arXiv:2601.18904v2 Announce Type: replace
Abstract: Auditory Large Language Models (LLMs) have demonstrated strong performance across a wide range of speech and audio understanding tasks. Nevertheless, they often struggle when applied to low-resource tasks. In case in-domain labeled data are scarce or mismatched with the true test distribution, direct fine-tuning can be brittle. In-Context Learning (ICL) provides a training-free, inference-time solution by adapting auditory LLMs through conditioning on a few in-domain demonstrations. In this work, we first show that $\textit{Vanilla ICL}$, improves zero-shot performance across diverse speech and audio tasks for selected models which suggest that this ICL adaptation capability can be generalized to multimodal setting. Building on this, we propose $\textbf{Meta Speech In-Context Learning (MetaSICL)}$, a post-training recipe utilizes only high resource speech data from various tasks intending to strengthen model's in-context learning capability. Experiments indicate our proposed method outperforms direct fine-tuning in low-resource scenario.
Science Journals
AI-based separation of turbulence from coherent background flows in decaying hydrodynamic turbulence
arXiv:2601.18163v2 Announce Type: replace
Abstract: Separating turbulent fluctuations from coherent large-scale background flows is a longstanding challenge in the analysis of numerical simulations and astronomical observations. Traditional approaches commonly rely on decomposition-based techniques such as Fourier or wavelet filtering, which assume that a meaningful separation can be achieved through scale selection. In realistic flows, however, coherent motions and turbulence often overlap across a broad range of scales and interact nonlinearly, making a unique separation inherently ambiguous. In this work, we investigate the robustness of an AI-based turbulence-background separation approach using two-dimensional incompressible Navier-Stokes simulations of decaying hydrodynamic turbulence. The simulations are initialized with a coherent background flow and divergence-free turbulent perturbations with a Kolmogorov-like spectrum and evolve without external forcing, providing a controlled physical testbed. A neural network trained exclusively on static synthetic images is applied to simulation snapshots at different evolutionary stages. The model recovers turbulent fluctuations during early and intermediate stages when partial scale separation is present. At later stages, where nonlinear interactions increasingly mix coherent and turbulent structures, the separation becomes less distinct; nevertheless, the recovered fields remain visually and spectrally consistent with the expected turbulent behavior. Quantitative comparisons with a Fourier filtering baseline show that the AI-based approach achieves comparable reconstruction accuracy while not requiring an explicit spectral cutoff scale. These results suggest that AI models trained on static data can provide a flexible diagnostic tool for turbulence-background separation in time-evolving flows, with potential applications to astrophysical datasets.
arXiv:2601.06580v2 Announce Type: replace
Abstract: Singlish is a creole rooted in Singapore's multilingual environment that continues to evolve alongside social and technological change. We examine diachronic stylistic change across a decade of informal digital messages and ask whether Large Language Models (LLMs) can generate temporally neutral outputs approximating the stable essence of the variety. Using lexical, pragmatic, psycholinguistic, and encoder-based features, we find that stylistic separability increases with temporal distance, driven primarily by structural features such as length and complexity. Evaluated against a null distribution baseline, most LLMs fail to achieve both authenticity and temporal neutrality simultaneously, revealing a structural trade-off: models generating realistic Singlish inherit its temporal biases, while temporally neutral models produce inauthentic outputs. These findings position temporal neutrality as a diagnostic metric for assessing sociolectal grounding in LLMs.
arXiv:2512.11280v2 Announce Type: replace
Abstract: Large language models (LLMs) have achieved remarkable performance across a wide range of tasks, but their increasing parameter sizes significantly slow down inference. Speculative decoding mitigates this issue by leveraging a smaller draft model to predict candidate tokens, which are then verified by a larger target model. However, existing approaches often require additional training, extensive hyperparameter tuning, or prior analysis of models and tasks before deployment. In this paper, we propose Adaptive Speculative Decoding (AdaSD), a hyperparameter-free decoding scheme that dynamically adjusts generation length and acceptance criteria during inference. AdaSD introduces two adaptive components: one to determine when to stop candidate token generation and the other to decide token acceptance, updated in real time based on token entropy and Jensen-Shannon distance. This approach eliminates the need for pre-analysis or fine-tuning and is compatible with off-the-shelf models. Experiments on benchmark datasets demonstrate that AdaSD achieves up to 1.46x speedup over vanilla speculative decoding while limiting accuracy degradation to under 1.8%, making it a practical solution for efficient and adaptive LLM inference.
arXiv:2512.06609v3 Announce Type: replace
Abstract: Vector-quantized variational autoencoders (VQ-VAEs) are discrete autoencoders that compress images into discrete tokens. However, they are difficult to train due to discretization. In this paper, we propose a simple yet effective technique dubbed Gaussian Quant (GQ), which first trains a Gaussian VAE under certain constraints and then converts it into a VQ-VAE without additional training. For conversion, GQ generates random Gaussian noise as a codebook and finds the closest noise vector to the posterior mean. Theoretically, we prove that when the logarithm of the codebook size exceeds the bits-back coding rate of the Gaussian VAE, a small quantization error is guaranteed. Practically, we propose a heuristic to train Gaussian VAEs for effective conversion, named the target divergence constraint (TDC). Empirically, we show that GQ outperforms previous VQ-VAEs, such as VQGAN, FSQ, LFQ, and BSQ, on both UNet and ViT architectures. Furthermore, TDC also improves previous Gaussian VAE discretization methods, such as TokenBridge. The source code is provided in https://github.com/tongdaxu/VQ-VAE-from-Gaussian-VAE.
arXiv:2511.20092v2 Announce Type: replace
Abstract: Spurious numerical mixing is a frequent phenomenon in ocean models. In this paper, we present an efficient and robust methodology that defines the vertical grid motion so that this mixing is reduced. This motion is defined as the solution of an optimization problem that -- using the ideas of the calculus of variations -- results in an elliptic partial differential equation, which is straightforward to analyze and discretize. This framework is generally applicable to any ocean model that uses an Arbitrary Lagrangian-Eulerian (ALE) vertical coordinate and can be tuned to fit the modeler's specific needs based on the guidelines presented herein. The method is applied to the nonhydrostatic solver presented by the authors in [Alexandris-Galanopoulos et al., 2024].
While the majority of spurious numerical mixing studies focus on large-scale processes, herein the proposed method is applied and tested in small-scale nonhydrostatic phenomena. Specifically, the effectiveness of the method in capturing fully nonlinear internal waves is investigated for the test cases of wave propagation, breaking and overturning. Overturning serves as a demanding test for the proposed scheme as it induces rapid vertical accelerations and thus the mesh-moving algorithm must incorporate this motion with the goal of reducing numerical mixing, while not suppressing physically relevant vertical mass transfer. These numerical benchmarks show the ability of the method to reduce spurious mixing, while attaining the physical relevance of the results.
arXiv:2605.27178v1 Announce Type: new
Abstract: We address the challenging task of 3D object segmentation in complex scene point clouds without relying on any scene-level human annotations during training. Existing methods are typically constrained to identifying simple objects, primarily due to insufficient object priors in the learning process. In this paper, we present FoundObj, a novel framework featuring a superpoint-based object discovery agent that incrementally merges suitable neighboring superpoints, guided by our innovative semantic and geometric reward modules. These modules synergistically leverage semantic and geometric priors from self-supervised 2D/3D foundation models, providing complementary feedback to the object discovery agent and enabling robust identification of multi-class objects through reinforcement learning. Extensive experiments on diverse benchmarks demonstrate that our approach consistently outperforms existing baselines. Notably, our method exhibits strong generalization in zero-shot and long-tail scenarios, underscoring its potential for scalable, label-free 3D object segmentation.
arXiv:2511.03175v2 Announce Type: replace
Abstract: Computer generated holography requires high-speed spatial light modulators (SLMs) for dynamically patterning light in 3D. Piston-motion micromirror-based SLMs support high-speed ($\geq$ 10 kHz) phase modulation; however, fabricating micromirror arrays with sufficient fill factor necessary for high diffraction efficiency is challenging. In particular, the larger mirrors of high fill factor designs are susceptible to stress-induced curvature that significantly degrades optical performance. In this work, we introduce an optical compensation method using a pitch-matched microlens array (MLA) to focus light onto just the center of each mirror. Our approach thus avoids curvature-induced artifacts and improves optical fill factor to nearly 100$\%$, independent of the original mechanical fill factor. Through simulations and experiments on a fabricated micromirror array with bowed mirrors, we show that the Pearson correlation coefficient of the imparted phase profile is improved from 0.11 to 0.85 and the brightness of a holographically-generated single spot is enhanced by 8$\times$ with our microlens array in place. Our hybrid optical-electromechanical strategy thus provides a scalable path toward high-speed, high-fidelity wavefront control for applications such as adaptive optics, holographic displays, and optogenetics.
arXiv:2511.02525v2 Announce Type: replace
Abstract: The capacitated location-routing problems (CLRPs) are classical problems in combinatorial optimization, which require simultaneously making location and routing decisions. In CLRPs, the complex constraints and the intricate relationships between various decisions make the problem challenging to solve. With the emergence of deep reinforcement learning (DRL), it has been extensively applied to address the vehicle routing problem and its variants, while the research related to CLRPs still needs to be explored. In this paper, we propose the DRL with heterogeneous query (DRLHQ) to solve CLRP and open CLRP (OCLRP), respectively. We are the first to propose an end-to-end learning approach for CLRPs, following the encoder-decoder structure. In particular, we reformulate the CLRPs as a markov decision process tailored to various decisions, a general modeling framework that can be adapted to other DRL-based methods. To better handle the interdependency across location and routing decisions, we also introduce a novel heterogeneous querying attention mechanism designed to adapt dynamically to various decision-making stages. Experimental results on both synthetic and benchmark datasets demonstrate superior solution quality and better generalization performance of our proposed approach over representative traditional and DRL-based baselines in solving both CLRP and OCLRP.
arXiv:2605.26403v1 Announce Type: new
Abstract: A long-standing goal of the research community is to develop highly interactive LLM-based dialogue agents. Recent research focuses on optimizing policies based on fixed offline logs (Static Context RL) or using a prompt-based simulator (Interactive RL). In this work, we theoretically show that both paradigms are fundamentally limited by context distribution shift--a mismatch between dialogue histories observed during training and those encountered in real conversations. This shift compounds quadratically over turns and severely degrades dialogue quality. Specifically, we attribute this shift to two distinct sources: (i) policy-induced shift, arising from training on static histories rather than self-generated trajectories; and (ii) simulator-induced shift, stemming from discrepancies between simulated and real human behaviors. To address these challenges, we propose Calibrated Interactive RL, a unified framework that couples interactive RL with simulator alignment. By aligning the simulator with human interaction patterns, our approach reduces the sim-to-real gap and mitigates compounding distribution shifts. Experiments across multiple dialogue tasks confirm our theoretical analysis: (i) Interactive RL significantly outperforms the Static Context baseline by mitigating policy distribution shift; and (ii) calibrating simulators with our alignment method further bridges the sim-to-real gap, yielding state-of-the-art downstream performance.
arXiv:2511.01724v3 Announce Type: replace
Abstract: Deep learning models are notoriously vulnerable to imperceptible perturbations. Most existing research centers on adversarial robustness (AR), which evaluates models under worst-case scenarios by examining the existence of deterministic adversarial examples (AEs). In contrast, probabilistic robustness (PR) adopts a statistical perspective, measuring the probability that predictions remain correct under stochastic perturbations. While PR is widely regarded as a practical complement to AR, dedicated training methods for improving PR are still relatively underexplored, albeit with emerging progress. Among the few PR-targeted training methods, we identify three limitations: i non-comparable evaluation protocols; ii limited comparisons to strong AT baselines despite anecdotal PR gains from AT; and iii no unified framework to compare the generalization of these methods. Thus, we introduce PRBench, the first benchmark dedicated to evaluating improvements in PR achieved by different robustness training methods. PRBench empirically compares most common AT and PR-targeted training methods using a comprehensive set of metrics, including clean accuracy, PR and AR performance, training efficiency, and generalization error (GE). We also provide theoretical analysis on the GE of PR performance across different training methods. Main findings revealed by PRBench include: AT methods are more versatile than PR-targeted training methods in terms of improving both AR and PR performance across diverse hyperparameter settings, while PR-targeted training methods consistently yield lower GE and higher clean accuracy. A leaderboard comprising 229 trained models across 7 datasets and 10 model architectures is publicly available at https://wellzline.github.io/PRBenchLeaderboard/.
arXiv:2510.08932v2 Announce Type: replace
Abstract: Recently, a growing body of research has focused on either optimizing CTR model architectures to better model feature interactions or refining training objectives to aid parameter learning, thereby achieving better predictive performance. However, previous efforts have primarily focused on the training phase, largely neglecting opportunities for optimization during the inference phase. Infrequently occurring feature combinations, in particular, can degrade prediction performance, leading to unreliable or low-confidence outputs. To unlock the predictive potential of trained CTR models, we propose a Model-Agnostic Test-Time paradigm (MATT), which leverages the confidence scores of feature combinations to guide the generation of multiple inference paths, thereby mitigating the influence of low-confidence features on the final prediction. Specifically, to quantify the confidence of feature combinations, we introduce a hierarchical probabilistic hashing method to estimate the occurrence frequencies of feature combinations at various orders, which serve as their corresponding confidence scores. Then, using the confidence scores as sampling probabilities, we generate multiple instance-specific inference paths through iterative sampling and subsequently aggregate the prediction scores from multiple paths to conduct robust predictions. Finally, extensive offline experiments and online A/B tests strongly validate the compatibility and effectiveness of MATT across existing CTR models.
arXiv:2605.27189v1 Announce Type: new
Abstract: This study examines the relationship between speech representations and the hierarchical structure of cognitive assessment in mild cognitive impairment. Utilizing 5,754 German neuropsychological assessment recordings, we evaluate six cognitive tasks across three score levels: task, domain, and global levels. We compare hand-crafted acoustic features with self-supervised learning (SSL) embeddings. Results show that although SSL representations generally outperform hand-crafted features at lower levels, this trend reverses for MCI classification. Furthermore, task-specific constraints influence performance: tasks with greater response freedom exhibit performance dilution as hierarchical levels increase, suggesting ``specialist'' representations, whereas the performance of highly structured tasks increases toward higher levels, suggesting ``generalist'' representations. These findings show links between task constraints and assessment hierarchy in automated clinical speech analysis.
arXiv:2605.26406v1 Announce Type: new
Abstract: 3D reconstruction techniques such as LiDAR scanning and photogrammetry have made it practical to build detailed geometric models of real-world environments. Such reconstructed models can potentially serve as the foundation for wireless digital twins and support network planning and optimization. The core challenge is that reconstructed models inevitably contain geometric artifacts such as holes and noisy surfaces, and wireless simulation is highly sensitive to such noise. To solve this problem, we propose a differentiable directional scattering model to approximate the noise-sensitive specular reflection. This approximation smoothly distributes reflected power among nearby ray directions, making the simulator inherently robust to local geometric artifacts in the reconstructed model. We prove mathematically that this approximation preserves asymptotic path-gain accuracy. Building on this idea, we propose mmDiff, an end-to-end differentiable framework for calibrating material properties from sparse mmWave measurements and predicting mmWave channels. We evaluate mmDiff on both real-world and synthetic datasets, and demonstrate its superior performance over prior methods using pure specular reflection in noisy reconstructed geometry.
arXiv:2605.27192v1 Announce Type: new
Abstract: Automata acceptance can, in several situations of interest, be captured game-theoretically via acceptance games. The existence of a winning strategy for Verifier then captures the existence of a winning run-tree of a given automaton over a model. However, such acceptance is rigid, in that it does not allow a measurable defect budget, which can be a challenge in software verification. In this paper, we draw inspiration from how bisimulation distance can be defined as an extension of bisimilarity to define epsilon-acceptance games. Our main theorem shows that a tree T is epsilon-accepted iff there is a tree T' that is accepted in the traditional (rigid) sense and the bisimulation distance of T' and T is at most epsilon. Our work also suggests a strong connection with measure theory, of which we give a preliminary exploration via appropriate examples. Our framework is defined over binary trees with leaves and infinite branches, and strictly contains the case in which binary nodes are seen as probabilistic choice and the defect measures the probability of the set of rejected branches.
arXiv:2510.07231v4 Announce Type: replace
Abstract: Socio-economic causal effects depend heavily on their institutional and environmental contexts. The same intervention can produce different, even opposite, effects across regulatory regimes, market conditions, time periods, or populations. This poses a challenge for large language models (LLMs) in decision-support roles: can they infer the direction of a causal effect under a specified context, and revise that judgment when the context changes?
To address this, we introduce EconCausal, a large-scale benchmark of 10,490 context-annotated causal triplets extracted from 2,595 high-quality empirical studies in top-tier economics and finance journals, constructed through a rigorous four-stage pipeline with multi-run consensus, context refinement, and multi-critic filtering.
Across models, LLMs often fail to condition their predictions on context. While top models reach 88% accuracy in fixed, explicit contexts, accuracy falls by 32.6~pp on cases that require revising the sign across contexts (73.9% to 41.3%), and drops below 50% once misleading signed evidence is introduced. Models also over-commit to directional (+/-) signs, recognizing null effects only 13.8% of the time while remaining poorly calibrated on these categories. The dataset and benchmark are publicly available at https://anonymous.4open.science/r/econcausal-benchmark-6F12.
arXiv:2510.06133v3 Announce Type: replace
Abstract: Diffusion large language models (dLLMs) generate text through iterative denoising. In commonly adopted parallel decoding schemes, each step confirms only high-confidence positions while remasking the others. By analyzing dLLM denoising traces, we uncover a key inefficiency: models often predict the correct target token several steps before its confidence becomes high enough to be decoded. This gap between early prediction and late decoding forces repeated remasking of already-correct tokens, causing redundant iterations and limiting acceleration. To exploit this temporal redundancy, we introduce Trace Credit to quantify a token's decoding potential by accumulating historical evidence. Building on this, we propose CreditDecoding, a training-free parallel decoding method that fuses Trace Credit with current logits to boost the confidence of correct but underconfident tokens, thereby accelerating denoising and improving robustness. On eight benchmarks, CreditDecoding achieves up to 5.48 times speedup with +0.48 accuracy on LLaDA-8B and consistently improves performance across diverse dLLM architectures and parameter scales. It further scales to long contexts and remains orthogonal to mainstream inference optimizations, making it a practical and widely applicable solution.
arXiv:2510.05141v2 Announce Type: replace
Abstract: When we read, we make predictions about upcoming words; these predictions influence our reading behavior. The success of large language models (LLMs), which, like humans, make predictions about upcoming words, has motivated their use as models of human linguistic prediction. Surprisingly, in the last few years, as LLMs' ability to predict the next word has improved, their ability to explain reading behavior has declined. We argue this is because current LLMs can predict upcoming words much better than human readers can. This 'superhumanness' is driven by LLMs' extensive training data, stronger long-term memory of training examples, and stronger short-term memory. We advocate for LLMs with human-like memory and for new experiments to measure the alignment between humans and LLMs, and outline directions towards achieving these goals.
arXiv:2605.26416v1 Announce Type: new
Abstract: Classical Proper Orthogonal Decomposition (POD)-based Galerkin projection models of chaotic flows typically require a large number of modes as well as stabilization or closure terms to achieve adequate accuracy and long-term stability. We present a novel differentiable programming framework that stabilizes low-rank POD-Galerkin models without increasing the number of modes or introducing additional closure terms, thereby delivering both high efficiency and high accuracy. Model stabilization is achieved by tuning the linear and quadratic tensors in the POD-Galerkin using differentiable programming, trained on short-term trajectory data. A key finding of this study is that a purely point-wise trajectory-based loss function yields poor long-term accuracy for chaotic systems. In contrast, a hybrid loss function that combines trajectory error with a physics-based conservation-of-energy term provides superior long-term performance. We demonstrate the approach on a chaotic lid-driven cavity flow at Re = 30,000. The stabilized ROM achieves an order-of-magnitude reduction in computational cost compared with the classical POD-Galerkin method: it remains accurate and stable with only 20 modes, whereas the classical ROM requires 80 POD modes.
arXiv:2605.27204v1 Announce Type: new
Abstract: Scientific paper evaluation often involves not only assessing a manuscript itself, but also relating it to contemporaneous research and prior literature. However, existing LLM-based methods typically model these signals separately and lack a unified mechanism for propagating review evidence across papers. We propose $\textbf{GraphReview}$, a graph-based LLM framework that formulates paper evaluation as review-signal message passing over a semantic paper graph. The graph jointly captures intrinsic quality, synchronic links among contemporaneous papers, and diachronic links to prior work. LLMs are used to estimate node-level quality priors and generate edge-level comparative evidence through pairwise paper comparisons, while Personalized PageRank integrates review signals for quality ranking, decision prediction, and review generation. To produce higher-quality graph evidence, we propose reward-induced maximum likelihood objectives for training the LLM backbones. Experiments show that GraphReview consistently outperforms the strongest baseline, achieving average improvements of 29.7% on decision and ranking metrics, including gains of 23.7% in Accuracy and 57.6% in Spearman's $\rho$. It also produces higher-quality review texts and generalizes effectively across time periods and conference venues. The code is available at https://github.com/ECNU-Text-Computing/GraphReview.
arXiv:2510.04533v2 Announce Type: replace
Abstract: Diffusion models achieve state-of-the-art image generation but often produce semantic inconsistencies, or hallucinations. Existing inference-time guidance methods rely on external signals or architectural modifications, adding computational overhead. We propose $\mathbf{T}$angential $\mathbf{A}$mplifying $\mathbf{G}$uidance $\mathbf{(TAG)}$, a training-free, architecture-agnostic, plug-and-play guidance method that operates purely on trajectory signals. TAG uses an intermediate sample as a projection basis and amplifies the tangential components of the estimated score to correct the sampling trajectory. A first-order Taylor analysis shows that this steers the state toward higher-probability regions of the data manifold, reducing inconsistencies and improving fidelity while adding negligible overhead to existing samplers. Code is available at our Project Page (https://hyeon-cho.github.io/TAG/).
arXiv:2605.27209v1 Announce Type: new
Abstract: Recent advances in large language models (LLMs) have facilitated the widespread deployment of LLMs as interactive agents capable of reasoning, planning, and tool use. Despite strong performance on existing benchmarks, such agents often exhibit notable degradation when deployed in real-world settings, where environments are inherently stochastic and imperfect. We argue that this discrepancy arises from a fundamental mismatch between idealized training settings and real-world interaction dynamics, where current paradigms rely on carefully curated task instructions and stable, well-controlled environments. To address this gap, we propose NoisyAgent, an agentic training framework that explicitly incorporates environmental imperfections into the agent learning process. We identify two major sources of interaction noise in real-world scenarios: user noise, which captures ambiguity and variability in user interaction, and tool noise, which reflects failures and anomalies in tool execution. We introduce such perturbations into the training pipeline by modifying user interaction patterns and simulating tool execution results within the training environment. To stabilize training while encouraging agents to handle increasingly challenging imperfections, noise is applied to only a subset of rollouts and progressively increased in difficulty as the model adapts to the current noise level. Extensive experiments demonstrate that our approach consistently improves agent robustness under noisy and dynamic environments. Our analysis reveals that training under noise conditions also yields performance gains on idealized benchmarks, suggesting that controlled exposure to environmental noise promotes more generalizable reasoning and decision-making behaviors. Our findings highlight the importance of modeling interaction imperfections for bridging the gap between agent training and real-world deployment.
arXiv:2510.03352v3 Announce Type: replace
Abstract: Diffusion models have been used as priors for solving inverse problems. However, existing approaches typically overlook side information that could significantly improve reconstruction quality, especially in severely ill-posed settings. In this work, we propose a novel framework that incorporates side information into existing diffusion-based inverse problem solvers via inference-time search, in a plug-and-play, training-free manner. Through extensive experiments across a range of inverse problems, including inpainting, super-resolution, and several deblurring tasks, and across multiple diffusion-based inverse problem solvers (DPS, DAPS, and MPGD), we show that augmenting each solver with our framework consistently improves the quality of the reconstructions over the corresponding original method. To demonstrate the generality of our approach, we consider diverse forms of side information, including reference images, textual descriptions, and anatomical MRI scans. The code is available at this \href{https://github.com/mahdi-farahbakhsh/DISS}{repository}\footnote{https://github.com/mahdi-farahbakhsh/DISS}.
arXiv:2510.01481v2 Announce Type: replace
Abstract: We present the Social Influence Game (SIG), a framework for modeling adversarial persuasion in social networks with an arbitrary number of competing players. Our goal is to provide a tractable and interpretable model of contested influence that scales to large systems while capturing the structural leverage points of networks. Each player allocates influence from a fixed budget to steer opinions that evolve under DeGroot dynamics, and we prove that the resulting optimization problem is a difference-of-convex program. To enable scalability, we develop an Iterated Linear (IL) solver that approximates player objectives with linear programs. In experiments on random and archetypical networks, IL achieves solutions within 7% of nonlinear solvers while being over 10x faster, scaling to large social networks. This paper lays a foundation for asymptotic analysis of contested influence in complex networks.
arXiv:2510.01336v2 Announce Type: replace
Abstract: Speculative decoding accelerates LLM inference by using a smaller draft model to speculate tokens that a larger target model verifies. Verification is often the bottleneck (e.g. verification is $4\times$ slower than token generation when a 3B model speculates for a 70B target model), but most prior works focus only on accelerating drafting. $\textit{``Intermediate"}$ verification reduces verification time by discarding inaccurate draft tokens early, but existing methods incur substantial training overheads in incorporating the intermediate verifier, increase the memory footprint to orchestrate the intermediate verification step, and compromise accuracy by relying on approximate heuristics.
We propose $\underline{\textit{Hi}}\textit{erarchical }\underline{\textit{Spec}}\textit{ulative Decoding (HiSpec)}$, a framework for high-throughput speculative decoding that exploits $\textit{early-exit (EE) models}$ for low-overhead intermediate verification. EE models allow tokens to exit early by skipping layer traversal and are explicitly trained so that hidden states at selected layers can be interpreted, making them uniquely suited for intermediate verification without drastically increasing compute and memory overheads. To improve resource-efficiency even further, we design a methodology that enables HiSpec to re-use key-value caches and hidden states between the draft, intermediate verifier, and target models. To maintain accuracy, HiSpec periodically validates the draft tokens accepted by the intermediate verifier against the target model. Our evaluations using various representative benchmarks and models show that HiSpec improves throughput by 1.28$\times$ on average and by up to 2.01$\times$ compared to the baseline single-layer speculation without compromising accuracy.