflict-free short-term flight paths in urban airspace. This paper presents a Prediction-Based Short-Term Decentralized Greedy (PB-STDG) guidance algorithm for a structured Drone Road System (DRS). PB-STDG extends the original STDG algorithm by introducing a prediction mechanism that enables drones to anticipate the decisions of neighboring drones using additional information shared in beacon packets, aiming to address the over-conservative behavior observed in the STDG algorithm. Several simulation scenarios are conducted to evaluate and compare the proposed algorithm with STDG. The results show that PB-STDG improves traffic efficiency while maintaining a safety level comparable to that of STDG.

We present two complementary neural routes for microscopic neutron-star EOS inference: truncated sequential neural posterior estimation with mixture importance sampling (TSNPE+MIS), which targets one observation, and amortized neural posterior estimation with importance sampling (A-NET+IS), which is reused across nuclear inputs and NICER sources. Both are corrected against the original prior and full likelihood, with the effective sample size as a diagnostic. We test nucleonic and hyperonic relativistic mean-field models with nuclear data, NICER, GW170817, and pQCD. For the fiducial analyses, both agree with UltraNest in the parameter posteriors and 90% mass-radius and tidal-deformability bands. TSNPE+MIS band-edge differences are only 0.024-0.033 km and 1.02-1.03%, while A-NET+IS differences are 0.019-0.050 km and 0.86-1.13%. TSNPE+MIS reduces the computing time by factors of 1.6-2.3; after training, A-NET+IS is 15-240 times faster per configuration. The frozen A-NET also analyzes three NICER sources absent from training, including the newly reported high-mass, compact PSR J1614-2230 posterior. For this source, the inferred $R_{1.4}$ agrees with UltraNest within 0.010 km for nucleonic matter and 0.009 km for hyperonic matter. We also introduce a Green-function Evidence Network (EN), a regression-based variant of Evidence Networks and, to our knowledge, the first Evidence Network applied to neutron-star EOS inference. After a one-time cost, its network queries return absolute evidences in about 0.03 s for new nuclear data and 3.5 s for a new NICER source; all 16 differences from UltraNest are below $1\sigma$, and independent importance-sampling evidences agree within the quoted EN uncertainties. Under the adopted likelihood, both evidence calculations favor nucleonic matter by Bayes factors of about 22:1-27:1.

We give a short and simple proof that an anisotropic generalized quadratic form over a quaternion division algebra with involution of the first kind, in arbitrary characteristic, remains anisotropic after scalar extension to any odd-degree field extension.

Deep search agents tackle complex knowledge tasks through iterative retrieval, multi-hop reasoning, and evidence synthesis across multiple sources. Existing approaches typically assume relatively stable retrieval systems and operate over short-horizon tool interaction. However, when retrieval is sensitive to query formulation, even a semantically appropriate query may fail to surface critical evidence because of mismatched entity names, aliases, or keyword combinations. Recovering from such failures requires repeated query reformulation and longer interaction trajectories. This setting poses a distinct training challenge, as the policy must sustain long-horizon query exploration while managing an expanding volume of retrieved content. We propose LexiHorizon, a framework for training search agents over long horizons that expands the trajectory context budget, manages accumulated retrieval content using a window over recent tool observations while preserving the reasoning history, and introduces an outcome-gated search-effort reward that provides a bounded bonus for tool invocations to trajectories with nonzero answer reward. Experiments on XBench, WebWalkerQA, and BrowseComp-ZH show that the resulting 9B model consistently outperforms both its base model and MiroThinker-1.7-mini, with maximum absolute gains of 8.7 and 23.8 percentage points, respectively. These results suggest that combining an extended context budget with reasoning-preserving context management benefits long-horizon deep search agents.

R Aqr is a resolved symbiotic Mira binary and a prototype target for the study of complex hydrodynamic processes including dust-driven mass loss, mass transfer, accretion, and jet outflows and the formation of strongly asymmetric structures. We investigate for R Aqr the central Halpha emission from the ionized gas and the polarization from scattering dust based on SPHERE/ZIMPOL maps with a resolution of about 25 mas (5 AU). Data for four epochs were taken, including the dust eclipse around 2021, to follow structural changes. The central Halpha gas traces the orbital motion of the hot component. The polarization from scattering dust shows interaction effects near the binary (<60 AU), including a flattened distribution in the orbital plane, corotating dust, and a reduction of the scattering signal during the dust eclipse. The polarization shows at >60 AU a steady outflow with about a dozen dust clouds, a few of which have a comet-like core and tail features. The dust velocity is about 20 km/s, and this is faster than for the CO gas measured with ALMA. Several prominent Halpha clouds also show scattering polarization because they are just partly ionized dust clouds from the Mira wind. We also found small faster-moving transient Halpha clouds, pointing to outflow instabilities and shocks. The observed angular shift of Halpha during the dust eclipse, can be explained by strong illumination changes from the hot component. At larger separation, the H$\alpha$ clouds move faster (>100 km/s), in agreement with previous studies. This acceleration is probably a result of gas heating by photoionization and shocks, and the produced overpressure gives rise to the two well-known jet-like features expanding in polar directions. The presented R Aqr data are very useful to clarify the complex hydrodynamics taking place in AGB binaries.

We construct smooth collapsing sequences with sectional curvature at least $-1$ whose limits are topological manifolds with nonzero Kirby–Siebenmann class. For every $n\geq6$, the source and limit are simply connected, have dimensions $n+3$ and $n$, and the metrics have a common positive Ricci lower bound. The limits are homeomorphic to the product of the Chern four-manifold with $S^{n-4}$. We also obtain spin examples and five-dimensional lens-space homotopy types, the latter from finite quotients of one fixed $2$-connected eight-manifold. Finally, we realize aspherical manifolds with no closed PL representatives of their homotopy types as collapsing limits with scalar curvatures tending uniformly to $+\infty$, and establish an obstruction to their approximation with almost nonnegative Ricci curvature.

The exceptional group $F_4(\mathbb{C})$ has two conjugacy classes of involutions: an involution $t$ with centralizer $\mathrm{Spin}(9,\mathbb{C})$, and the principal involution, with centralizer $(\mathrm{SL}(2,\mathbb{C})\times\mathrm{Sp}(6,\mathbb{C}))/\{\pm1\}$. We determine the values of all irreducible characters of $F_4(\mathbb{C})$ at $t$. For the irreducible representation with highest weight $a\omega_1+b\omega_2+c\omega_3+d\omega_4$ (in Bourbaki's labelling), we show that the character vanishes at $t$ if and only if $c$ and $d$ are both odd. In all other cases its value is $\pm 2^{-7}$ times the dimension of an explicit irreducible representation of $\mathrm{Spin}(9,\mathbb{C})$. The proof is a limiting argument in the Weyl character formula, combined with a decomposition over the three cosets of $W(B_4)$ in $W(F_4)$ and the triality action of $W(F_4)=W(D_4)\rtimes S_3$ on the weight lattice modulo the $D_4$-lattice. At the principal involution, Nadimpalli, Pattanayak and Prasad showed that the character value is $\pm 1/2$ times the dimension of an irreducible representation of $\mathrm{SL}(2,\mathbb{C})\times\mathrm{Spin}(7,\mathbb{C})$. Comparing the two results, we find that the characters of $F_4(\mathbb{C})$ vanish at the two involutions for exactly the same highest weights. Together these results determine all irreducible characters of $F_4(\mathbb{C})$ at all involutions. We also explain how our formula at $t$ relates to the generalized Weyl character formula of Gross, Kostant, Ramond and Sternberg.

Small targets occupy few patches in a vision-language encoder, so spatial features often mix object appearance with surrounding content. We propose Prototype Contrast and Local Magnification (PCLM), a support-conditioned localization method that uses a frozen CLIP encoder. Five masked support images per class define foreground and background prototypes through equally weighted regional features. Their difference provides a shared scoring direction for query patches, explicitly comparing target evidence with the demonstrated background. Nine overlapping query windows are enlarged and encoded independently to sample small targets more densely. Reprojection and coverage averaging combine their scores into a continuous localization map. The class direction occupies 2 KiB regardless of support count and transfers unchanged across datasets with mapped categories. On 5,047 small-target queries from VOC, COCO, ADE20K and Oxford-IIIT Pets, PCLM achieves higher mean pixel AP than every evaluated text-conditioned localization baseline on each dataset under our evaluation protocol. Gains over the strongest scene-dataset baselines range from 5.63 to 13.23 percentage points. At comparable measured latency, local magnification improves scene small-target AP by 4.51 to 5.68 points over whole-canvas enlargement. Factorial experiments show that prototype contrast increases the benefit of local observation, including under matched image-coordinate filtering. Support-budget experiments show that additional examples refine category estimation without increasing representation size or query-time scoring cost.

Large language models (LLMs) progressively revise their predictions across Transformer layers, yet we typically observe only the final output, discarding the trajectory through which it is formed. We introduce Belief-Trajectory Energy(BTE), a model-grounded measure that characterizes an input through the layerwise predictive revisions it induces in a model. By mapping intermediate states into a shared predictive space, BTE provides a principled measure of belief change that can be summarized as either a scalar or a structured depth profile. Theoretically, we show that local BTE corresponds to predictive revision under the Fisher-Rao geometry, while the sequence of revisions captures information beyond the initial-to-final belief change. Empirically, scalar BTE provides a model-relative signal of difficulty across diverse reasoning tasks, while richer BTE representations support human-LLM review detection and fine-grained generator attribution, reaching up to $0.998$ macro-AUROC and $95.6\%$ eight-way attribution accuracy. Further analysis shows that BTE develops throughout pretraining and is selectively reshaped by targeted training, demonstrating that the resulting measurement reflects what the scoring model has learned. Together, our results establish belief trajectories as a principled model-grounded signal and suggest a broader perspective in which learned models can themselves serve as instruments for characterizing the data they process. More demonstrations can be found at https://yingjiahao14.github.io/BTE-web/.

It is well-known that a random Steiner process converges almost surely to a Euclidean ball. On the other hand, as we show in this article, a planar random shaking process converges almost surely to a triangle (in the Banach-Mazur distance). Asking whether these phenomena are generic or exceptional, we study random iterations of an operation called $\lambda$-Steiner symmetrization, which includes Steiner symmetrization and shaking as special cases. We establish abstract criteria for almost sure convergence and divergence of such processes, obtaining criteria for many other symmetrization operations from convex geometry as a byproduct. Based on a joint characterization of ellipses and triangles, we give a full resolution to our question in the planar case.

Asynchronous circuits have some benefits over their synchronous counterparts, such as lower idle power consumption, potentially lower latency, absence of clock distribution issues, among others. However, industry-scale electronic design automation (EDA) support, particularly for high-level synthesis (HLS) of asynchronous systems from imperative programs, remains limited. We present an approach to synthesize asynchronous circuits from imperative programs using the AHIR HLS framework. The framework models the control path in the form of a restricted form of Petri nets that are 1-safe and live by construction. In this work, we start with this representation in AHIR and generate asynchronous circuits with a delay insensitive controller, a single-rail data path with bundled-data interaction between the controller and the data path that resembles the micropipelines architecture. We formalize the timing constraints required for correct functioning of the generated circuit and show that those constraints are one sided (feasible), necessary and sufficient. We have also developed a static timing analyzer to check that the circuit meets these constraints. Our flow integrates seamlessly with off-the-shelf ASIC toolchains and standard synchronous cell libraries. To demonstrate our approach, we show an end to end implementation of an AES encryption algorithm and use post layout circuit simulation to verify the correctness and estimate the performance of the resultant circuit.

Multimodal inputs are inherently heterogeneous, not only across modalities but also in the information pathways required for effective prediction. To address this limitation, we propose InstMoE, an adaptive expert routing framework for multimodal learning. InstMoE dynamically routes each input to specialized unimodal and cross-modal experts, allowing the model to adapt its information pathways to the characteristics of the input. However, routing can be misled when modality-specific variations obscure task-relevant semantics. Such irrelevant variations may distort routing decisions, causing inputs to be assigned to inappropriate experts. We therefore introduce a Contrastive Semantic Alignment module, which encourages semantically similar inputs to share task-relevant representations while suppressing irrelevant modality-specific variations. Experiments on multimodal sentiment analysis benchmarks demonstrate that InstMoE achieves state-of-the-art performance on CMU-MOSEI and CH-SIMS v2 while using substantially fewer parameters than competitive baselines. Further analysis shows that different inputs exhibit distinct expert preferences, demonstrating that InstMoE moves beyond fixed fusion toward adaptive multimodal computation.

We develop a family of compact, dispersion-corrected marker-and-cell (MAC)-based finite-difference discretizations for the elastic Helmholtz equation. While dispersion correction is well established for scalar Helmholtz problems, fewer approaches exist for the elastic problems, and most use non-compact stencils. To this end, we introduce a generalized MAC (GMAC) framework that enables compact discretization of the elastic Helmholtz equation. We prove that even though GMAC is non-symmetric, this discretization produces no numerical dissipation and no shear-mode separation. For certain weights, GMAC has lower polarization error than standard MAC. We also introduce an elastic dispersion correction (EDC) process, extending real-shifted-wavenumber dispersion correction methods from acoustic to elastic Helmholtz problems. We prove and demonstrate numerically that GMAC is second-order accurate. Furthermore, we show that EDC substantially reduces dispersion for standard MAC and can provide additional correction in some instances of GMAC. Dispersion analysis shows that the resulting method outperforms existing compact finite-difference discretizations in dispersion accuracy and is competitive with some non-compact schemes. We give numerical evidence for improved phase alignment in mildly heterogeneous media.

Recently, unified high-definition image restoration has attracted considerable attention; however, existing models tend to excessively increase their depth in pursuit of improved generalization, which often yields only limited gains. Meanwhile, loop-based learning paradigms have drawn widespread attention due to their low parameter counts and strong regression capability, as exemplified by GPT-6 and looped Transformers. In this paper, we introduce the loop learning paradigm to address restoration tasks that require cross-domain learning. Specifically, we propose LoopMoEVR, a loop-based mixture-of-experts model capable of handling degraded ultra-high-definition (UHD) inputs. First, a degradation-conditioned low-rank loop embedding is designed to construct input-dependent stage conditions. Second, a spatio-temporal iterative adaptive normalization module, termed IterAda3DN, is developed to fuse local features with global loop context, thereby performing position-wise affine modulation. Finally, the expert branches further integrate the attention-updated local and global video states with the loop conditions to generate dedicated modulation parameters, while an input-conditioned depth predictor adaptively configures the number of loop iterations. With only approximately 0.884M trainable parameters, the proposed model uniformly handles UHD video dehazing, deraining, denoising, and low-light enhancement tasks, achieving state-of-the-art restoration performance on both public benchmarks and real-world scenarios.

Diffusion models achieve strong generative performance but often struggle to align generated samples with human preferences measured by a reward model. A simple yet effective algorithm for test-time alignment is Best-of-$N$ (BoN) sampling, which draws $N$ i.i.d. samples from a pre-trained diffusion model and outputs the single highest-reward sample. Despite its empirical success, BoN makes limited use of reward information, as it is incorporated only at the final selection stage without influencing the reverse diffusion trajectory during sampling. Consequently, BoN sampling does not improve the average alignment of generated samples and is primarily suited to single-output settings. We propose Best-of-$N$ Guidance (BoNG), a novel method that integrates the principle of BoN sampling directly into the reverse diffusion process. BoNG performs online BoN selection over denoising particles and adjusts the reverse diffusion process to steer the particle population toward higher-reward regions during generation. Specifically, by introducing an asymmetric guidance interaction among denoising particles, BoNG uses the current BoN particle as a guidance signal to the rest of the particle population. This particle-level interaction reshapes the sampling process toward higher-reward regions, enabling BoNG to improve not only the final best sample beyond Vanilla BoN sampling, but also the average quality of generated samples. Over 36 empirical comparisons, BoNG achieves the best performance in 29 cases, ranking first in 80.56% of the comparisons against SMC and Vanilla BoN sampling. BoNG also supports multi-output capability, achieving 1.3$\times$ ImageReward score of the latest sample-based guidance method with a 1.6$\times$ speedup. We release the code at https://github.com/aailab-kaist/BoNG.

Search agents repeatedly make short decisions about relevance, evidence sufficiency, and search actions. Using generative language models for these decisions introduces latency and unreliable confidence. We present SearchJev, a calibrated System-1 model that separates search decisions from System-2 reasoning and generation. Given a search state and a decision schema, SearchJev scores legal options without autoregressive output generation. We propose Soft-Label Learning for Calibrated Decisions (SLCD) to learn decision probabilities from uncertain supervision and improve the reliability of decision confidence. In a dual-system search agent, SearchJev handles short decisions and delegates uncertain judgments to System 2, which retains planning, query generation, and answer composition. We also introduce SearchDecision-Bench, a benchmark unifying six types of search decisions for training and evaluation. On SearchDecision-Bench, SearchJev improves decision quality over same-size Qwen3.5 autoregressive models, achieves 5.2-5.3 times faster decisions, and reduces average expected calibration error by 41-74%. On BrowseComp-Plus, the dual-system agents achieve a 3.7-4.7 times speedup in active search time while improving answer accuracy from 45% to up to 54%.

Long-form reasoning makes inference expensive, and speculative decoding mitigates this cost by verifying multiple draft tokens in parallel. Its speedup, however, can fade as context grows and draft acceptance declines. We focus on attention-mass dilution: as softmax normalizes over more visible Keys, the mass concentrated on the highest-scoring Keys can decrease. We introduce SharpDraft, a training-free method that counteracts this effect through cardinality-aware Query scaling, without the computational overhead of online adaptation. Under explicit assumptions, we derive an exact top-$k$ mass correction and deploy a closed-form fixed-slope approximation. Across AIME-26, GPQA-Diamond, and LongGenBench Diary, SharpDraft achieves $2.59$-$3.19\times$ geometric-mean end-to-end speedups over target-only autoregressive decoding when applied to DFlash, PARD, and EAGLE 3.1. With DFlash, it improves decoding speed and outperforms full-parameter and LoRA-based online adaptation in end-to-end speedup, while matching the unmodified drafter's reported peak allocated GPU memory.

Split learning has emerged as a promising paradigm for privacy-preserving LLM fine-tuning, yet its practical deployment is severely hindered by the sequential communication-computation bottleneck. In conventional synchronous pipelines, clients remain idle while waiting for server-side gradients, resulting in substantial training inefficiency. We propose CommuteProp, an asynchronous split-learning algorithm that decouples the training process into two concurrent phases: a cross-block forward-backward pass and an in-block weight update. By overlapping computation with communication, CommuteProp reduces the marginal cycle time from a sum of all stage latencies to the dominant computational bottleneck. We provide a rigorous asynchronous error and convergence analysis. Moreover, we derive an NS preconditioner method based on our analysis to further mitigate staleness-induced noise. Comprehensive experiments indicate that our algorithm achieves substantial throughput gains while maintaining accuracy comparable to synchronous methods; furthermore, it functions as a plug-and-play module that not only accommodates but actively enhances existing privacy enhancement methods.

In this paper, we present a machine-learning-based framework for identifying leptonically decaying, highly boosted top quarks, $t\to b\ell\nu$, at the Future Circular Collider: hadron-hadron mode (FCC-hh) operating at a center-of-mass energy of $\sqrt{s}=100$~TeV. At these energies, the large Lorentz boost causes the top-quark decay products to become highly collimated, with the lepton embedded within the resulting top-quark jet. In this regime, it is not a priori clear whether constituent-based jet taggers, which rely on information from individual jet constituents, retain the high discrimination power achieved at LHC energies. To answer this question, we employ two complementary approaches: a variable-based tagger trained with a deep neural network (DNN) using physics-motivated observables and a constituent-based Particle Transformer (ParT) acting on the jet constituents. Both approaches are optimised for the characteristic non-isolated lepton signature of boosted leptonic top decays and achieve comparable performance in discriminating leptonic top jets from hadronic top jets and heavy-flavor $b$- and $c$-jet backgrounds. We then incorporate the tagging framework into an event-level search for single production of a vector-like $B^\prime$ quark decaying through $B^\prime\to tW$, and use it to evaluate sensitivity to $B^\prime$ in the range of $3$ to $10$~TeV. For the reference $B^\prime$ benchmark, the proposed search strategy achieves a $5\sigma$ discovery reach of approximately $7$~TeV under the width requirement $\Gamma_{B^\prime}/m_{B^\prime}<25\%$, in both the statistics-only and $10\%$ background systematic uncertainty scenarios. Relaxing the width requirement to $\Gamma_{B^\prime}/m_{B^\prime}\lesssim 50\%$ extends the discovery reach to approximately $10$~TeV.

Reliable learning-based vulnerability detection requires high-quality labels, yet datasets built from vulnerability-fixing commits may label functions as vulnerable simply because they were changed by a security patch. We present VulValidate, a framework that uses LLM agents to coordinate dynamic analysis tools and construct vulnerability-triggering experiments from runtime feedback. Given a labeled function and its fixing patch, VulValidate reconstructs vulnerable and fixed revisions, selects suitable tools and execution paths, refines triggering inputs, and compares runtime behavior to assess function-level attribution. We audit all 35,849 instances originally labeled vulnerable in BigVul, PrimeVul, and DiverseVul. We confirm 20,510 (57.2%), correct 6,819 labels (19.0%), leave 7,981 attacked but undecided (22.3%), and cannot successfully measure 539 (1.5%). After conflict resolution and byte-exact deduplication, the corrected release contains 15,890 distinct confirmed vulnerable function bodies. In a blinded review of 581 sampled decisions, expert consensus supports 90.0%–92.0% of confirmations and 92.6%–99.0% of label corrections. With model parameters fixed, corrected evaluation lowers F1 for all five tested detectors on both BigVul and DiverseVul; retraining with corrected labels improves F1 for four of five detectors on each dataset. We also release a reusable VulValidate skill, corrected datasets, and reproducible evidence for future vulnerability-detection research.

We study differentially private learning problems in the realizable setting, where a hypothesis is specified by $k$ components. A direct iteration of private component learners is obstructed by a simple difficulty: an approximate choice of the next component may destroy exact realizability of the labeled sample, even when the next component is locally accurate. We restore realizability using the LabelBoost procedure of Beimel, Nissim, and Stemmer [SODA '15, Algorithmica '21] and recycle data through two alternating reservoirs. The resulting learner, for a target privacy $\varepsilon$, pays only $\widetilde O(\sqrt{k}/\varepsilon)$ overhead relative to the active sample requirement of a single component learning step at target accuracy $\Theta(\alpha/k)$. For learning $d$-dimensional halfspaces over a finite coordinate grid of size $L$, exact realizability makes the direct component-depth objective quasi-concave. Instantiating the framework with the IPConcave algorithm of Nissim, Tsfadia, and Yan [SODA '26] and with the quasi-concave optimizer of Cohen, Lyu, Nelson, Sarl'os, and Stemmer [STOC '23] yields a realizable sample complexity of $\widetilde{O}\left(\frac{1}{\varepsilon \alpha}\cdot \min\{d^{2.5} \log^*L, \:\: d^{2.5} + d^{1.5} 2^{\log^*L}\}\right),$ which improves on the previously known bound of $\widetilde{O}\left(\frac{1}{\varepsilon \alpha}\cdot\min\{\frac{1}{\alpha}\cdot d^{5.5}\log^*L,\:\: d^{2.5}2^{\log^*L}\}\right).$ We also apply the framework to Boolean compositions: given proper private learners for classes $H_1,\ldots,H_k$, we obtain a proper private learner for $G(H_1,\ldots,H_k)$ for any fixed Boolean function $G:\{0,1\}^k\to\{0,1\}$. Compared with the closure theorem of Alon, Beimel, Moran, and Stemmer [COLT '20], this reduces the overhead on a common component sample bound from $\widetilde O(k/\varepsilon)$ to $\widetilde O(\sqrt{k}/\varepsilon)$.

With the development of pulse-tube cryocoolers, the availability and use of low-temperature environments have increased exponentially in recent years. However, replacing liquid helium with a mechanical cooling device has, since its introduction, highlighted the need to attenuate the mechanical vibrations generated by the cryocooler itself. In the case of dilution refrigerators, which are capable of reaching temperatures on the order of few mK, the issue of mechanical vibrations reaching the experimental setup is, in many situations, extremely critical. This has led to the development of several customized experimental approaches aimed at attenuating vibrations. However, the inertial measurement of vibrations at cryogenic and, in particular, ultracryogenic temperatures has relied on customized sensors, since no commercially available, calibrated instruments exist for measurements at these temperatures. Nevertheless, many research groups employ commercial geophones that are recalibrated at low temperatures. This approach, however, is not free from systematic errors and presents significant challenges in quantitatively assessing the absolute response. In this work, we describe the development of cryogenic geophones exhibiting an almost temperature-independent response, from room temperature down to approximately 10 mK.

State-of-the-art neural operators scale to complex meshes via slice-and-process architectures, yet many rely on linear compatibility scores for latent tokenization. Under common feature normalization, such scores are equivalent to isotropic Euclidean clustering, while without normalization they induce unbounded linear decision regions. In both cases, they lack slice-specific anisotropic locality, which can lead to redundant and entangled latent slices. To address this, we propose Metric-Enhanced Token Routing Operator (METRO), a geometry-aware routing mechanism that replaces linear projection with a learnable Mahalanobis metric. By enabling each latent slice to learn a local anisotropic tensor, METRO shapes receptive fields into exponentially localized, oriented ellipsoids that naturally align with flow features like boundary layers and wakes. As a drop-in replacement, METRO yields consistent improvements across both Transformer and Mamba backbones. Empirically, our method achieves substantial performance gains on irregular domains, outperforming baselines on both standard PDE benchmarks and complex industrial design tasks. Finally, METRO exhibits enhanced robustness in out-of-distribution regimes across varying Reynolds numbers and geometric configurations.

Locating relevant files from natural-language requests is a core subtask for agents operating over code repositories. We investigate whether this task can be delegated to compact, specialized models to enable on-device search while reducing the token usage, latency, and inference cost of larger agents. We show that semantic search improves file localization, with gains in accuracy, cross-language transfer, and inference efficiency. To study this setting, we introduce a training framework for file-localization agents built around ColGREP, a local semantic search tool based on late-interaction retrieval models. Our recipe combines weighted supervised fine-tuning on teacher trajectories, assigning turn-level credit based on retrieval outcomes, with reinforcement learning on localization quality. We train three model families with fewer than two billion parameters to formulate search queries, inspect retrieved content, and identify relevant files. On localization tasks derived from SWE-bench Lite and Multi-SWE-bench Flash, ColGREP-equipped agents substantially improve over their base models and outperform corresponding GREP-based agents. In addition to improving localization accuracy, ColGREP reduces mean end-to-end trajectory latency by 44.1% on CPU while using 29.1% fewer tokens, and enables better generalization to programming languages unseen during fine-tuning. These results suggest that compact, tool-specialized localization agents can provide an efficient interface between natural-language requests and large codebases.

Many latent neural operators represent input and output fields in a stationary latent chart. In particular, common latent routing mechanisms use fixed or shared assignment weights for feature projection and reconstruction, limiting their ability to model transport-dominated systems where coherent structures move relative to fixed coordinate frames. We propose Advectra, a transport-aware latent operator that introduces a regularized kinematic coordinate map to decouple source and target coordinate systems. This yields an approximately co-moving latent reference frame and enables asymmetric feature aggregation and reconstruction. Combined with a geometry-aware ordering mechanism for state-space models, Advectra captures advective dynamics while maintaining stable global interactions. Advectra achieves the best performance among evaluated geometry-constrained and form-free baselines on advection-dominated benchmarks, including passive scalar transport in Navier–Stokes flows and Rayleigh–Taylor instability, while demonstrating strong generalization on real-world engineering tasks. These results highlight the benefit of explicit moving-frame structure in neural operators for non-stationary physics.

As multimodal large language models (MLLMs) reason for longer, attention to the initial visual input diminishes, weakening visual grounding. Visual memory reintroduces visual evidence during reasoning. We conduct a controlled analysis of visual memory along three axes: curation, organization, and access. We find that local evidence benefits from global context, compact latent representations balance accuracy and visual-context cost, and the utility of memory access depends on the reasoning state. Guided by these findings, we propose ReMAP (Reasoning-Time Memory-Augmented Perception), which couples two complementary latent memories: a static, question-conditioned Global memory that preserves scene and cross-image context, and a dynamic Local memory that uses this context as an anchor while selecting and re-encoding region-level evidence according to the current reasoning state. Both memories return compact latent tokens inserted into the reasoning sequence, and a reinforcement-learning access policy trained with branched rollouts decides when to continue reasoning or invoke Global or Local memory. On ten benchmark families, ReMAP outperforms prior visual-memory methods on all four multi-image benchmarks, exceeding the strongest prior results on MuirBench and MIMIC by 8.38 and 14.84 percentage points. Across four backbone families, enabling memory access improves over the same trained model with memory disabled, and on shared V*Bench, CV-Bench-2D, and MuirBench questions ReMAP reduces the visual tokens entering the reasoning sequence by 51.0-76.8% relative to the native-resolution backbone. Further analyses show that Global and Local memory form distinct yet complementary latent representations. Together, these components restore the perceptual cycle by letting the reasoning state trigger targeted visual retrieval, with the retrieved evidence guiding subsequent reasoning.

Motivated by Igusa's relation between relative cyclic homology and the logarithm of the graded Cartan determinant, we isolate the homological step behind the passage from logarithmic determinants to traces of powers. For a bounded chain complex $(C_*,d)$ and a degree-zero graded endomorphism $f$, not assumed to commute with $d$, we construct the universal quotient on which $f$ becomes a chain map. The kernel of this reflection iterates to a canonical functorial minimal filtration, whose Rees module gives a flat degeneration to an uncurved associated graded. For finite-dimensional chain groups all power supertraces of $f$ are represented by Lefschetz traces on this associated graded, so the resulting logarithmic-determinant and zeta series admit a canonical homological representation.

Latent Dirichlet Allocation (LDA) and models derived from it remain widely used topic models. LDA observes each document as a bag-of-words and models each topic by a categorical distribution over the vocabulary, so that it uses neither the internal structure of the document nor the similarity in meaning between words. Earlier work has responded to this limitation in two ways: many models have introduced embeddings, and some have assigned topics to sentences rather than to words. Their combination, a topic model that observes sentence embeddings, remains little explored. We propose vMF Sentence LDA (vSLDA), which keeps the admixture structure of LDA, observes each sentence as its L2-normalized embedding and models each topic by a von Mises-Fisher (vMF) distribution, which matches the cosine geometry of sentence embeddings. Its per-topic parameter count and per-iteration cost are linear in the embedding dimension, versus quadratic for the full-covariance Gaussian distribution in the existing model over sentence embeddings. We evaluate vSLDA where the limitation is expected to matter most, among topics that share much of their vocabulary: the topics that subdivide the one subject of a collection, and the narrow topics that result when a corpus is divided into a large number of topics. On two corpora, vSLDA attains the best mean rank against eight baselines when the fine categories within each coarse category are classified from the document-topic distributions. On the whole corpus, its advantage appears or widens as the number of topics grows. Weighting the word frequencies of each sentence by its topic posterior yields expected topic-word counts of the same form as those of LDA, so that the standard topic coherence and diversity measures apply to models that assign topics to sentences. On their product, topic quality, vSLDA leads in most within-category conditions of both corpora.

Researchers increasingly use Large Language Models as judges (LLM-as-a-judge) to evaluate model outputs. Yet there are no standards for how to design these judges. Typically, researchers choose the prompt, rating scale, and model intuitively. If these choices change the judge's verdicts, two studies can reach different conclusions about the same facts. To address this risk and to provide an empirical basis for judge designs, we evaluate 10 reasoning models across multiple designs on two tasks: a scalar rating of sentence sentiment and toxicity (over 500 items per category), as well as a binary accuracy classification of question-answer pairs (n=600). For the rating tasks, despite judges showing significant disagreements with the human ground truth, the practical size of differences is small enough to consider most judges reliable (mean absolute deviation of 0.11 points on a 1 - 7 scale); toxicity judges even outperform standard classifiers. Judges are also highly accurate on average (96.5%) for the accuracy classification task. However, design choices can produce shifts: changing the rating scale alone can shift measured bias by up to 0.93 points (rating task), and while accuracy levels are rarely impacted, design choices consistently impact judge leniency (classification task; leniency drop of 28.9 percentage points when using detailed prompts, and up to 56.1 percentage points when switching models). Counterintuitively, lower reasoning effort affects neither accuracy nor leniency. Across both tasks, model identity is the dominant source of variance. These findings suggest that while LLM judges are broadly trustworthy in aggregate, design choices can be meaningful sources of variance. Given the growing reliance on automated evaluation in LLM research, we intend this study as a methodological reference for designing more robust and replicable LLM-as-a-judge pipelines.