Optimization
-
Personality Latents: Conditioning Frozen Language Models on Individuals from Their Own Writing
2026-10-27 08:08:06Gunhee Cho, Yun-Gyung CheongAbstract
Controlling the personality of a language model usually relies on written profiles or named trait categories, so it can only express what an author writes down or what a taxonomy names. We instead propose Personality Latents (PL), which learn a latent for each person from their own writing, with no personality labels and no authored profile, and inject it into a frozen language model as a soft prefix. Following generative latent optimization, each training person receives a free latent that is optimized through the frozen model itself, while an encoder learns to predict these latents so that a new person needs only one forward pass, with optional refinement at test time. We evaluate on held-out people by having the conditioned model answer neutral and interview questions and reading personality from its answers with scorers trained on real human writing. On the MBTI axes PL reaches an AUC of , ahead of persona embedding and prompt-based baselines under identical settings, and the ranking holds when the answers are scored in three different embedding spaces. A latent learned with Qwen2.5-3B-Instruct also transfers personality to Llama-3.1-8B-Instruct through a new adapter alone.
Abstract (translated)
URL
https://arxiv.org/abs/2610.36805
PDF
-
Batching the Reflective Optimization Loop: Parallel Proposals Make GEPA Faster and Better
2026-10-27 07:41:40Jialin Zhang, Lakshya A Agrawal, Shangyin Tan, Donghyun Lee, Lutfi Eren Erdogan, Dan Klein, Koushik Sen, Alex Dimakis, Matei ZahariaAbstract
Reflective text optimization makes agentic systems better, but the optimization loop is itself a resource-hungry workload whose sequential structure leaves parallel evaluation capacity idle. We extend the open-source GEPA optimizer to propose and evaluate a batch of candidates on each optimization step instead of one candidate at a time. In our sweep on two tasks, most batched runs finished in half the wall-clock time or less, and the fastest in about a quarter to a third. Batched settings also achieved higher held-out test scores, from 68.9% to 72.1% on LiveBench-Math (with $2\times2$) and from 49.0% to 60.0% on HoVer (with $8\times1$). A simple runtime model explains the measured speedups under a finite worker pool, and batched settings also spend their metric-call and dollar budgets more efficiently. We frame the axes along which a reflective optimization step can scale in parallel, a first look at allocating a fixed optimization budget across them.
Abstract (translated)
URL
https://arxiv.org/abs/2610.36774
PDF
-
DocSkill: Skill-Integrated Agentic Reasoning for Multilingual Multi-Hop Document Question Answering
2026-10-27 07:14:59Jing Huang, Zhiya Tan, Changtao Miao, Shutao Gong, Yang Wang, Joey Tianyi Zhou, Jianshu Li, Ying YanAbstract
Multilingual multi-hop document question answering (MMHop-DocQA) requires locating and integrating evidence across complex layouts, document pages, and diverse languages. Existing approaches include single-pass processing and adaptive tool use, but learning to coordinate evidence acquisition, cross-page integration, and answer verification remains challenging. To address this challenge, we present DocSkill, an agentic framework with a learned skill-orchestration policy for adaptive evidence acquisition and reasoning. The policy selects, revisits, or skips skills according to the evolving evidence state. To construct supervision for skill orchestration, bounded Monte Carlo Tree Search (MCTS) explores alternative skill trajectories offline. These trajectories provide demonstrations for supervised fine-tuning (SFT) and preference pairs for Direct Preference Optimization (DPO). We further introduce MMHop-DocBench, a multilingual multi-hop document QA dataset with training and test splits, containing 60K QA pairs across 10 languages and eight document domains. Its HopGen pipeline builds hierarchical page-element graphs to derive traceable cross-page evidence paths and generate grounded QA pairs with page- and element-level evidence annotations. Experiments show that DocSkill with Qwen3-VL-8B achieves 41.0 accuracy and 49.2 F1 on MMHop-DocBench, exceeding the strongest evaluated baseline by 5.5 and 6.3 points, respectively, and achieving competitive performance on MMLongBench-Doc and DocBench.
Abstract (translated)
URL
https://arxiv.org/abs/2610.36759
PDF
-
Robust LLM Data Selection via Content-Invariant Gradient Attribution
2026-10-27 07:12:01Jia Gu, Liang Pang, Huawei ShenAbstract
Gradient-based data attribution has become an important tool for high-quality data selection in the instruction fine-tuning of large language models (LLMs), as it estimates the contribution of training examples through their influence on model optimization. However, we find that existing global gradient attribution methods suffer from surface-form bias: minor changes to semantically irrelevant details, such as wording or formatting, can produce highly inconsistent attribution scores even when the underlying content remains unchanged. To address this issue, we propose Content-Invariant Gradient Attribution (CIGA), a robust framework for instruction data selection. CIGA first identifies layers whose gradients are both influential for attribution and stable under semantics-preserving surface-form perturbations, thereby reducing formatting noise in gradient features. It then dynamically fuses these attribution signals with semantic representations of the input, combining optimization-level influence with content-level information. Extensive experiments across multiple benchmarks show that CIGA produces more stable rankings under surface-form perturbations while preserving downstream capabilities and improving average performance. CIGA also reduces attribution cost, offering a robust and efficient approach to data selection for LLM instruction tuning.
Abstract (translated)
URL
https://arxiv.org/abs/2610.36755
PDF
-
PsyJev: Psychological Analysis via System-1 Judgment
2026-10-27 06:03:37Fangyuan Liu, Sirui Zhao, Xiaohao Wang, Guo Cheng, Peng Zhang, Enhong ChenAbstract
Self-disclosures on the Web support diverse psychological analyses, from recognizing expressed experiences to identifying symptom-relevant information. We study a shared, question-conditioned decision interface for this heterogeneous problem in content analysis. We organize PsyJudgeBench, comprising 13 annotated datasets and 82 questions across six psychological areas, while preserving source-specific criteria and masks for unobserved labels. We introduce PsyJev, a System-1 model that directly produces distributions over answers specified in each request. Joint supervised training establishes the shared judgment function, followed by Judgment-Anchored RLCD (JA-RLCD), which adapts logit-space reward optimization to heterogeneous psychological judgments through balanced supervision, selective retention of correct SFT predictions, and question-group reward optimization. The Qwen3.5-based 2B, 4B, and 9B models achieve 64.33, 66.21, and 67.56 average test F1, respectively. The 9B model exceeds the strongest external comparison by 2.54 points. Controlled 2B comparisons examine the training stages and anchored refinement. Further analyses examine question-conditioned judgment, inference cost, and probability quality, complemented by qualitative comparisons across models.
Abstract (translated)
URL
https://arxiv.org/abs/2610.36689
PDF
-
Image Modeling: From Handcrafted Priors to Foundation Image Models
Abstract
This paper unifies six decades of image modeling, evolving from handcrafted mathematical tractable priors to solve inverse problems in imaging to data-driven deep foundation image models for both image generation and solving inverse problems, under a common Bayesian framework that formulates image modeling as an implicit or explicit probabilistic characterization of a data manifold. Every image restoration or generative task fundamentally seeks to answer the same question: What makes an image look like a real image? In the Bayesian framework, this question is encapsulated by the prior probability density p(x), or its more optimization-friendly counterpart, the log-prior log p(x). While early works in solving inverse problems, such as Tikhonov, Total Variation (TV), and Non-local priors relied on explicit handcrafted definitions of log prior to model local smoothness, modern techniques, such as Plug-and-Play (PnP) priors and Denoising Diffusion models, are rooted in algorithmic uncoupling and implicit manifold learning that apply to both inverse problems and conditional image generation. By grounding imaging inverse problems and conditional image generation under a single mathematical engine: the gradient of the log-prior, or the Stein Score Function, we bridge the gap between signal processing and machine learning communities, turning what seems like parallel developments into a single, evolution of image modeling from handcrafted analytical models to models learned from big data, demonstrating that the pursuit of the denoising score is the common thread connecting variational optimization (in MAP estimation) to posterior sampling (in modern foundation image models).
Abstract (translated)
URL
https://arxiv.org/abs/2610.36646
PDF
-
PIVOT: Bridging Planning and Execution in LLM Agents via Trajectory Refinement
2026-10-27 04:50:15Tuo Zhang, Alin-Ionut Popa, Yan Xu, Rui Song, Dimitrios DimitriadisAbstract
Large language model (LLM)-based agents frequently generate seemingly coherent plans that fail upon execution due to infeasible actions, constraint violations, and compounding errors over extended horizons. PIVOT (Plan-Inspect-eVOlve Trajectories) addresses this plan-execution misalignment through a self-supervised framework that treats trajectories as optimizable objects iteratively refined via environment interaction. The framework comprises four stages: PLAN generates candidate trajectories; INSPECT executes them and computes structured losses with textual gradients encoding plan-execution discrepancies; EVOLVE applies these signals to produce improved trajectories; and VERIFY performs a final global check against task constraints. A monotonic acceptance process ensures a non-decreasing solution quality. Empirical evaluations on DeepPlanning and GAIA demonstrate state-of-the-art performance: with human-in-the-loop (HITL) feedback, PIVOT establishes a strong upper bound up to 94% relative improvement in constraint satisfaction, while its fully autonomous variant retains substantial gains, showing that the core trajectory-refinement mechanism remains effective without external supervision. At the same time, PIVOT remains computationally efficient, requiring up to 3x to 5x fewer tokens than competing refinement methods. These findings establish that (self- or human-supervised) feedback-based trajectory optimization is a principled methodology for mitigating plan-execution gaps in autonomous agent systems.
Abstract (translated)
URL
https://arxiv.org/abs/2605.11225
PDF
-
Polarity--Time Interaction and Multi-Level Reliability for Event-Based Spacecraft Pose Estimation
2026-10-27 04:29:19Yonglong Zhang, Zongwu Xie, Yang LiuAbstract
Sparse event responses produce uneven keypoint errors whose effects on spacecraft pose depend on correspondence geometry and the reliability of successive observations. We present a reliability-aware framework connecting polarity--time localization, geometric recovery, event--CAD refinement, and trajectory optimization. A three-branch representation combines positive events, negative events, and recent-event timing through cross-branch attention and gating, followed by spatiotemporal aggregation. An ordered semantic-point head estimates relative correspondence precision for geometrically constrained pose recovery. Event-edge and CAD-projection agreement then guide local pose refinement under keypoint-geometry safeguards. At the sequence level, a ten-dimensional descriptor combines heatmap shape, reprojection consistency, and geometric conditioning to set time-aligned observation scales. On the official SPADES challenge test set, the complete system achieves a combined pose score of 0.0474, approximately 5.4\% below PASAN+IRIS. Validation ablations reduce the score from 0.0772 for base localization with uniform PnP to 0.0514 for the complete cascade; on identical single-frame observations, quality weighting reduces the score from 0.0549 with constant-weight optimization to 0.0514.
Abstract (translated)
URL
https://arxiv.org/abs/2610.36625
PDF
-
God Does Not Play Dice: Enforcing Batch Invariance in Agentic Kernel Generation
2026-10-27 04:27:11Youwei Bao, Chengpeng Wang, Simin ChenAbstract
Large language model (LLM) serving systems dynamically batch user requests for GPU execution, so the same request may be processed alongside different requests across runs. If the GPU kernels lack batch invariance (BI), this changing batch context can alter the numerical output for an unchanged input, undermining reproducibility. Existing kernel generation agents focus on correctness and efficiency but \emph{overlook BI}. We find 19 confirmed BI violations among 83 batch-applicable released artifacts. To address this limitation, we propose \tool, which combines contract-guided static dependency analysis with dynamic validation of bitwise consistency across batch contexts. \tool is designed as a plug-in component for general kernel agents, allowing their original optimization capabilities to be reused while adding BI enforcement. We integrate it as successive acceptance gates in the optimization loop, rejecting structurally risky or dynamically non-invariant candidates before performance benchmarking and returning feedback to the agent. Experiments across three agent frameworks under baseline and BI-aware generation on real world LLM serving tasks show that \tool improves the BI pass rate from 67.6\% to 91.9\% while retaining 91.8\% of the baseline geometric mean speedup. These results demonstrate that \tool can be incorporated into kernel optimization by agents as an additional reliability objective alongside correctness and speed.
Abstract (translated)
URL
https://arxiv.org/abs/2610.36623
PDF
-
Dream-COP: Guiding Algorithm Discovery for Combinatorial Optimization Problems in an Experience World
2026-10-27 04:22:22Kuan Xu, Jianming Wang, Ni Zhang, Yaoxin Wu, Yong Wang, Zhiguang CaoAbstract
Large language models (LLMs) enable algorithm discovery for combinatorial optimization problems (COPs) through iterative program generation and evaluation. A central challenge is how to organize search to discover better algorithms within a limited budget. Our key insight is that past search trajectories and their recorded outcomes form an experience world, which can be used to facilitate the COP solving through replay. We introduce Dream-COP, in which a \emph{policy agent} learns executable policies through \emph{dreaming} in this experience world to guide a \emph{discovery agent} in generating and refining COP programs. Specifically, policy replay provides feedback on the quality of the programs revealed and the search steps used, to guide policy selection more effectively. Across five COPs, \textsc{Dream} discovers better algorithms than fixed exploration in both heuristic component design and complete solver generation. Particularly, it matches the optimal values on KP and reduces the gap on OBPP from 2.112\% to 1.199\% while using 42.5\% fewer search requests. Moreover, the learned exploration policies transfer effectively across different COPs with the assitance of the experience world.
Abstract (translated)
URL
https://arxiv.org/abs/2610.36618
PDF
-
Mimetic Embodiment Transfer of Human-Humanoid Interaction from In-the-Wild Videos
2026-10-27 04:18:31Nghia Le (Hann), Khang Nguyen (Hann), Ba Long Tran (Hann), Thi Quynh Nhu Pham (Hann), Binh Nguyen (Hann), Tuong Do (Hann), Erman Tjiputra (Hann), Hoan Nguyen (Hann), Tung D. Ta (Hann), Baoru Huang (Hann), Hai-Nguyen (Hann), Nguyen (Hann), Anh NguyenAbstract
Enabling a humanoid to interact naturally with human partners requires preserving not only individual motions but also the spatial, semantic, and contact relationships underlying the interaction (i.e., hugging, dancing, hand-shaking, etc.). Furthermore, a practical solution should generalize across various humanoid embodiments with distinct morphologies and kinematic constraints. To tackle this challenge, we propose HumiTransfer, a mimetic embodiment transfer framework that converts in-the-wild human-human interaction videos into physically plausible human-humanoid interactions. HumiTransfer first reconstructs 3D human motion with consistent foot-ground contact, then transfers one participant to a target humanoid embodiment via interaction modeling, using weighted joint-relative directions and anatomical offsets to prioritize interaction-critical inter-body relationships, while post-retargeting optimization enforces balance, contact, and collision avoidance. Subsequently, we apply HumiTransfer to 1,200 in-the-wild videos, producing approximately 500,000 frames of diverse interactions for the 26-DoF Unitree R1 Edu and the 23-DoF HighTorque Mini-hi humanoid robot, resulting in our HumiMimic dataset. Across both robots, HumiTransfer achieves superior motion fidelity, preserves inter-body relationships, and maintains contact accuracy compared with existing baselines. Moreover, we demonstrate that HumiMimic supports motion generation methods for synthesizing interactive humanoid motions following a human partner. Finally, our real robot deployments confirm the physical executability of the transferred motions in real-world settings.
Abstract (translated)
URL
https://arxiv.org/abs/2610.36615
PDF
-
Less is more: learning what to profile for GPU kernel optimization
2026-10-27 03:31:23Yihan Wang, Peiyu Liu, Runyu Chen, Wei XuAbstract
Large language models have recently shown strong potential for automatically optimizing GPU kernels, yet their effectiveness depends critically on the hardware information available to the kernel optimization model. Existing systems typically rely on predefined profiling strategies, leaving what to profile largely outside the optimization process. We show that effective profiling is not simply about collecting more information: increasing the number of profiling metrics does not monotonically improve kernel optimization, while the most useful profiling information varies across kernels. Our theoretical analysis further shows that, under such heterogeneous profiling utility, kernel-conditioned selection can outperform any single fixed profiling configuration. Motivated by this observation, we propose AutoProfiler, a feedback-driven framework for learning a deterministic profiling tree from optimization feedback. The tree is iteratively refined through REWRITE and SPLIT operations, with an LLM generating candidate modifications and empirical evaluation determining their acceptance. Once constructed, the tree quickly gathers the required profiling information for any kernel, allowing the LLM rewriter to be invoked only once. Experiments on KernelBench show that our approach improves kernel optimization over fixed profiling configurations under constrained profiling budgets. These results suggest that hardware profiling should evolve from static metric collection toward profiling reasoning, making profiling itself a new dimension for improving kernel optimization. The codes are available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.36576
PDF
-
QESChunker: A Single Objective Unifies Overlapping and Non-Overlapping Chunking for RAG
2026-10-27 03:05:09Yifan Zhao, Qianyang Zhao, Xinyu Guan, kai wei, Yuming DengAbstract
Chunking is a fundamental preprocessing step in retrieval-augmented generation (RAG), yet most existing chunkers rely primarily on heuristics. We introduce the Question--Evidence-Supervised Chunker (QESChunker), to our knowledge the first method for constructing static chunk indices from question--evidence supervision. QESChunker constructs document-specific chunk indices by optimizing over human-annotated or LLM-generated question--evidence pairs, without access to held-out test queries. We formulate non-overlapping and overlapping index construction through a unified question--evidence utility-maximization objective, turning chunking into a supervised combinatorial optimization problem. Under the non-overlap constraint, the objective decomposes over chunks and can be optimized exactly in polynomial time by dynamic programming. When overlap is allowed, the objective becomes monotone submodular; exact optimization is NP-hard, but our greedy achieves a $(1-1/\sqrt{e})$-approximation. Across two self-constructed and three public datasets, both variants outperform existing chunking methods in complete-evidence recall in nearly all evaluated settings, under the same overlap regime and a fixed hybrid retriever. These results recast chunking as supervised, budget-aware index construction for evidence retrieval.
Abstract (translated)
URL
https://arxiv.org/abs/2610.36557
PDF
-
NextEval: An Agent-Based Solver and Benchmark for Derivative-Free Optimization
2026-10-27 03:02:04Shuo Chen, Li Cui, Cunxin Huang, Ya-Feng Liu, Minghang Xie, Meng Xu, Zaikun ZhangAbstract
Derivative-free optimization (DFO) addresses problems where derivatives are unavailable or costly to obtain. This setting often arises in simulation-based and experimental applications, where evaluating the black-box objective can itself be costly. Thus, it remains a central challenge in DFO to effectively exploit accumulated observations to select informative evaluation points under a limited budget. In this light, we introduce the NextEval solver, an agent-based DFO solver, which integrates the capabilities of a large language model (LLM) to guide the sequential selection of evaluation points. Specifically, we establish a system that combines an LLM agent with a numerical environment. The environment provides tools for analyzing past evaluations and executing numerical methods, while maintaining a shared evaluation history and method-specific state to support continuation across method switches. Using observations and tool feedback, the LLM agent selects tools, configures their parameters, and allocates the evaluation budget, while numerical methods compute and evaluate candidate points. Building on this system, we also introduce NextEval Bench, a benchmark protocol for comparing agent configurations on shared DFO tasks using the same numerical tools under matched evaluation budgets and information constraints. Numerical results on low-dimensional DFO problems demonstrate that the NextEval solver outperforms the state-of-the-art DFO baselines in function-evaluation efficiency.
Abstract (translated)
URL
https://arxiv.org/abs/2610.36553
PDF
-
Agentic Data Sampling Framework for Multilingual ASR via Linguistic-Aware Error-Driven Optimization
2026-10-27 02:49:36Zhiheng Qian, Junhao Du, Jing Peng, Hanqi Li, Kunyang Peng, Feng Xue, Lu Chen, Shuai Wang, Kai YuAbstract
Multilingual automatic speech recognition (ASR) has achieved rapid progress with large-scale pretrained models and multilingual speech corpora. Previous studies have proved the pivotal role of data sampling in multilingual ASR performance. However, existing sampling strategies largely rely on time-consuming manual, heuristic-based designs. To address this, we propose a linguistic-aware agentic data sampling framework that introduces a data sampling LLM agent into the multilingual ASR training loop. The agent is driven by a linguistically grounded rubric that analyzes ASR errors from phonological, morphological, and syntactic perspectives, and uses the resulting feedback together with training statistics, budget constraints, and algorithm profiles to select domain-level, sample-level, and token-level sampling strategies. This design turns evaluation results into adaptive data-selection decisions rather than treating WER/CER as final endpoints only. We conduct experiments on twelve Common Voice languages using Whisper-medium as the ASR backbone. The results show that our method achieves performance better than full-data training while using less training data and lower computational resources, and substantially accelerates training convergence compared to conventional sampling strategies.
Abstract (translated)
URL
https://arxiv.org/abs/2610.36538
PDF
-
GarmentSegAgent: Autonomous Garment Segmentation for Web Fashion Images via Textual Self-Refinement
2026-10-27 02:10:25Jia Chen, Chao Huang, Xinrong Hu, Mingdi Huang, Ruyang Zhao, Jun BaiAbstract
Web fashion images support applications such as visual product search and consumer-to-shop matching, yet their visual heterogeneity makes reliable garment segmentation challenging. Despite advances in promptable segmentation, autonomously identifying and correcting garment-specific errors without human-provided spatial prompts remains difficult. We propose GarmentSegAgent, an autonomous framework that integrates semantic planning, spatial execution, and visual evaluation through iterative textual self-refinement. At its core, we introduce CAST (Constrained Attribution-guided Selective TextGrad), which connects visual error attribution with targeted textual optimization. CAST identifies instruction components responsible for segmentation errors, selectively updates them through textual gradients, and validates the resulting instructions against garment-specific constraints. This enables image-specific error correction at inference time while keeping the underlying foundation models frozen. Experiments on DeepFashion2, ModaNet, and Fashionpedia demonstrate consistent improvements in region overlap and boundary accuracy, alongside more reliable refinement than conventional TextGrad. By enabling visual feedback to guide controlled instruction updates, GarmentSegAgent offers a promising direction for autonomous garment understanding, with potential applications in scalable fashion content analysis, visual retrieval, and consumer-to-shop matching on the Web.
Abstract (translated)
URL
https://arxiv.org/abs/2610.36506
PDF
-
ProCerDCL: Query-Efficient Black-Box Probabilistic Verification of Deep Learning Compiler Optimization
2026-10-27 01:57:08Zexin Li, Weibo Zhang, Linyi Li, Arvind Easwaran, Simin ChenAbstract
We present \tool, a query-efficient framework for black-box probabilistic verification. Given a fixed, sampleable input distribution and a binary property oracle, \tool assesses the property's violation probability $p$ against a threshold $\theta$. Under independent sampling and stateless oracle behavior, acceptance asserts $p\le\theta$, rejection asserts $p>\theta+\eta$ for a user-specified separation margin, and an inconclusive verdict indicates insufficient evidence for either conclusion. The probability of each type of erroneous verdict is bounded by its specified error budget. To support this three-valued contract efficiently, \tool uses a single sequential test, eliminating the interval family required by the baseline. The test accounts for repeated sampling checks through either corrected exact binomial confidence intervals or an anytime-valid confidence sequence. We apply \tool to probabilistic local model semantic consistency, where the property oracle detects decision-level inconsistencies between reference and compiled executions under a specified perturbation distribution. Our evaluation covers 24 tasks across TorchInductor and TensorRT and examines sampling efficiency, conclusiveness, and diagnostic behavior. For sampling efficiency, we use paired one-model clean-label radius searches: the baseline requires 3.6-4.2 times as many unique samples under matched numerical parameters while providing a weaker decision contract. For model semantic consistency, checks at $\theta=10^{-3}$ yield 817 acceptances, 19 inconclusive verdicts, and 4 rejections across 840 input--backend instances from 21 tasks. Case studies of two documented compiler faults further demonstrate what these local verdicts can diagnose.
Abstract (translated)
URL
https://arxiv.org/abs/2610.36499
PDF
-
Proposing Better Rollouts for On-Policy Distillation
2026-10-27 01:33:51Sirui Chen, Yunzhe Qi, Haobo Xu, Jiaru Zou, Mengting Ai, Zihao Li, Jingrui HeAbstract
On-policy distillation (OPD) improves language-model post-training by letting the student learn from its own sampled trajectories under teacher supervision, aligning the training process with inference-time behavior. However, the rollout policy in OPD is usually treated as fixed: each round samples directly from the current student. We show that this choice can create a compounding teacher--student mismatch, because it requires the teacher to provide effective supervision on whichever trajectories the student samples, including ones less reachable by the teacher itself. On such trajectories, teacher supervision can become noisy or even misleading, leading to undesirable student patterns such as repetition. Moreover, since the updated student in turn defines the rollout policy of subsequent rounds, these patterns persist and potentially amplify over training. This failure mode suggests that improving OPD requires treating the rollout policy itself as an optimization target, rather than only modifying the distillation objective. We propose Mixture Proposal On-Policy Distillation (MPOPD), which derives a student-anchored, teacher-guided rollout distribution from a local trust-region objective based on relative teacher support. The resulting target is a geometric mixture of the student and teacher policies, and we introduce a proposal policy that approximates this mixture, enabling rollout generation in a single forward pass. Experiments on 16 math, knowledge, and medical reasoning benchmarks show consistent gains over OPD and recent variants. On math reasoning, our method improves average accuracy by 6.91 over OPD on Qwen3-4B-Base while reducing average generation length by 41%, suggesting that rollout policy design provides an effective way to improve OPD.
Abstract (translated)
URL
https://arxiv.org/abs/2610.36477
PDF
-
A Singularity Avoiding Track Model for Race Car Optimization
2026-10-27 00:24:49Adwait Verulkar, Behzad Danaei, Srikanth Sivaramakrishnan, Ryan TrinterAbstract
A track fitting pipeline is presented that converts a measured racetrack point cloud into a closed, $C^4$ differentiable, three-dimensional surface model suitable for high-fidelity Modelica race-car optimization. The surface is a Monge patch, parametrised by a curvilinear arc-length coordinate measured along the minimum-curvature line (MCL) and a signed lateral offset, with a height polynomial of user-selectable degree capturing the cross-sectional elevation profile. The choice of MCL as the reference line is deliberate since anchoring the same surface on the geometric centre-line can cause adjacent normals to cross inside the drivable region on tight corners. The resulting non-injective parametrisation assigns multiple curvilinear addresses to the same three-dimensional point and returns invalid road geometry. The MCL is obtained from a quadratic program that minimises integrated curvature subject to track-boundary, periodicity, and jerk-regularisation constraints. By reducing total curvature while remaining within the track bounds, the MCL moves the region of parametrisation breakdown outside the drivable strip on all circuits studied, keeps the normal map injective across the full track width, and yields a more linear time-to-arc-length transformation for minimum lap-time problems. The fitted surface and the associated track widths are exported as periodic piecewise-quintic $C^4$ B-splines, solved as a nonlinear program.
Abstract (translated)
URL
https://arxiv.org/abs/2610.36430
PDF
-
Repulsive Population Policy Gradients for Behaviorally Diverse Reinforcement Learning
2026-10-27 00:16:33Saeed Hedayatian, Harsh Shah, Celine Mneimneh, Stefanos NikolaidisAbstract
The goal of Quality-Diversity Reinforcement Learning (QD-RL) is to find a population of high-performing policies that solve a decision-making problem in diverse ways. Despite recent progress, QD-RL methods predominantly rely on evolutionary operators--including black-box optimizers, mutations, and crossovers--to facilitate exploration and induce diversity. Scaling these methods becomes increasingly costly as policies and behavior spaces grow larger. Building on the Soft QD view of quality-diversity optimization, we introduce \textbf{Re}pulsive \textbf{P}opulation \textbf{P}olicy \textbf{O}ptimization (\textbf{RePPO}), which adapts PPO to jointly optimize a compact population of diverse, high-quality policies using a population-dependent, dynamically shaped reward. On standard continuous-control tasks, RePPO achieves a strong quality-diversity tradeoff compared with state-of-the-art QD-RL methods. Ablations further characterize this tradeoff by examining how each component affects individual quality and collective diversity. To assess RePPO's scalability, we extend our evaluation to learning from pixels and fine-tuning large language models. In pixel-based tasks, where large policies make most traditional QD-RL methods infeasible, RePPO maintains quality on par with standard PPO while achieving substantially higher diversity. Its successful application to language model fine-tuning further demonstrates, for the first time, that QD-RL methods can scale to this domain. Together, these results show that directly optimizing compact populations with RePPO provides a scalable approach to QD-RL across increasingly complex domains.
Abstract (translated)
URL
https://arxiv.org/abs/2610.36425
PDF
