Academic Intelligence · Curated Daily

探索全球前沿学术脉络

AcademicHub 汇聚顶级期刊与预印本平台的实时文献。定制您的专属科研雷达,利用大语言模型自动生成交叉领域文献分析简报。

01.
arXiv (quant-ph) 2026-06-11

Planted-Solution Pauli Hamiltonians as a Quantum Benchmarking Primitive

arXiv:2606.11455v1 Announce Type: new Abstract: We introduce a construction of Pauli Hamiltonians with exactly known ground-state energies, intended as reference instances for ground-state energy estimation algorithms. The construction embeds a planted block-product state as the simultaneous ground state of a sum of frustration-free local clauses on overlapping supports, exposes the resulting model only as a polynomial-size linear combination of Pauli operators, and admits optional Clifford conjugation that preserves the spectrum. The framework subsumes classical planted constraint-satisfaction problems as a diagonal special case, providing a direct embedding channel through which classical hardness properties can be inherited. Open-source software, certification keys, and example instances are made publicly available.

02.
arXiv (CS.CL) 2026-06-18

STARE: Surprisal-Guided Token-Level Advantage Reweighting for Policy Entropy Stability

Reinforcement Learning with Verifiable Rewards algorithms like GRPO have emerged as the dominant post-training paradigm for complex reasoning in LLMs, yet commonly suffer from policy entropy collapse during training. We conduct a first-order gradient analysis of token-level entropy dynamics under GRPO and identify a token-level credit assignment mismatch: the per-token entropy variation decomposes into the product of the trajectory-level advantage and an entropy sensitivity function over the next-token distribution, yielding an advantage-surprisal four-quadrant structure and a near-criticality property. Motivated by it, we propose STARE (Surprisal-guided Token-level Advantage Reweighting for policy Entropy stability), which identifies entropy-critical token subsets via batch-internal surprisal quantiles, selectively reweights their effective advantages, and incorporates a target-entropy closed-loop gate for stable entropy regulation. Across model scales from 1.5B to 32B and three task families (Short CoT, Long CoT, and Multi-Turn Tool Use), STARE sustains stable RL training over thousands of steps while maintaining policy entropy within the target band. On AIME24 and AIME25, STARE outperforms DAPO and other competitive baselines by 4%-8% in average accuracy, with reflection tokens and response length growing in tandem, indicating sustained exploration-exploitation balance that further unlocks RL training potential.Code is available at https://github.com/hp-luo/STARE.

03.
arXiv (CS.CV) 2026-06-12

Analyzing and Improving Fine-grained Preference Optimization in Medical LVLMs

Large Vision-Language Models (LVLMs) have achieved strong performance across medical imaging tasks, yet they remain prone to factual inconsistencies, poor visual grounding, and misalignment with clinically meaningful feedback. Existing post-training alignment approaches, including Direct Preference Optimization (DPO) and its variants, face three critical limitations in the medical domain: (1) sequence-level reward signals treat clinically critical tokens identically to generic filler text; (2) reliance on static supervised fine-tuning references as preferred responses introduces an off-policy distribution shift, steering optimization toward stylistic artifacts over clinical correctness; and (3) alignment objectives lack explicit visual grounding constraints, leaving models insensitive to subtle yet diagnostically decisive pathological features. Our method leverages a bidirectional token-wise KL regularizer alongside a visual-contrastive grounding objective that pairs clean and lesion-corrupted images to penalize responses generated without adequate visual evidence. Together, these components form a fine-grained, on-policy alignment framework that constructs preference pairs by minimally editing model-generated outputs, correcting only clinically erroneous spans while preserving the original linguistic style. Extensive experiments across medical imaging tasks and clinical text generation benchmarks validate the effectiveness of our approach.

04.
arXiv (CS.AI) 2026-06-19

Beyond Static Leaderboards: Predictive Validity for the Evaluation of LLM Agents

arXiv:2606.19704v1 Announce Type: new Abstract: Agent benchmarks are growing fast, but no single benchmark touches more than four or five of the dimensions that deployment exposes. This paper aggregates the largest coordinated deep-dive of one MCP-based industrial-agent benchmark to date: fourteen parallel implementation studies covering new asset classes (including a multi-modal visual extension), alternative orchestrations, retrieval strategies, reasoning modes, infrastructure optimizations, and evaluation-methodology probes. Consolidating those studies with seven prior agent benchmarks, we argue that aggregate-score leaderboards systematically underspecify deployed-agent evaluation. Rankings derived from aggregate scores do not transfer to out-of-distribution settings; recent public-to-hidden competition retrospectives provide direct empirical evidence of this rank instability. We propose ranking configurations by predictive validity, the correlation between in-sample and out-of-sample rank, rather than in-sample mean, and report a twelve-tier measurement apparatus that exposes the deployment-relevant dimensions HELM and its agent-era successors collapse. The position is operationalized through three falsifiable out-of-distribution criteria with explicit thresholds; existing evidence partly supports it but is too thin to confirm. We close with a pre-registered pilot design and a field-level vision for what the next generation of agentic benchmarks should report.

05.
arXiv (CS.AI) 2026-06-16

HoloRec: Holistic Encoding and Interleaved Reasoning for Generative Recommendation

arXiv:2606.15331v1 Announce Type: cross Abstract: Generative recommendation models that formulate the task as sequence generation overcome the objective fragmentation problem of traditional cascade architectures, yet existing approaches still suffer from flat semantic representations lacking hierarchical structure for multi-step reasoning and an externally constructed chain-of-thought (CoT) that requires expensive annotations and remains disconnected from the generation objective. We propose HoloRec, an endogenous chain-of-thought recommendation mechanism that unifies representation, reasoning, and generation by constructing a hierarchical semantic encoding matrix via multi-granularity nested residual quantization optimized by a holistic reconstruction loss. HoloRec supports two inference modes: a non-thinking mode that uses lightweight multi-granularity supervised alignment for fast prediction, and a thinking mode that employs an interleaved reasoning scheme to generate CoT steps on the fly, directly embedding reasoning into the generation process without external data. Experiments on multiple public recommendation datasets demonstrate that HoloRec consistently outperforms baselines, with especially significant gains in sparse scenarios, and the thinking mode achieves better accuracy than the non-thinking mode with only modest inference overhead.

06.
arXiv (CS.AI) 2026-06-16

Topological Flow Matching

arXiv:2606.15897v1 Announce Type: cross Abstract: Flow matching is a powerful generative modeling framework, valued for its simplicity and strong empirical performance. However, its standard formulation treats signals on structured spaces, such as fMRI data on brain graphs, as points in Euclidean space, overlooking the rich topological features of their domains. To address this, we introduce topological flow matching, a topology-aware generalization of flow matching. We interpret flow matching as a framework for solving a degenerate Schrödinger bridge problem and inject topological information by augmenting the reference process with a Laplacian-derived drift. This principled modification captures the structure of the underlying domain while preserving the desirable properties of flow matching: a stable, simulation-free objective and deterministic sample paths. As a result, our framework serves as a drop-in replacement for standard flow matching. We demonstrate its effectiveness on diverse structured datasets, including brain fMRIs, ocean currents, seismic events, and traffic flows.

07.
arXiv (CS.AI) 2026-06-16

Fusion is not one-size-fits-all: Cross-Modal Representation Alignment for Time-to-Event Modeling

arXiv:2606.15038v1 Announce Type: new Abstract: Accurate time-to-event (TTE) prediction from multimodal clinical data remains challenging due to modality imbalance and distribution shift. We introduce a foundation model-driven framework for cross-modal representation alignment between CT imaging and longitudinal EHR data, designed to generalize across tasks and institutions. CT and EHR modalities are encoded independently using domain-specific foundation models and aligned in a shared latent space through four principled fusion strategies: late fusion, contrastive alignment, cross-attention, and co-attention. We evaluate two clinically distinct TTE tasks: pulmonary embolism (PE) mortality and cardiovascular disease (CVD) outcomes, on large-scale multi-institutional cohorts (PE: N=3,099 train; 1,098 internal; 435 external; CVD: N=2,951 train; 837 internal; 682 external). Fusion consistently improves concordance index by 1.5-5.4% over unimodal baselines when modalities contribute comparably. Overall, contrastive multimodal fusion, particularly with CLMBR representations, provided the most consistent and statistically robust improvements, especially for PE mortality prediction. For MACE, cross-attention (one-hot) achieved the highest internal performance and image-guided co-attention achieved the best external performance. We therefore introduce a generalizable foundation model-based cross-modal alignment framework and provide the first systematic analysis of fusion behavior under modality imbalance in TTE prediction. Our results establish task-aware multimodal alignment as a necessary design principle for robust generalization and scalable clinical deployment.

08.
arXiv (CS.AI) 2026-06-18

Towards Understanding What State Space Models Learn About Code

arXiv:2602.06774v2 Announce Type: replace Abstract: State Space Models (SSMs) have emerged as an efficient alternative to the Transformer architecture. Prior work shows that, when trained under comparable conditions, SSMs can match or surpass Transformers on code understanding tasks. However, their internal mechanisms remain a black box. We present the first systematic analysis of what SSM-based code models learn along with the direct comparison between SSM and Transformer models in this domain. Our analysis shows that SSMs capture syntactic and semantic structure more effectively than Transformers during pretraining but forgets certain relations during fine-tuning on some tasks. To investigate this behavior, we introduce SSM-Interpret, a frequency-domain framework that exposes a spectral shift toward short-range dependencies during fine-tuning. Guided by these findings, we propose architectural modifications that significantly improve the performance of SSM-based code model by upto +6 MRR on NLCodeSearch. This demonstrates that our analysis not only explains model behavior but also leads directly to better designs.

09.
arXiv (math.PR) 2026-06-17

Limit theorems for descents and inversions of shelf-shuffles

arXiv:2510.00343v2 Announce Type: replace Abstract: We prove central limit theorems for the number of descents and inversions of permutations produced by shelf-shuffles. These are a model for casino card shuffling machines. We show the asymptotic normality of the number of descents in two limiting regimes depending on the ratio of cards to shelves. On the other hand, we study the inversions by employing a modification of the techniques from Islak's analysis of the statistics of riffle shuffles. In particular, we obtain a bound for the rate of convergence for inversions that is independent of the number of shelves.

10.
arXiv (CS.LG) 2026-06-12

Mirror Descent on Riemannian Manifolds

arXiv:2603.17527v2 Announce Type: replace-cross Abstract: Mirror Descent (MD) is a scalable first-order method widely used in large-scale optimization, with applications in image processing, policy optimization, and neural network training. This paper generalizes MD to optimization on Riemannian manifolds. In particular, we develop a Riemannian Mirror Descent (RMD) framework via reparameterization and further propose a stochastic variant of RMD. We also establish non-asymptotic convergence guarantees for both RMD and stochastic RMD. As an application to the Stiefel manifold, our RMD framework reduces to the Curvilinear Gradient Descent (CGD) method proposed in [26]. Moreover, when specializing the stochastic RMD framework to the Stiefel setting, we obtain a stochastic extension of CGD, which effectively addresses large-scale manifold optimization problems.

11.
arXiv (CS.LG) 2026-06-16

A Comparative Study of Graph Neural Network Layer Selection for Interaction Modelling in Driving Trajectory Prediction

arXiv:2606.14956v1 Announce Type: new Abstract: Autonomous driving systems rely on precise trajectory prediction to plan safe and efficient movement. Graph Neural Networks (GNNs) have become a promising approach for modelling spatiotemporal interactions among road agents. However, designing GNN architectures for trajectory prediction remains non-standardized, with little guidance on which graph layers effectively capture spatial interactions and temporal dynamics. This paper offers a detailed comparative study of 19 graph layer types, focusing on their spatial and temporal processing capabilities to discover the most effective architectures for trajectory prediction. Within the explored hyperparameter setting, we highlight five standout layer combinations, with ARMA, Chebyshev, and topology-aware layers consistently performing better than others. Beyond performance metrics, our findings yield practical design principles: sum-based aggregation is more effective than mean-based methods, multi-head attention mechanisms enable richer interactions, and assigning different weights to different hop distances significantly improves prediction accuracy. These findings offer useful guidance for designing more interpretable and effective trajectory prediction models.

12.
arXiv (CS.LG) 2026-06-18

Model-Free Reinforcement Learning Control for Resilient Cyber-Physical Systems

arXiv:2606.19069v1 Announce Type: cross Abstract: This paper compares the performance of model-free controllers on a nonlinear system under cyberattacks, including false data injection and denial-of-service attacks. Four RL reward types are analyzed for accuracy, cost, and resilience. Results show that the Lyapunov reward offers the best resilience with low tracking error. Exponential mode also provides good trade-offs with acceptable resilience under moderate training conditions. Progressive and linear rewards converge faster but are less robust. RL-MPCs show strong steady-state resilience but require longer training times; RL-PID controllers are faster with significantly less training time. Proximal Policy Optimization outperforms Deep Deterministic Policy Gradient with a significant reduction in KPI variance. This study serves to highlight how well-designed RL rewards can improve performance and resilience against cyber threats.

13.
arXiv (CS.LG) 2026-06-16

Mean-Field Parallel Decoding for Discrete Diffusion Language Models

arXiv:2606.15805v1 Announce Type: new Abstract: Discrete diffusion language models enable parallel token generation, offering a pathway to low-latency decoding. However, selecting tokens independently by marginal confidence limits effective parallelism: tokens that appear reliable in isolation can form incompatible configurations when several positions are updated at once. We introduce a training-free decoding framework that coordinates these parallel updates. At each forward pass, the method assigns a commit score to each masked position and refines these scores using pairwise interactions derived from the model's predictive distributions. A variational relaxation yields a simple fixed-point update that suppresses conflicting simultaneous commitments within a single forward pass. This mechanism allows the decoder to commit more tokens in parallel while maintaining competitive generation quality. The method is lightweight, requires no auxiliary model or retraining, and drops into existing diffusion decoding pipelines without modification. Experiments on reasoning and code-generation benchmarks show consistent improvements in the quality-latency trade-off.

14.
medRxiv (Medicine) 2026-06-22

Association of Digoxin Use at Norwood Discharge with Fontan Completion: A Study from the Pediatric Heart Network Public Dataset

Background: Digoxin use after the Norwood procedure has been associated with improved interstage survival in hypoplastic left heart syndrome and related conditions. Whether this benefit translates into improved longer-term outcomes through staged palliation remains unknown. We aimed to determine the association of digoxin use at Norwood discharge with transplant-free survival and Fontan completion. Methods: We conducted a retrospective cohort study using the Pediatric Heart Network (PHN) Single Ventricle Reconstruction trial public dataset, including 549 infants enrolled at 15 North American centers between 2005 and 2008. Competing risk analysis was used to evaluate Fontan completion and Cox regression to assess death or transplantation within 6 years after the Norwood procedure. Mixed-effects models compared pre-Fontan hemodynamic and echocardiographic right ventricular indices between patients treated with and without digoxin after accounting for center clustering and adjustment for sex, shunt type, heart failure medications at Norwood discharge, and census block poverty level. Results: The 6-year cumulative incidence of Fontan completion was higher among patients discharged on digoxin than among those not receiving digoxin (82% vs 71%; p = 0.013). Competing-risk analysis accounting for death and transplant demonstrated a greater likelihood of Fontan completion among digoxin users (aHR 1.31; 95%CI 1.09-1.58; p = 0.005), without significant difference in the hazard of death or transplant (aHR 0.78; 95%CI 0.53-1.15; p = 0.208). No significant differences in pre-Fontan hemodynamic or echocardiographic indices were observed between groups. Initiation of digoxin post Stage II procedure was not associated with improved survival or likelihood to complete Fontan. Conclusion: Digoxin use at the time of Norwood discharge was associated with a 30% greater likelihood of Fontan completion by 6 years, without accompanying improvement in transplant-free survival. These findings extend prior observations of improved interstage outcomes associated with digoxin use and suggest that treatment may facilitate progression through staged palliation.

15.
arXiv (CS.AI) 2026-06-16

Post-Hoc Merging is Not Enough: Many-Shot Model Merging with Loss-Gap Balancing

arXiv:2606.16501v1 Announce Type: new Abstract: Model merging has become a practical post-training strategy for building a single multi-task large language model (LLM) by combining multiple task-specialized models. However, most existing approaches rely on post-hoc merging, in which task-specific models are merged only once after training. This one-shot aggregation often suffers from task interference, leading to information erasure across individual tasks. In this work, we show that replacing post-hoc merging with an iterative many-shot merging protocol is effective in improving multi-task performance. Building on this insight, we propose METIS, Mitigating Erasure from Task Interference for Stable many-shot merging. METIS is a loss-aware many-shot merging method that addresses information erasure in post-hoc merging through task-wise loss-gap weighting and consensus-based masking. Notably, METIS exhibits significant performance improvement on the worst-performing task, effectively mitigating information erasure. (Project page: https://imkyungjin.github.io/METIS/)

16.
medRxiv (Medicine) 2026-06-22

Reliable quantification of renal function from frozen blood samples

BACKGROUND: Differences in renal function may affect Alzheimer disease (AD) blood biomarker levels independent of AD pathology. Although renal function was unaccounted for in foundational AD blood biomarker studies, there is potential to address this through quantification of estimated glomerular filtration rate (eGFR) from frozen serum and plasma samples. However, the validity of eGFR evaluation from long-term frozen blood samples is unknown. METHODS: Adults aged 50-85 with at least 2 vascular risk factors were recruited from vascular surgery or cardiology clinics in Tucson, Arizona from 2022-2025. Individuals with creatinine assessments in point-of-care whole blood (POC-WB) and frozen serum and plasma samples using the iSTAT (Abbott) were included. eGFR was calculated using the 2021 CKD-EPI creatinine equation without race. Agreement between POC-WB and frozen blood samples was assessed using Cohen's kappa with linear weights. RESULTS: 134 participants (mean [SD] age: 72.6 [7.5] years, 39.6% female, 23.1% chronic kidney disease) had POC-WB eGFR available. Frozen serum and plasma samples had strong agreement with POC-WB for eGFR (Kw= 0.90-0.95, P

17.
arXiv (CS.CV) 2026-06-15

A Unified Theory of Sinusoidal Activation Families for Implicit Neural Representations

Implicit Neural Representations (INRs) model continuous signals with compact neural networks and have become a standard tool in vision, graphics, and signal processing. A central challenge is accurately capturing fine detail without heavy hand-crafted encodings or brittle training heuristics. Across the literature, periodic activations have emerged as a compelling remedy: from SIREN, which uses a single sinusoid with a fixed global frequency, to more recent architectures employing multiple sinusoids and, in some cases, trainable frequencies and phases. We study this family of sinusoidal activations and develop a principled theoretical and practical framework for trainable sinusoidal activations in INRs. Concretely, we instantiate this framework with Sinusoidal Trainable Activation Functions (STAF), a Fourier-like activation whose amplitudes, frequencies, and phases are learned. Our analysis (i) establishes a Kronecker-equivalence construction that expresses trainable sinusoidal activations with standard sine networks and quantifies expressive growth, (ii) characterizes how the Neural Tangent Kernel (NTK) spectrum changes under trainable sinusoidal parameterization, and (iii) provides an initialization that yields standard normal post-activations without asymptotic central limit theorem (CLT) arguments. Empirically, on images, audio, shapes, inverse problems (super-resolution, denoising) and NeRF, STAF is competitive and often stronger on distortion-oriented reconstruction metrics such as PSNR/SSIM across the evaluated INR tasks, with favorable parameter efficiency under layer-wise sharing. While periodic activations can alleviate practical manifestations of spectral bias, our results indicate they do not eliminate it; instead, trainable sinusoids can improve the observed capacity-optimization trade-off in the evaluated settings.

18.
arXiv (CS.CL) 2026-06-16

Anything Goes? A Crosslinguistic Study of (Im)possible Language Learning in LMs

Do language models (LMs) offer insights into human language learning? A common argument against this idea is that because their architecture and training paradigm are so vastly different from humans, LMs can learn arbitrary inputs as easily as natural languages. We test this claim by training LMs to model impossible and typologically unattested languages. Unlike previous work, which has focused exclusively on English, we conduct experiments on 12 languages from 4 language families with two newly constructed parallel corpora. Our results show that while GPT-2 small can largely distinguish attested languages from their impossible counterparts, it does not achieve perfect separation between all the attested languages and all the impossible ones. We further test whether GPT-2 small distinguishes typologically attested from unattested languages with different NP orders by manipulating word order based on Greenberg's Universal 20. We find that the model's perplexity scores do not distinguish attested vs. unattested word orders, while its performance on the generalization test does. These findings suggest that LMs exhibit some human-like inductive biases, though these biases are weaker than those found in human learners.

19.
arXiv (CS.AI) 2026-06-12

What Type of Inference is Active Inference?

arXiv:2606.04935v2 Announce Type: replace Abstract: Active inference casts decision-making as inference, with the Expected Free Energy (EFE) unifying goal-directed and information-seeking behavior. Recent work showed that EFE minimization can be written as Variational Free Energy (VFE) minimization on a generative model augmented with epistemic priors. We prove that the VFE of the augmented model can be rewritten as the VFE of the predictive model plus explicit entropy-correction terms, making the EFE contribution transparent. We then show that proper EFE-based planning requires combining these epistemic corrections with a planning correction that turns marginal inference into policy optimization, yielding a full variational characterization of EFE-based planning. This clarifies which corrections are needed for cross-entropy planning and for full EFE-based planning. The same entropy-corrected formulation leads to a detailed message-passing scheme for EFE-based planning together with simpler ablations. Experiments on three grid-world environments show that full EFE-based planning outperforms ablations that omit either the planning correction or the epistemic corrections.

20.
arXiv (CS.CL) 2026-06-17

LVLMs and Humans Ground Differently in Referential Communication

For generative AI agents to partner effectively with human users, the ability to accurately predict human intent is critical. But this ability to collaborate remains limited by a critical deficit: an inability to model common ground. We present a referential communication experiment with a factorial design involving director-matcher pairs (human-human, human-AI, AI-human, and AI-AI) that interact with multiple turns in repeated rounds to match pictures of objects not associated with any obvious lexicalized labels. We show that LVLMs cannot interactively generate and resolve referring expressions in a way that enables smooth communication, a crucial skill that underlies human language use. We release our corpus of 356 dialogues (89 pairs over 4 rounds each) along with the online pipeline for data collection and the tools for analyzing accuracy, efficiency, and lexical overlap.

21.
arXiv (CS.CV) 2026-06-17

Edit3DGS: Unified Framework for Dynamic Head Editing via 2D Instruction-Guided Diffusion and 3D Gaussian Splatting

We present Edit3DGS, a unified framework for dynamic 3D head editing that integrates 2D instruction-guided diffusion with 3D Gaussian splatting. Unlike prior approaches that separately address frame-based edits or static 3D reconstruction, our method couples semantic controllability in the image domain with photorealistic, temporally consistent 3D representations. Given an input video, editable facial regions are masked and modified using a text-conditioned diffusion model to support fine-grained operations such as expression transformation, attribute modification, and appearance refinement. The edited frames are then aggregated through 3D Gaussian splatting to produce a coherent, high-fidelity avatar that preserves both identity and motion dynamics. To enforce consistency, Edit3DGS incorporates multi-view batch editing and lightweight inpainting strategies that recover lost expressions across timesteps. Experimental results demonstrate that our framework enables controllable, artifact-free head editing with smooth temporal transitions, offering practical applications in virtual avatars, immersive communication, film production, and interactive media.

22.
arXiv (quant-ph) 2026-06-12

Electric Field Distortions in Surface Ion Traps with Integrated Nanophotonics

arXiv:2503.20387v3 Announce Type: replace Abstract: The integration of photonic components into surface ion traps provides a scalable approach for trapped-ion quantum computing, sensing, and metrology, enabling compact systems with enhanced stability and precision. However, the introduction of optical apertures in the trap electrodes can distort the trapping electric field. This can lead to excess micromotion (EMM) and ion displacement which degrade the performance of quantum logic operations and optical clocks. In this work, we systematically investigate the electric field distortion in a surface ion trap with integrated waveguides and grating couplers using Finite Element Method (FEM) simulations. We analyze methods to reduce these distortions by exploiting symmetries and transparent conductive oxide materials.

23.
arXiv (CS.AI) 2026-06-11

Generalizing Beyond Suboptimality: Offline Reinforcement Learning Learns Effective Scheduling through Random Solutions

arXiv:2509.10303v2 Announce Type: replace-cross Abstract: Online reinforcement learning (RL) approaches have demonstrated strong performance on Job Shop Scheduling (JSP) and Flexible JSP (FJSP) problems by learning scheduling policies through direct interaction with simulated environments. However, these methods often require extensive training interactions, limiting their sample efficiency and practical applicability. Motivated by this challenge, we introduce Conservative Discrete Quantile Actor-Critic (CDQAC), an offline RL algorithm that learns effective scheduling policies directly from static, suboptimal datasets. CDQAC couples a quantile-based critic with delayed policy updates to estimate the return distribution of machine-operation pairs. Extensive experiments on JSP and FJSP benchmarks demonstrate that CDQAC consistently outperforms the data-generating heuristics, surpasses state-of-the-art offline and online RL baselines, and is highly sample efficient, requiring only 1 to 5% of the original dataset to learn high-quality policies. Our analysis suggests that, in scheduling, offline RL performance is governed mainly by state-action coverage rather than the quality of individual trajectories. Scheduling couples a dense reward aligned with the makespan objective with equal-length trajectories across heuristics, enabling effective learning from a broad range of behaviors. Consistent with this observation, datasets generated by a simple random heuristic with broader coverage let it outperform policies trained on datasets produced by stronger heuristics such as Genetic Algorithms.

24.
arXiv (CS.CL) 2026-06-12

Keep Policy Gradient in Charge: Sibling-Guided Credit Distillation for Long-Horizon Tool-Use Agents

Long-horizon tool-use reinforcement learning can learn from outcome verification, but its trajectory-level advantage is broadcast across many reasoning, API, and answer tokens. Self-distillation promises a denser signal by reusing a policy's own rollouts or a privileged teacher. We show, however, that direct token-level self-distillation can silently destroy tool use: it rehearses teacher behavior without knowing which actions the verifier rewards, so useful skills and harmful shortcuts are amplified together. We introduce Sibling-Guided Credit Distillation (SGCD), which uses distillation for credit assignment rather than as a competing actor loss. Dynamic sampling produces mixed successful and failed sibling rollouts; an external LLM summarizes their contrast into a training-only stepwise credit reference; dense teacher/student divergence drives credit reassignment; and bounded detached credit weights reshape GRPO token advantages. The deployed student sees no external LLM, sibling evidence, or oracle. Across AppWorld and $\tau^3$-airline, SGCD improves over matched GRPO comparators: AppWorld TGC $42.9 \to 45.6$ on test_normal and $24.7 \to 27.0$ on test_challenge, and $\tau^3$-airline pass@1 $0.583 \to 0.602$.

25.
arXiv (CS.CL) 2026-06-19

MENTOR: Reinforcement Learning via Flexible Teacher-Optimized Rewards for Tool-Use Distillation

Distilling the tool-use capabilities of large language models (LLMs) into small language models (SLMs) is essential for their practical application. The predominant approach, supervised fine-tuning (SFT), suffers from poor out-of-domain (OOD) generalization due to its rigid alignment with static teacher trajectories. While reinforcement learning (RL) offers an alternative, the capacity limitations of SLMs pose a severe dilemma: sparse outcome rewards provide insufficient guidance, whereas strict trajectory matching imposes overly restrictive constraints. To bridge this capacity-driven gap, we propose MENTOR, which introduces a flexible yet process-aware reward structure. Instead of enforcing rigid replication, MENTOR uses the teacher's reference to guide tool-use behavior, balancing behavioral alignment with downstream performance. Extensive experiments on controlled executable-tool benchmarks demonstrate that MENTOR improves OOD tool-use performance compared to SFT and strict RL baselines. Our findings suggest that within verifiable tool-use environments, flexible tool-use alignment offers a more effective approach than strict trajectory replication for developing adaptable small models.