Computer Vision
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Thursday 3 September
Seeing frames in order does not mean representing time. Modern VideoLMs receive ordered video streams, yet their main supervision acts on generated text rather than video-token representations where event dynamics should first emerge. This mismatch allows models to learn temporal answers from shortcuts such as objects, scenes, and language priors, without requiring internal video representations to capture event progression. To address this, we propose VT-Contrast, a representation-level temporal counterfactual objective for VideoLMs. Its design asks where temporal supervision should act and what temporal differences it should expose. VT-Contrast supervises selected late-layer last-frame video tokens, where temporal information is expected to be integrated before language generation, and contrasts order-preserving views with same-video reordered counterfactuals graded by Kendall tau distance. It requires no architectural changes, is compatible with diverse VideoLM training tasks, and improves overall performance across temporal understanding benchmarks. Our code is available at https://github.com/ANDgate99/VT-Contrast.
Materials Science
Strengthened Silicate Glasses by Residual Stress: Depth of Compression and Surface Flaws Stability Conditions
The application of silicate glasses in severe service environments requires a precise evaluation of structural strength under mechanical loads and surface tribological conditions. Because glass strength is governed by surface flaws and microcracks rather than being an intrinsic material property, residual surface compression fields, balanced by interior tensile zones, are widely implemented to inhibit flaw opening. Rather than relying on conventional allowable stress criteria to establish product acceptance, this study adopts a fracture mechanics framework based on the stress intensity factor KI and fundamental material limits: the critical stress intensity factor KIC for rapid fracture and the threshold stress intensity factor KIth for time-delayed static fatigue failure. Using the Weight Function Method (WFM), KI is evaluated across generic surface flaw depths for two-dimensional continuous (2D-Continuous) surface cracks subjected to non-uniform internal residual stress fields and external loads. Flaw stability criteria are established for both zero-risk and moderate-risk design methodologies. Finally, the interaction and superposition of externally applied mechanical and thermal stresses with internal residual stress fields are evaluated.
Read the full abstract
Computation and Language
Sequential Beats Joint: On the Interplay between On-Policy Distillation and RLVR
Reinforcement learning with verifiable rewards (RLVR) and on-policy distillation (OPD) have emerged as two dominant methods for post-training reasoning LLMs. Prior work uses OPD's dense token-level supervision to complement the sparse RL reward, fusing the two signals within a single step: either as a weighted-additive combination or a teacher-modulated rescaling of the RL advantage. In this paper, we show that a simple two-stage scheme, OPD-then-RL, consistently outperforms pure OPD, pure RLVR, and all such joint baselines across logic and math reasoning benchmarks. Beyond the empirical results, we further provide a systematic understanding of this through pass@$k$ behavior, learning dynamics, and parameter updates, yielding a consistent explanation: OPD expands the student's coverage of teacher-supported solutions and RL sharpens within that support, while jointly optimizing the two signals causes them to interfere. To provide a practical recipe, we find that the OPD validation score is the key signal for when to switch to RL, and that OPD is a better cold start for RL than SFT. Together, our results establish OPD-then-RL as a simple yet strong way to combine the two methods, turning two entangled signals into complementary stages.
Read the full abstract
High Energy Physics (Theory)
The growth of $SO(9)$ super-representations in Type II string theory
We study the growth of massive $SO(9)$ super-representations in type II and type I string theories. We do this first directly by finding an empirical formula that specifies which representations appear at any level, and then compute the multiplicities from a refined partition function evaluated over finite fields up to level 501 for individual representations, and level 226 for all representations. We then derive asymptotic formulae for the growth of any representation, which are constructed by integrating about the peaks of the refined partition function. Using a Rademacher sum we can find very accurate approximations for the multiplicities of the representations. Unlike the superstring partition function, which only receives contributions from the odd Rademacher terms, the multiplicities for any representation have contributions from both even and odd terms. Finally, we apply the same techniques to the ordinary representations, where a simplified refined partition function allows for better computational speed and simpler expressions for the asymptotic approximations.
Read the full abstract
Atomic Physics
Sub-barrier peaks in atom-atom-ion three-body recombination
Hybrid atom-ion experiments have recently entered the few-partial wave regime in which magnetically tunable Feshbach resonances dominate the three-body loss dynamics. This necessitates a treatment of three-body recombination that can incorporate the role of the resonant dimer channel. Here, we develop a two-step Lindemann-type model for resonant atom-atom-ion three-body recombination that combines multichannel quantum-defect theory for the long-range polarization potential with a universal quantum-capture treatment of the inelastic atom-dimer step. We show that within our model, rate competition can lead to a peak in the three-body-recombination cross section of higher partial-wave channels at energies well below their corresponding centrifugal barrier. Investigating the resulting loss rates under numerically obtained non-thermal collision-energy distributions, we find that these effects survive the energy averaging and can be observed in state-of-the-art hybrid experimental setups. Our findings pave the way for a detailed characterization of atom-ion Feshbach resonances via the analysis of inelastic three-body-recombination processes.
Read the full abstract
Machine Learning
Hardware-Aware FP4 FlashAttention-4
Blackwell's 4-bit floating-point (FP4) tensor cores do not automatically make attention faster because softmax conversion and on-chip dependencies dominate once its matrix products shrink. We address this with Direct-P for noncausal inference and a causal path that passes the forward quantization directly into backward. Direct-P maps scores directly to FP4 probabilities and reaches up to 2.13$\times$ the bfloat16 (BF16) forward throughput on an NVIDIA GB200. The causal path reconstructs probabilities from saved quantized queries and keys and uses 8-bit floating-point (FP8) gradient operands, accelerating a complete single-GPU 8-billion-parameter update by up to 1.14$\times$. Matched distributed training retains FP8 probabilities and values; every tested MXFP4 probability/value training trajectory diverges.
Read the full abstract
Quantum Physics
Gravitationally Induced Entanglement Across an Event Horizon
It has long been assumed that a particle–having crossed a black hole event horizon unentangled with another particle in the exterior universe–can no longer dynamically entangle with that particle. Here, we first establish the possibility of post-crossing entanglement via an electromagnetic mediator. We then show within a gravitational time-delayed-potential model that entanglement can be created between freely falling spatial superpositions across an event horizon. Radial escape, however, triggers bremsstrahlung emission–imposing a dephasing bound that grows with entangling phase, $\Gamma > \frac{6}{11\pi}\Phi$, within a constrained leading-quadrupole model–and is ultimately ruled out by macroscopic which-path leakage. By contrast, equivalent macroscopic optical masses in principle allow local harvesting of entanglement tangentially, via quantum erasure, thus revealing a remarkable geometric duality: Spacetime irreversibly degrades entanglement under radial escape, while allowing the transverse teleportation of the entangled qubit to infinity, independent of Hawking evaporation.
Read the full abstract
Robotics
Corner Cases: Headland Coverage Path Planning for Autonomous Driving in Arable Farming
This paper presents a new method for headland coverage path planning for arable fields. Several earlier approaches suggest covering the headland with nested polygons and smooth turns, however, covering the field corners entirely requires manoeuvres with reversing. In the new method, the polygon corners are modified to allow a reversing turn. A comparison to two other methods considering gap, overlap, and crossing the field boundary shows an improvement in the coverage result especially in field corners of around 90 degrees, and 240 degrees and above. Applicability of the new method is shown with several examples of real polygonal field maps.
Read the full abstract
Sound
Compressing Streaming Neural Audio Encoders via Latent-Space Distillation
System-wide Dictation on Apple devices runs entirely on-device, and the speech it transcribes reaches the foundation model through a tokenizer: an encoder that maps short windows of waveform onto the representation the language model reads. Because that model is sparsely activated under Instruction-Following Pruning, only a small subset of its experts occupies DRAM at any time, so the always-on tokenizer competes for the same memory, and its parameter count bears directly on power and latency. In this work we study how to compress such a tokenizer by distillation, taking as the supervision target neither the discrete token nor the output distribution but the pre-quantizer latent the model actually consumes - the last representation the two token interfaces share. We train only the student encoder to regress the teacher's per-frame latent under a squared-error objective, with a single affine layer absorbing the teacher-student width mismatch. Because the target precedes both the quantizer and the language-model bridge, one recipe covers both token interfaces we support, and applies both to a tokenizer pretrained alone and to one jointly trained with a language model. At 2.8x compression the distilled student stays within 1.9% relative WER of its teacher on five of six teacher-student pairs without any fine-tuning, and improves on an independently trained tokenizer of identical capacity by 3.9% relative.
Read the full abstract
Classical Analysis and ODEs
Endpoint Estimates for Stein's Purely Quadratic Carleson Operator
We study the near $L^1$ behavior of the maximally quadratically modulated Hilbert transform \[ \mathcal{C}_2f(x) := \sup_{\lambda} \left|\operatorname{p.v.} \int_{\mathbb{R}}f(x-y)e^{2 \pi i \lambda y^2} \frac{\mathrm{d} y }{y} \right| \] and its lacunary counterpart obtained by restricting $\lambda$ to $2^{\mathbb{Z}}.$ We prove that if $\Phi$ is a Young function satisfying $ \Phi(t)=o\bigl(t\log_2t\bigr)$ then $\mathcal{C}_{2,\mathsf{lac}}$, and therefore $\mathcal{C}_2$ as well, does not satisfy a corresponding $\Phi$-modular estimate. In particular, neither the lacunary nor the full quadratic operator is of weak type $(1,1)$. In the positive direction, we establish an $L\log_1 L$ modular estimate for $\mathcal{C}_2$ and a $L (\log_2L)^2 \log_4L$ estimate for $\mathcal{C}_{2,\mathsf{lac}}.$
Read the full abstract
Optimization and Control
Zero sum two-player differential game under three regimes
This paper investigates a two player zero-sum stochastic differential game characterized by three distinct regimes, reflecting the regime switching dynamics within the system. We aim to derive an explicit solution for a number of configurations of the switching system by means of the viscosity solutions approach. In particular, we analyze the associated Hamilton-Jacobi-Bellman-Isaacs equations and examine how the interactions between the different regimes affect the structure of the value function. The proposed framework provides a systematic approach to characterize the optimal strategies of the two players and to obtain explicit representations of the value function un der suitable assumptions. We also illustrate how to derive the value functions in case we know the qualitative structure of switching regions.
Read the full abstract
High Energy Physics (Theory)
SCFT/VOA correspondence and R-twisted reductions of $(A_2,D_{3n-2})$ Argyres–Douglas theories
We study the SCFT/VOA correspondence for the $(A_2,D_{3n-2})$ Argyres–Douglas theories with $n\geq2$. From the matching of the central charges and the equality of the Schur index with the vacuum supercharacter to all orders, we propose that the associated VOA is the logarithmic doublet algebra $\mathcal A(4n-2)$ of Feigin, Feigin, and Tipunin. For $n=2$, this correspondence was studied in detail by Buican and Nishinaka, and our proposal extends it to $n\geq3$. Using the Nahm sum expression for the vacuum supercharacter, we further propose a family of 3d $\mathcal N=2$ abelian Chern–Simons matter theories that flow to the 3d $\mathcal N=4$ SCFTs obtained by the R-twisted circle reduction, generalizing the $n=2$ construction. We determine the monopole superpotential and show that the Coulomb branch is $\mathbb C^2/\mathbb Z_2$ for every $n$. The superconformal index provides quantitative evidence for the proposed infrared fixed points, including the expected enhancement to $\mathcal N=4$ supersymmetry and the Higgs and Coulomb branch structure.
Read the full abstract
Artificial Intelligence
Why Gated DeltaNet Survives 4-Bit Quantization: NVFP4 W4A4 for the Recurrent Half of a Hybrid 27B LLM
Hybrid LLMs pair softmax attention with linear-attention layers such as Gated DeltaNet (GDN), whose recurrent state summarizes the context in fixed size. Early community 4-bit quantizations of Qwen3.8-27B (48 GDN layers, 16 attention layers) left the GDN block in 8- or 16-bit precision – especially its decay and write-strength gates – on the intuition that errors in a recurrence accumulate over long contexts. We test that intuition by building Minima: NVFP4 W4A4 on all 496 linear layers, GDN included. Across perplexity at 4K/32K, MMLU-Pro, GSM8K, AIME'25, GPQA-Diamond, LiveCodeBench, and RULER retrieval to 64K, Minima matches BF16 within seed noise (5-task average -0.52) while being the smallest (17.5 GiB) and fastest-prefill (+14-19%) recipe we compare, and its 32K perplexity gap shrinks with position. A four-part mechanism study explains why: (i) NVFP4's 16-element block scaling localizes the residual stream's extreme outliers, equalizing activation error across layer roles; (ii) the supposedly fragile gate projections are the least sensitive – softplus/exponential and sigmoid parameterizations compress ~11% GEMM error to ~2% output error; (iii) the delta-rule recurrence holds injected noise at a flat plateau over 32K tokens and forgets a state impulse within hundreds of steps, because each write overwrites the state along the current key direction; (iv) the per-token quantization cost washes out with context instead of compounding. We also repair a global-scale mismatch that arises when per-module-calibrated NVFP4 checkpoints are served by kernels that fuse those modules into one GEMM, and show calibrated FP8 KV-cache scales are performance-free. The result: a practical recipe – quantize everything, ship KV scales – and a mechanistic account of why the recurrent half of a hybrid LLM is the easy half to quantize. Checkpoint: https://huggingface.co/minima-ai/mnma_qwen3.8_27b_nvfp4
Read the full abstract
Mesoscale and Nanoscale Physics
Hydrodynamics of two-dimensional electrons due to scattering by disorder
The hydrodynamic regime of electron transport, induced by fast inter-electron collisions, was discovered in high-quality nanostructures in recent ten years. However, signs of hydrodynamic transport, primarily, the giant negative magnetoresistance, were observed even at very low temperatures, when electron-electron scattering is too weak to affect the transport. To address this puzzle, here we develop a theory of mixed, hydrodynamic and non-Markovian, magnetotransport of two-dimensional electrons at zero temperature in samples with weak but still important disorder. Namely, we account for both the memory effects at electron scattering by localized defects in magnetic field and an unconventional viscosity effect due to electron scattering by defects in bulk and by rough sample edges. Solution of the model yields a strong negative magnetoresistance, which exhibits at zero magnetic field a sharp maximum in narrower samples or a blunt maximum in wider samples. This and other our results explain various properties of the giant negative magnetoresistance observed on ultra-high-quality GaAs quantum wells, thereby we apparently reveal the nature of low-temperature magnetotransport in these systems.
Read the full abstract
Robotics
Adaptive Vision-Language Grasping via Composable Foundation Priors and Generalizable Grasp Synthesis
This paper proposes AdaRoboVLG, a task-adaptive Vision-Language-Grasp (VLG) framework that supports generalizable grasp synthesis across different robotic hands. Unlike existing VLG methods that tightly couple foundation models with end-to-end grasp policies, AdaRoboVLG learns an efficient generalizable base policy that generates and evaluates physically feasible grasp candidates through explicit kinematic mapping and force-closure-based stability estimation, while offloading task-dependent understanding to specialized foundation-model modules. These modules provide composable priors that are integrated into the grasp synthesis process, enabling contextually adaptive grasp synthesis without retraining the underlying grasp policy. Through extensive simulation and real-world experiments, we demonstrate that (i) the base policy exhibits efficient learning and strong cross-hand generalization, (ii) the framework effectively incorporates spatial, cognitive, and temporal priors to address three representative grasping challenges without compromising grasp synthesis performance compared to state-of-the-art methods, and (iii) these priors can operate jointly to enable functional grasping in cluttered and dynamic environments. These results indicate that decoupling physical grasp synthesis from task-dependent understanding provides a scalable paradigm for robotic grasping, allowing future advances in foundation models to be directly translated into improved grasp capabilities without redesigning or retraining the underlying grasp policy. Supplementary videos are available at https://adarobovlg.github.io/
Read the full abstract
Algebraic Geometry
Log-concavity and unimodality of Hodge numbers of Hilbert schemes of points over a surface
Let \(S\) be a smooth projective complex surface with irregularity \(q=h^{1,0}(S)\) and geometric genus \(g=h^{2,0}(S)\), and let \(S^{[n]}\) denote its Hilbert scheme of \(n\) points. We prove that, for every \(n\ge0\), the sequence \[ \left(h^{p,0}\bigl(S^{[n]}\bigr)\right)_{p=0}^{2n} \] is log-concave if and only if \(g\le\binom{q+1}{2}\). Moreover, log-concavity of all these sequences is already equivalent to log-concavity of the sequence for \(n=2\). We also prove that these sequences are unimodal for every \(n\ge0\) if and only if \(q\ge1\) or \(g=0\), and that this condition is already detected by the sequence for \(n=1\).
Read the full abstract
Artificial Intelligence
DRACO: Fine-Grained Credit Assignment with Dynamic Rubrics for Long-Horizon Agent Training
Reinforcement Learning from Verifiable Rewards works well when a task has a programmatic checker, but most long-horizon agent domains have none. We work in the outcome-blind setting, where ground-truth success signals are not available. Multi-criteria rubrics are a popular way to supply such a reward; they are scored once per trajectory, but a single scalar is a poor signal across tens of steps. We propose DRACO: Distributing Rubric-based Advantage for Credit Optimization. It generates rubrics dynamically during training to track the policy's evolving capability, scores those rubrics once per completed trajectory, and redistributes that judgment over the steps responsible for annotated rubrics to produce differentiated per-step advantages in GRPO. The redistribution is closed-form and does not introduce any trained attribution module. On AppWorld, DRACO gains 15.9 points over the base model and 5.3 points over GRPO trained with a sparse ground-truth reward, despite not using any verifiers itself. On out-of-domain Tau-Bench, it gains 5.3 points over the base model even without a frontier judge, beating both ground-truth-reward training and other rubric-based training settings. The code for DRACO is available at https://github.com/IBM/draco.
Read the full abstract
Mesoscale and Nanoscale Physics
Extrinsic orbital Edelstein effect from asymmetric scattering
The generation and manipulation of orbital angular momentum (OAM) by an external electric field constitute one of the central themes of orbitronics. In particular, the electrically induced nonequilibrium OAM polarization, known as the orbital Edelstein effect (OEE), has attracted considerable attention in recent years. While the intrinsic band-geometric mechanism and the role of conventional symmetric impurity scattering in the OEE are well understood, the contribution from disorder-induced asymmetric scattering remains unclear. Here, we develop a semiclassical theory that separates the OEE into intrinsic, Drude, side-jump, and third- and fourth-order skew-scattering channels. Unlike the Drude channel, the intrinsic, side-jump, and skew-scattering responses survive only in systems with broken time-reversal symmetry. We find that in a magnetized Rashba two-dimensional electron gas (2DEG), these disorder-induced mechanisms can substantially exceed the intrinsic contribution. Remarkably, we find that for a system with Rashba coupling of $1$ $eV \AA$, the orbital magnetization is about one order of magnitude larger than the spin magnetization for the chosen parameters, highlighting the crucial role of orbital degrees of freedom in the Edelstein effect.
Read the full abstract
Number Theory
Short character sums of inhomogeneous polynomials
Let $p$ be a prime. We prove nontrivial bounds on short sums of Dirichlet characters mod $p$ evaluated at a class of polynomials, not necessarily homogeneous, in $n$ variables and of degree $k$. For large $n$, we further achieve nontrivial bounds for sums over boxes with side-lengths as short as $p^{1/(k-1)+\varepsilon}$, which breaks past the Burgess barrier of $p^{1/4+\varepsilon}$ as soon as $k\geq 6$. In the proof, we develop a new variation of the Burgess amplification method that reduces the problem to bounding additive character sums. This is the first case in which a Burgess-type method succeeds in the inhomogeneous setting.
Read the full abstract
Statistical Mechanics
Effective Hamiltonian description on monitored Majorana chains: correlated power-law hoppings and unconventional entanglement scaling
We investigate the structure of effective Hamiltonians governing monitored dynamics of a one-dimensional Majorana chain using Lyapunov spectral analysis. We focus on the gapless phase of the Lyapunov spectrum, where the effective Hamiltonians behave differently from conventional critical Hamiltonians in isolated systems; the spectral gap closes faster than $1/L$ but slower than $1/L^2$ and the entanglement entropy grows as $[\ln(L)]^2$ where $L$ is the system size. We find that the effective Hamiltonians have random long-range power-law hoppings with nontrivial magnitude correlations, rather than being independently and identically distributed. To elucidate the role of these non-Gaussian correlations, we construct random power-law hopping models that capture the essential features of the effective Hamiltonians. The spectral gaps of the constructed models decay faster than $1/L$ but slower than $1/L^2$. We find that, in the absence of hopping correlations, the ground-state entanglement exhibits $\ln(L)$ scaling. In the presence of correlations, by contrast, the entanglement entropy is enhanced and its system-size dependence is consistent with $[\ln(L)]^2$ scaling over the system sizes studied. These results suggest that correlations among long-range hopping magnitudes are responsible for the entanglement scaling rarely encountered in the ground states of conventional isolated quantum systems.
Read the full abstract
Machine Learning
Conditioning Degenerate Diffusion Models
Current conditioned generative models heavily rely on score functions for guidance during training. When the generative model is a diffusion process with a singular diffusion coefficient and the underlying (conditional) densities either do not exist or are not smooth, we use causal optimal transport to define approximate loss functions that identify a minimum-entropy control for guidance under minimal assumptions. Our approach relies on causal optimal transport and its characterization through the predictable representation property of (conditioned) diffusion processes whose associated martingale problem is well posed, à la Üstünel.
Read the full abstract
Representation Theory
On the new bases attached to Weyl groups of exceptional type
We define the new basis of the Grothendieck group of unipotent representations in an exceptional Weyl group, making more precise and correcting results in previous papers.
Computer Vision
Efficient Semantic Understanding from Digital Foveation
Dense semantic segmentation allocates computational resources uniformly across the entire image, regardless of scene complexity or task relevance. Inspired by biological vision, we investigate whether semantic understanding can be achieved more efficiently through digital foveated perception. We introduce a lightweight active-vision pipeline that combines saliency-driven fixation selection, high-resolution foveal observations, low-resolution contextual information, semantic accumulation, and adaptive computation. Beyond conventional dense prediction metrics, we use object-level evaluation to measure semantic understanding under sparse observations. On ADE20K-Object, a single foveated observation achieves 95.9% of the baseline Top-1 accuracy and 96.9% of the baseline Top-3 accuracy while requiring only 4.7% of the computational cost. At the scene level, semantic accumulation recovers 90.6% of the baseline object recall while using 58.6% of the computation. These results suggest that substantial semantic understanding can emerge from sparse observations when computation is allocated selectively, highlighting active vision as an efficient alternative to uniform dense processing and motivating evaluation protocols beyond conventional pixel-wise segmentation metrics.
Read the full abstract
Computational Finance
Global Multi-Maturity SPX-VIX Calibration Beyond Markovian Stitching
We develop a global martingale optimal transport framework for joint S&P 500 (SPX)-VIX smile calibration across multiple maturities, without the conditional independence that Markovian stitching imposes between months. We prove that exact local and global feasibility are equivalent, yet the stitched laws, which carry earlier months only through the current SPX level, can form a strict subset of the global ones, so adjacent smiles cannot identify the dependence across months. We show that relative entropy against the standard Markov reference selects the stitched law, so this dependence must be supplied deliberately. We supply it through a finite-memory reference, in which a latent autoregressive state drives the VIX levels while every VIX marginal stays exact, and calibrate the resulting law on a junction tree whose size grows linearly with the number of maturities. On market data with up to nine SPX maturities, a law with memory and a law without it fit the same quotes but price VIX-conditioned SPX Asian claims differently: on every calendar, memory makes a claim on the maximum VIX cheaper and a claim on the minimum VIX dearer. Varying the memory strength on one calendar moves these differences steadily from zero, so the vanilla quotes, which fix the law of each date, do not determine prices that depend on the joint law across dates.
Read the full abstract
Cryptography and Security
A Non-Formulable Theorem: A Fundamental Limit of Finite Syntactic Systems and Its Consequences for Security and AI
For every coherent and sufficiently expressive finite syntactic system S, we prove the existence of at least one theorem that S cannot produce autonomously. The result is a metatheorem: it proves the existence of a theorem, and applies to every finite syntactic system - security mechanisms, AI systems, formal verifiers, legal systems, economic models, and the formal system in which it is itself proved.
Read the full abstract
Audio and Speech Processing
Deep Neural Compression for RIR-Characterized Acoustic Environments with Structure-Aware Constraints
Room impulse responses (RIRs) characterize the acoustic environment of a room by capturing how sound propagates and decays within an enclosed space. In applications such as immersive audio rendering, accurate acoustic reconstruction often relies on spatially densely sampled RIRs. This consequently gives rise to a large volume of RIR data, imposing a substantial burden on storage. Although recent neural audio codecs provide an effective framework for low-bitrate compression, their training objectives are mainly tailored to speech and general audio, and are therefore not well aligned with the acoustic characteristics of RIRs. Therefore, we propose an EnCodec-based neural RIR compression method, which incorporates RIR structure-aware constraints at two levels. Specifically, at the RIR level, structure-aware constraints are imposed on the global decay behavior and local energy distribution of RIRs through energy decay curve (EDC) regularization and a short-time window energy constraint, while at the reverberant-speech level, reverberant-speech supervision is further introduced to constrain the consistency of the reverberant speech generated by the reconstructed RIRs. Experimental results show that, at a low bitrate of 375 bps, the proposed method achieves lower RIR reconstruction error and better reverberant-speech perceptual consistency than audio-oriented codecs.
Read the full abstract
Optimization and Control
Transversality Conditions for Boundary Constraints Defined by Differential Equations
What are the transversality conditions for an optimal control problem when the boundary conditions are defined by differential equations? This seemingly bizarre question is motivated by trajectory optimization problems in the $N$-body system. The question, however, is more fundamental and goes beyond problems in astrodynamics to nonintegrable dynamical systems in general. The main contribution of this paper is the development of generic initial- and final-time transversality conditions for optimal control problems whose boundary conditions are defined in terms of differential equations with side conditions. The mathematical definition of differential boundary conditions are part of the foundations developed in this paper. To support the new fundamentals, the concept of coordinated/uncoordinated clock times and weak adjoint covectors are introduced. In the case of uncoordinated clock times, the new transversality conditions reveal that there exists a special situation where a weak adjoint covector is orthogonal to the vector field of the boundary differential equation. This condition is sharply different from the classical statement of orthogonality with respect to the endpoint manifold. The theorems developed in this paper are generic. An application of the theorems to several cases in the three-body problem are described in separate papers.
Read the full abstract
Computer Vision
CORE: Improving Compositional Reasoning in MLLM Embedding via Reranker Distillation
MLLM-based embedding models remain limited in compositional retrieval, often failing to distinguish scenes containing the same concepts but different attribute-object bindings. Yet the same backbone can resolve such distinctions when used as a cross-attentive reranker, motivating us to distill its compositional judgments into the embedding model. We propose CORE, which synthesizes candidate lists spanning five compositional matching levels and introduces a Rank-KL objective that trains the embedding model to reproduce the reranker's fine-grained ranking. We further introduce a graded evaluation protocol and compare contrastive learning, pairwise CoSENT, and listwise Rank-KL under the same data and tuning budget. Our comparison shows that both CoSENT and Rank-KL use the multi-level supervision more effectively than contrastive learning, with Rank-KL achieving the strongest overall performance. Across three compositional reasoning benchmarks (COLA, SUGARCREPE++, NEGBENCH), CORE-RERANKER-8B achieves an 82.7% total average, outperforming Jina-Reranker by 10.7 points, while CORE-EMBED-8B achieves the best total average (0.666) among all evaluated embedding models. The improvements transfer to the MCMR benchmark without sacrificing retrieval performance on COCO and Flickr30K.
Read the full abstract
Statistical Methodology
Model-assisted estimation with a training subsample: a two-phase sampling approach with design-based variance estimation
When a flexible prediction model is fitted on a training subsample drawn from a probability sample, the model-assisted estimator actually reported arises from one realized partition, yet existing theory quantifies uncertainty only for partition-averaged, cross-fitted, or symmetrized versions of it. We represent the training subsample as a second phase of sampling and derive, exactly and for any algorithm, a two-term variance decomposition and the variance family linking the single-partition estimator to its Rao-Blackwellized average, whose design bias it shares. For cell-mean rules under equal-probability designs the estimator is, exactly, a convex combination of a poststratified plug-in at the training-sample size and the exactly unbiased single-split estimator of Sanguiao Sande and Zhang, which locates its bias precisely. For tree-type predictors the second-phase variance is computable in closed form when the cell structure is fixed or $s$-measurable, and in that regime, for cell counts $H=o(n)$, its share of total variance grows with the number of cells – a design-based mechanism distinct from residual shrinkage and consistent with the variance underestimation documented for flexible learners. We propose an analytic and a replication variance estimator, neither altering the point estimate, and evaluate them by simulation under simple random and stratified designs, in the adaptive regime and in an $s$-measurable configuration in which the theory is exact: in the populations studied, budgeting the second phase recovers much of the coverage lost by ignoring it, and the residual shortfall is attributable to the bias of the point estimator rather than to its variance.
Read the full abstract
Quantum Physics
Frequency-Multiplexed Parallel Gates for Quantum LDPC Codes in a Two-Dimensional Ion Crystal
Quantum low-density parity-check (qLDPC) codes admit high encoding rates but require nonlocal entangling gates for syndrome measurement. Instead of physically moving the qubits which slows down with the increasing qubit number, here we propose to achieve parallel nonlocal entangling gates on a two-dimensional (2D) ion crystal using frequency-multiplexing. Adiabatic conditions ensure the suppression of gate infidelity and crosstalk error, as well as their robustness against slow drift in the trap frequency which is a leading error source in ion trap. We consider a numerical example of a $[[248,10,18]]$ bivariate bicycle code on a 2D crystal of 512 ions. By optimizing the mapping of the qubits and the assignment of the frequency bands for multiplexing, we show that a moderate laser power is sufficient for parallelism, and that a logical error rate of $10^{-12}$ can be achieved under realistic noise parameters.