arXiv:2607.17901v1 Announce Type: cross
Abstract: In an incomplete round robin tournament, teams play against different sets of opponents. Given varying strengths of the teams, this raises a fairness issue. We establish the complexity of determining whether a fair draw exists under various scenarios involving the presence of pots (i.e., strength based groups of teams), teams coming from the same association, connectivity requirements, and others. We also show experimentally how fair incomplete round robin tournaments can be generated.
Science Journals
arXiv:2412.11475v3 Announce Type: replace
Abstract: We present OmniVLM, a sub-billion-parameter vision-language model for efficient on-device inference. OmniVLM introduces a token compression mechanism that reduces visual token sequence length from 729 to 81 tokens, significantly reducing computational overhead while preserving visual-semantic fidelity. Through a multi-stage training pipeline of pretraining, supervised fine-tuning, and minimal-edit Direct Preference Optimization (DPO), OmniVLM matches the performance of larger models. On multiple benchmarks including ScienceQA, POPE, and MMMU, OmniVLM outperforms existing baselines like nanoLLAVA within a 968M-parameter footprint. Empirical results on the same laptop demonstrate 9.1x faster time-to-first-token (0.75s vs 6.82s) and 1.5x higher decoding speed (29.41 vs 19.20 tokens/s) compared to nanoLLAVA, enabling efficient deployment on edge devices. The model weights can be accessed on huggingface: https://huggingface.co/NexaAIDev/OmniVLM-968M, and the inference examples can be find in Appendix B.
arXiv:2502.04411v3 Announce Type: replace
Abstract: Model merging aggregates Large Language Models (LLMs) finetuned on different tasks into a stronger one. However, parameter conflicts between models leads to performance degradation in averaging. While model routing addresses this issue by selecting individual models during inference, it imposes excessive storage and compute costs, and fails to leverage the common knowledge from different models. In this work, we observe that different layers exhibit varying levels of parameter conflicts. Building on this insight, we average layers with minimal parameter conflicts and use a novel task-level expert routing for layers with significant conflicts. To further reduce storage costs, inspired by task arithmetic sparsity, we decouple multiple fine-tuned experts into a dense expert and several sparse experts. Considering the out-of-distribution samples, we select and merge appropriate experts based on the task uncertainty of the input data. We conduct extensive experiments on both LLaMA and Qwen with varying parameter scales, and evaluate on real-world reasoning tasks. Results demonstrate that our method consistently achieves significant performance improvements while requiring less system cost.
arXiv:2509.20174v2 Announce Type: replace
Abstract: Higher-order interactions, where groups of nodes interact collectively rather than pairwisely, are central to many complex systems, from neural and ecological networks to social contagion. However, simulating dynamical processes on such higher-order structures remains computationally challenging due to the combinatorial growth of possible interactions. Here, we develop efficient and statistically exact Gillespie algorithms for Markovian spreading dynamics on large and heterogeneous hypergraphs. By incorporating phantom processes $-$events that advance time without altering the system's state$-$, we drastically reduce the computational complexity of standard algorithms ($\mathcal{O}(N^2)$), achieving up to linear scaling with system size. Relying on the susceptible-infected-susceptible model with critical mass thresholds as a benchmark, we show that the optimized algorithms outperform standard approaches by several orders of magnitude, enabling simulations of networks with millions of nodes and broad heterogeneity in both degree and interaction order. Efficient sampling methods, needed to overcome the bottlenecks imposed by either a high maximum order or number of interactions, and other dynamical processes on higher-order networks are tackled. These results establish a general framework for scalable, continuous-time simulations of higher-order contagion and related dynamical processes.
arXiv:2509.20593v3 Announce Type: replace
Abstract: This paper presents an uncertainty-aware framework for the active tracking of marine pollution sources using Unmanned Surface Vehicles (USVs). The proposed framework employs an Informative Path Planning (IPP) strategy driven by Bayesian inference, modelling the belief of source location as a categorical distribution. This work presents a high-fidelity simulation pipeline, coupling Computational Fluid Dynamics (CFD) for realistic pollutant dispersion with Gazebo-based hydrodynamics and ArduPilot for USV control. Furthermore, this paper introduces the Smallest Credible Interval (SCI) as a metric to quantify estimation uncertainty and to serve as an autonomous termination criterion. Extensive simulations across diverse wave conditions and source locations demonstrate that the proposed framework achieves a 95.8% success rate, significantly outperforming baseline methods in both localisation accuracy and environmental adaptability. This framework provides a scalable and ROS-compatible foundation for fully autonomous environmental monitoring and rapid incident response.
arXiv:2509.22337v2 Announce Type: replace
Abstract: Belief Propagation (BP) is a widely used approximate inference algorithm in probabilistic graphical models (PGMs), but is computationally expensive when applied to large-scale program analysis. Existing GPU-based approaches are unable to support flexible update strategies and have yet to integrate logical constraints with GPU acceleration, leading to challenges in both generality and efficiency.
We present FastLBP, a GPU-accelerated BP framework for program analysis. We propose a unified representation for specifying flexible update strategies required in program analysis, along with a dependency analysis algorithm to enable parallel execution. Furthermore, we implement BP with local structures on GPUs by assigning individual threads to message computations and utilizing a memory-efficient representation. Experiments on SmartFL and BINGO show that FastLBP achieves average speedups of $17.42\times$ and $2.82\times$ over CPU-based approaches on SmartFL and BINGO, respectively, and $6.14\times$ over GPU-based approach on SmartFL, while preserving accuracy. Moreover, FastLBP supports update strategies that existing GPU-based approaches cannot support, demonstrating its improved generality for real-world program analysis.
arXiv:2509.23071v2 Announce Type: replace
Abstract: Retrieval-augmented generation (RAG) agent development is hindered by the lack of executable ground-truth agent-environment interaction trajectories. Existing datasets provide questions, answers, and evidence, but lack fine-grained supervision for retriever invocation, dynamic planning, and stepwise decision-making. Reinforcement learning offers a potential solution, but often suffers from sparse rewards and cold-start failures when base large language models (LLMs) lack sufficient reasoning capability. Meanwhile, existing data synthesis methods mainly generate post-hoc rationales rather than executable environment-interaction trajectories. In this paper, we propose EviPath, an evidence-anchored reasoning path synthesis paradigm for RAG agent development. EviPath reverse-engineers executable trajectories from question-answer pairs and supporting evidence through three stages: (i) Abductive Subtask Planning, which decomposes questions and plans dependency-aware solution paths; (ii) Faithful Sub-question Answering, which uses supporting evidence as a proxy environment to generate grounded intermediate thoughts and answers; and (iii) Conversational Fine-Tuning, which converts complete trajectories into a dialogue format for supervised fine-tuning. Experiments on widely used question-answering benchmarks show that an 8B model trained on our synthetic corpus significantly and consistently outperforms state-of-the-art baselines, achieving a 14.7% absolute Exact Match gain in open-domain question answering.
arXiv:2607.17946v1 Announce Type: new
Abstract: Large Language Models (LLMs) often struggle to navigate value conflicts when trained with the compressed scalar rewards of Reinforcement Learning from Human Feedback (RLHF). To address this challenge, we investigate how chain-of-thought (CoT) reasoning can help improve performance in this domain. Geometrically, we show that CoT correlates with further smoothing the model's loss landscape in its sharpest direction, helping resolve the optimization instability of traditional scalar rewards. We also demonstrate via relevant downstream benchmarks that value conflict-focused CoT may generalize to different kinds of moral reasoning, demonstrating that this CoT has the potential to be an effective mechanism for better moral reasoning. To capitalize on this potential, we create a new value conflict-focused CoT design that further smooths the sharpest direction of the loss landscape and increases moral reasoning performance. This finding shows that explicitly modifying and improving the design of reasoning dynamics offers a promising avenue for improving model performance on user requests with complex value conflicts, advancing pluralistic alignment in LLMs.
arXiv:2509.23185v2 Announce Type: replace
Abstract: We present an integrated planning framework for quadrupedal locomotion over dynamically changing, unforeseen terrains. Existing methods often depend on heuristics for real-time foothold selection-limiting robustness and adaptability-or rely on computationally intensive trajectory optimization across complex terrains and long horizons. In contrast, our approach combines reactive synthesis for generating correct-by-construction symbolic-level controllers with mixed-integer convex programming (MICP) for dynamic and physically feasible footstep planning during each symbolic transition. To reduce the reliance on costly MICP solves and accommodate specifications that may be violated due to physical infeasibility, we adopt a symbolic repair mechanism that selectively generates only the required symbolic transitions. During execution, real-time MICP replanning based on actual terrain data, combined with runtime symbolic repair and delay-aware coordination, enables seamless bridging between offline synthesis and online operation. Through extensive simulation and hardware experiments, we validate the framework's ability to identify missing locomotion skills and respond effectively in safety-critical environments, including scattered stepping stones and rebar scenarios.
arXiv:2509.24511v2 Announce Type: replace
Abstract: Retractions have increased sharply in recent years, alongside a growing number of papers that receive post-publication comments questioning their reliability (commented papers). Together, retracted and commented papers undermine the credibility of scientific research and may also threaten public health. In this study, we examine problematic papers in the field of non-coding RNA (ncRNA) from multiple perspectives to identify common patterns and inform strategies for addressing large-scale fraudulent publications. We find that studies on under-investigated ncRNAs are more likely to become problematic papers. These papers often show substantial textual similarity, and many additional papers with similar text also display suspicious image duplication. Healthcare institutions, particularly those with lower publication output, appear especially vulnerable to producing such papers. Most problematic papers are concentrated in a small set of journals, many of which do not adequately address concerns raised after publication. Overall, our findings indicate that a substantial number of problematic papers may remain undetected and that their shared characteristics can support more effective strategies for identifying and curbing large-scale fraudulent publications.
arXiv:2607.17458v1 Announce Type: new
Abstract: In conventional optical microscopy, accessing subwavelength information is classically constrained by the Abbe diffraction limit which remains a significant challenge in nanophotonics. To overcome this limitation, we introduce a scheme that generates a tightly confined near-field focal spot under non-diffracting Bessel beam illumination, employing a series of high refractive-index nanoscale structures. Furthermore, the utilization of higher-order Bessel beams enables additional resolution enhancement beyond the diffraction limit. Our results demonstrate a subwavelength resolution of (lambda/53) with a pyramidal nanostructure. To elucidate the factors governing the focusing behavior, we systematically investigate the influence of different geometrical parameters related to the incident Bessel beam on the focusing performance and shows that these factors critically determine the stability and spatial profile of the focal spot. These findings highlight the potential of the proposed approach for various applications in super-resolution imaging and quantum technology.
arXiv:2607.16890v1 Announce Type: new
Abstract: Spread codes are a well-known family of constant-dimension subspace-metric codes. For constant dimension $k$ and ambient space dimension $n$ being a multiple of $k$, these codes have minimum distance $2k$ and a rich geometric structure. In this paper, we study the decoding capabilities of the Nearest Neighbor Decoder for Desarguesian spread codes, establishing that unique decoding is still achievable beyond half the minimum distance. Motivated by this, we develop a new decoding algorithm to uniquely decode Desarguesian spread codes in the presence of both insertions and deletions, which increase and decrease, respectively, the dimension of the transmitted codeword. Even when the sum of the dimensions of insertions and deletions exceeds half the minimum distance, provided that deletions are of dimension at most $k-2$, the algorithm succeeds with a small decoding failure. We also propose two refinements to this algorithm that, empirically, can handle nearly as many insertions as the Nearest Neighbor Decoder.
arXiv:2607.17059v1 Announce Type: new
Abstract: Laser-plasma accelerators have been the subject of extensive research in recent years. The electron beams they generate exhibit a broad energy spread. To conveniently characterize beams from laser wakefield acceleration (LWFA), electron spectrometers employing scintillating screens coupled with CCD cameras are typically used. In this work, we calibrate a series of DRZ phosphor screens and measure the spectra of the light they emit. The calibration was performed using the radio-frequency linear electron accelerator at Tsinghua University, which provided monoenergetic electron beams with peak energy of approximately 30 MeV.
arXiv:2607.16410v1 Announce Type: cross
Abstract: The EWF-(FCI,SQD) method, a wave-function-based embedding approach combining full configuration interaction (FCI) and sample-based quantum diagonalization (SQD), is a promising new tool for the simulation of molecular systems. However, applications of EWF-(FCI,SQD) have so far been limited to single-point calculations, whereas the study of complex chemical processes requires the ability to explore potential energy surfaces. In this work, we demonstrate geometry optimization with EWF-(FCI,SQD), scaling our simulations to molecules as large as menthone and benzidine within the STO-3G basis set. Without fragmentation, these systems comprise 73 and 82 molecular orbitals respectively, presenting an intractable Hilbert space for conventional exact or high-level subspace solvers and establishing a clear necessity for fragmentation-based methodologies. The underlying fragment SQD simulations in the EWF-(FCI,SQD) geometry optimizations use up to 70 qubits. The resulting geometries show exceptional accuracy relative to the classical reference, with deviations below 4 picometers.
arXiv:2510.01483v3 Announce Type: replace
Abstract: Vision-language models (VLMs) demonstrate strong image-level scene understanding, but reasoning over long egocentric video remains costly: because VLMs maintain no persistent memory or explicit spatial representation, all sampled frames must be re-processed for every new query. We present VL-KnG, a training-free framework that constructs spatiotemporal knowledge graphs from monocular egocentric video, bridging fine-grained scene graphs and global topological graphs without 3D reconstruction. VL-KnG processes video in chunks, maintains persistent object identities via large language model (LLM)-based Spatiotemporal Object Association (STOA), and answers questions through Graph-Enhanced Retrieval (GER), which combines subgraph retrieval with visual grounding. Once constructed, the knowledge graph removes the need to revisit raw video at query time, decoupling query latency from video length. Across three embodied question answering benchmarks -- OpenEQA, NaVQA, and our newly introduced WalkieKnowledge -- VL-KnG achieves accuracy competitive with frontier VLMs while answering queries at substantially lower latency and with explainable, graph-grounded reasoning; it further surpasses prior persistent-representation baselines and open-weight VLMs in several settings. Deployment on a real robot demonstrates practical applicability, with query latency remaining stable as observation history grows. As a persistent, queryable scene memory, VL-KnG provides a concrete substrate for memory maintenance and knowledge updating in embodied AI agents.
arXiv:2510.05969v3 Announce Type: replace
Abstract: Large language models (LLMs) are increasingly deployed on complex reasoning tasks, yet little is known about their ability to internally evaluate problem difficulty, which is an essential capability for adaptive reasoning and efficient resource allocation. In this work, we investigate whether LLMs implicitly encode problem difficulty in their internal representations. Using a linear probe on the final-token representations of LLMs, we demonstrate that the difficulty level of math problems can be linearly modeled. We further locate the specific attention heads of the final Transformer layer: these attention heads have opposite activation patterns for simple and difficult problems, thus achieving perception of difficulty. Our ablation experiments prove the accuracy of the location. Crucially, our experiments provide practical support for using LLMs as automatic difficulty annotators, potentially substantially reducing reliance on costly human labeling in benchmark construction and curriculum learning. We also uncover that there is a significant difference in entropy and difficulty perception at the token level. Our study reveals that difficulty perception in LLMs is not only present but also structurally organized, offering new theoretical insights and practical directions for future research. Our code is available at https://github.com/Aegis1863/Difficulty-Perception-of-LLMs.
arXiv:2510.09171v2 Announce Type: replace
Abstract: Instance-level recognition (ILR) focuses on identifying individual objects rather than broad categories, offering the highest granularity in image classification. However, this fine-grained nature makes creating large-scale annotated datasets challenging, limiting ILR's real-world applicability across domains. To overcome this, we introduce a novel approach that synthetically generates diverse object instances from multiple domains under varied conditions and backgrounds, forming a large-scale training set. Unlike prior work on automatic data synthesis, our method is the first to address ILR-specific challenges without relying on any real images. Fine-tuning foundation vision models on the generated data significantly improves retrieval performance across seven ILR benchmarks spanning multiple domains. Our approach offers a new, efficient, and effective alternative to extensive data collection and curation, introducing a new ILR paradigm where the only input is the names of the target domains, unlocking a wide range of real-world applications. The code and pretrained models are publicly available at https://github.com/yankungou/ILGen.
arXiv:2510.09344v2 Announce Type: replace
Abstract: Elderly speech poses unique challenges for automatic processing due to age-related changes such as slower articulation and vocal tremors. Existing Chinese datasets are mostly recorded in controlled environments, limiting their diversity and real-world applicability. To address this gap, we present WildElder, a Mandarin elderly speech corpus collected from online videos and enriched with fine-grained manual annotations, including transcription, speaker age, gender, and accent strength. Combining the realism of in-the-wild data with expert curation, WildElder enables robust research on automatic speech recognition and speaker profiling. Experimental results reveal both the difficulties of elderly speech recognition and the potential of WildElder as a challenging new benchmark. The dataset and code are available at https://github.com/NKU-HLT/WildElder.
arXiv:2510.13634v2 Announce Type: replace
Abstract: Quantum reservoir computing (QRC) offers a hardware-friendly approach to temporal learning, yet most studies target univariate signals and overlook near-term hardware constraints. This work introduces a gate-based QRC for multivariate time series (MTS-QRC) that pairs injection and memory qubits and uses a Trotterized nearest-neighbor transverse-field Ising evolution optimized for current device connectivity and depth. On Lorenz-63 and ENSO, the method achieves a mean square error (MSE) of 0.0087 and 0.0036, respectively, performing on par with classical reservoir computing on Lorenz and above learned RNNs on both, while NVAR and clustered ESN remain stronger on some settings. On IBM Heron R2, MTS-QRC sustains accuracy with realistic depths and, interestingly, outperforms a noiseless simulator on ENSO; singular value analysis indicates that device noise can concentrate variance in feature directions, acting as an implicit regularizer for linear readout in this regime. These findings support the practicality of gate-based QRC for MTS forecasting on NISQ hardware and motivate systematic studies on when and how hardware noise benefits QRC readouts.
arXiv:2607.17097v1 Announce Type: new
Abstract: Hand-Object Interaction (HOI) synthesis is a cornerstone for animation production and embodied AI. Despite the strong priors of video foundation models, multi-view consistent HOI synthesis remains challenging due to complex hand motions and occlusions. We present HarmoHOI, a unified diffusion framework that jointly and harmoniously generates synchronized multi-view HOI videos and globally aligned 3D point tracks. Our core insight is that robust multi-view consistency fundamentally requires globally aligned 3D geometry and motion. To this end, we propose a Mixture of Multi-view Diffusion Transformer that co-models RGB videos and 3D point tracks. By representing point tracks as pseudo-videos, we align 3D geometric signals with the 2D latent space of foundation models, thereby minimizing the domain gap and easing adaptation of priors. To further ensure geometry consistency, we introduce Global Motion Aligning Diffusion, which refines coarse point tracks into metric-scale, globally aligned 3D trajectories. HarmoHOI enables on-the-fly co-evolution of 2D appearance and 3D motion during denoising. To overcome the scarcity of multi-view HOI data, we employ a hybrid data curriculum learning strategy that successfully transfers generic priors from single-view data to synchronized multi-view generation. Experimental results show that HarmoHOI achieves state-of-the-art performance in visual quality, motion plausibility, and multi-view geometric consistency. Project page available at https://droliven.github.io/HarmoHOI_project.
arXiv:2607.17029v1 Announce Type: new
Abstract: We study degeneracy guided list compression for greedy graph coloring when graph structure is available before colors are sampled. Our exposure calibrated ordering framework assigns each vertex an independent uniform list according to its backward neighborhood in a color independent order. Its certified instantiation, Profiled Structure Aware Asymmetric Palette Sparsification, or P-SAPST, reverses a minimum degree removal sequence and obtains every backward exposure from the removal profile. For each fixed profile, we characterize the exact local budget required by independent uniform lists under history robust greedy recovery. The profile yields linear list volume on high degree forests and on a core fringe family where reciprocal rank allocation requires Theta(n log^2 n) sampled colors. Exact conflict expectation, concentration, and a dense exposure barrier complete the theoretical description.
The evaluation contains 40,320 runs over SAPBench and two SNAP networks. At the theorem scale, P-SAPST reduces mean list size by 47.6 percent relative to calibrated APST while attaining 99.8 percent observed greedy success. P-SAPST Lite replaces peeling with a degree order and provides a lower latency order choice within the same framework. On stress graphs with 250,000 vertices and up to 1,251,868 edges, Lite obtains a payload ratio of 0.865, while calibrated APST obtains 7.886. On email Enron, the corresponding ratios are 0.193 and 5.814. Compression is strongest on hub dominated and power law graphs and disappears near the dense exposure barrier. The method complements edge oblivious streaming APST by addressing an offline regime in which structural plans can be reused.
arXiv:2607.17590v1 Announce Type: new
Abstract: Recently, Merino and M\"utze (FOCS'23+SICOMP'24) presented an algorithm for computing a Hamilton path on the skeleton of any 0/1-polytope ${\rm conv}(X)$, where $X\subseteq\{0,1\}^n$. The algorithm uses as a black box an algorithm for solving the classical linear optimization problem $\min\{w\cdot x\mid x\in X\}$ for some weight vector $w\in\mathbb{R}^n$. The resulting delay per visited vertex on the Hamilton path is only by a $\log n$ factor larger than the time to solve one instance of the optimization algorithm. In this paper, we make the Hamilton path algorithm simpler and faster. Namely, we obtain an amortized delay that is only by a constant factor larger than the running time of the optimization algorithm, thus removing the $\log n$ factor. As concrete results, this yields improved algorithms for generating bases and independent sets in a matroid, spanning trees, forests, matchings and maximum matchings in a graph, vertex covers, minimum vertex covers, independent sets and maximum independent sets in a bipartite graph, and antichains, maximum antichains and ideals in a poset. All of these listings correspond to Hamilton paths on the corresponding polytopes. Furthermore, we obtain an $\mathcal{O}(t_{\rm LP})$ amortized delay algorithm for the vertex enumeration problem on 0/1-polytopes $\{x\in\mathbb{R}^n\mid Ax\leq b\}$, where $A\in \mathbb{R}^{m\times n}$ and $b\in\mathbb{R}^m$, and $t_{\rm LP}$ is the time needed to solve the linear program $\min\{w\cdot x\mid Ax\leq b\}$. This improves upon the $\mathcal{O}(t_{\rm LP} \log n)$ delay algorithm of Merino and M\"utze, and the previous $\mathcal{O}(t_{\rm LP}\,n)$ delay algorithm of Bussieck and L\"ubbecke from 1998.
arXiv:2503.23536v3 Announce Type: replace
Abstract: Unlearnable data (ULD) has emerged as an innovative defense technique to prevent machine learning models from learning meaningful patterns from specific data, thus protecting data privacy and security. By introducing perturbations to the training data, ULD degrades model performance, making it difficult for unauthorized models to extract useful representations. Despite the growing significance of ULD, existing surveys predominantly focus on related fields, such as adversarial attacks and machine unlearning, with little attention given to ULD as an independent area of study. This survey fills that gap by offering a comprehensive review of ULD, examining unlearnable data generation methods, public benchmarks, evaluation metrics, theoretical foundations and practical applications. We compare and contrast different ULD approaches, analyzing their strengths, limitations, and trade-offs related to unlearnability, imperceptibility, efficiency and robustness. Moreover, we discuss key challenges, such as balancing perturbation imperceptibility with model degradation and the computational complexity of ULD generation. Finally, we highlight promising future research directions to advance the effectiveness and applicability of ULD, underscoring its potential to become a crucial tool in the evolving landscape of data protection in machine learning.
arXiv:2510.15936v3 Announce Type: replace
Abstract: The study explores the role of large language models (LLMs) in the context of the architectural design studio, understood as the pedagogical core of architectural education. Traditionally, the studio has functioned as an experiential learning space where students tackle design problems through reflective practice, peer critique, and faculty guidance. However, the integration of artificial intelligence (AI) in this environment has been largely focused on form generation, automation, and representation-al efficiency, neglecting its potential as a pedagogical tool to strengthen student autonomy, collaboration, and self-reflection. The objectives of this research were: (1) to identify pedagogical challenges in self-directed, peer-to-peer, and teacher-guided learning processes in architecture studies; (2) to propose AI interventions, particularly through LLM, that contribute to overcoming these challenges; and (3) to align these interventions with measurable learning outcomes using Bloom's taxonomy. The findings show that the main challenges include managing student autonomy, tensions in peer feedback, and the difficulty of balancing the transmission of technical knowledge with the stimulation of creativity in teaching. In response to this, LLMs are emerging as complementary agents capable of generating personalized feedback, organizing collaborative interactions, and offering adaptive cognitive scaffolding. Furthermore, their implementation can be linked to the cognitive levels of Bloom's taxonomy: facilitating the recall and understanding of architectural concepts, supporting application and analysis through interactive case studies, and encouraging synthesis and evaluation through hypothetical design scenarios.
arXiv:2510.19244v3 Announce Type: replace
Abstract: Deep reinforcement learning (RL) achieves remarkable performance but lacks interpretability, limiting trust in policy behavior. The existing SILVER framework (Li, Siddique, and Cao 2025) explains RL policy via Shapley-based regression but remains restricted to low-dimensional, binary-action domains. We propose SILVER with RL-guided labeling, an enhanced variant that extends SILVER to multi-action and high-dimensional environments by incorporating the RL policy's own action outputs into the boundary points identification. Our method first extracts compact feature representations from image observations, performs SHAP-based feature attribution, and then employs RL-guided labeling to generate behaviorally consistent boundary datasets. Surrogate models, such as decision trees and regression-based functions, are subsequently trained to interpret RL policy's decision structure. We evaluate the proposed framework on two Atari environments using three deep RL algorithms and conduct human-subject study to assess the clarity and trustworthiness of the derived interpretable policy. Results show that our approach maintains competitive task performance while substantially improving transparency and human understanding of agent behavior. This work advances explainable RL by transforming SILVER into a scalable and behavior-aware framework for interpreting deep RL agents in high-dimensional, multi-action settings.