Optimization
-
High Scores, Hidden Failures: Evaluating Codebook-Based LLM Content Analysis
2026-10-26 10:40:54Elena Senger, Yuri Campbell, Rob van der Goot, Barbara PlankAbstract
Large language models (LLMs) are increasingly used for codebook-based content analysis in computational social science, where codebooks translate theoretical concepts into labeling schemes. While prior work reports strong predictive performance, it remains unclear whether these systems produce reliable and robust text annotations. We conduct a large-scale empirical failure-mode analysis of LLM-based coding pipelines across four datasets and five models by systematically evaluating their stability, robustness, and conceptual validity under controlled perturbations, domain shifts, and optimization procedures. We uncover multiple failure modes that are largely invisible under standard metrics, showing that high aggregate scores frequently coexist with severe label-level disparities, instability across runs, domain sensitivity, and shifts in label interpretation. Importantly, these failure modes persist under widely used optimization strategies and remain largely invisible under standard metrics such as accuracy and micro-F1. Our results demonstrate that current LLM-based coding pipelines cannot be treated as stable measurement tools without targeted validation beyond aggregate performance.
Abstract (translated)
URL
https://arxiv.org/abs/2610.35179
PDF
-
AutoEnergyDesign: An LLM-Driven Agent for Automatic Energy System Design Optimization
2026-10-26 10:25:55Amedeo Buonanno, Salvatore Fabozzi, Maria ValentiAbstract
Designing building energy systems is a high-dimensional, mixed-integer search problem: when each design is evaluated by a year-long simulation, black-box optimizers typically need hundreds of evaluations. This paper presents AutoEnergyDesign, a framework in which a Large Language Model (LLM) agent reasons over previous simulation results, proposes configuration changes and learns in context from each outcome, while a git-based mechanism automatically reverts unsuccessful changes. The agent couples a reasoning model with an explicit memory organisation (procedural, semantic, episodic and reflective), a constrained action interface and an environment-grounded keep/revert signal. The simulator covers eight technologies (photovoltaics, micro-wind, combined heat and power, solar thermal, gas boilers, batteries, heat pumps and inverters) and computes the 25-year Net Present Cost (NPC), including capital and operating costs and salvage value. On a case study with measured building load profiles, the agent, run with Claude Haiku 5.5, Sonnet 5.5 and Opus 5.5 (ten runs of 50 simulations each), is compared with random search, Tree-structured Parzen Estimator (TPE) Bayesian optimization and a $(1+1)$ evolution strategy sharing the agent's acceptance rule (300 evaluations, ten seeds). With all three models the agent reaches an NPC of about 3.53 MEUR, 1.20% below the best TPE result and 0.23% below the best evolution-strategy result. It comes within 1% of this design after a median of 18 (Opus) to 27.5 (Haiku) simulations, whereas among the baselines only the evolution strategy does so, in half of its seeds and after a median of 192 evaluations. Opus reaches the best design in all ten runs, Sonnet in six (two runs never come within 1%), and Haiku comes within 1% in every run but reaches it exactly only once. Every decision is recorded in an auditable trace.
Abstract (translated)
URL
https://arxiv.org/abs/2610.35160
PDF
-
UTA: Universal Training Adaptive Learning Rate Decay
2026-10-26 09:53:01Anda Tang, Junjie Wang, Xuanlin Yang, Cong Fang, Zhouchen LinAbstract
Learning rate decay (LRD) is critical for training high-performance models, yet most designs remain empirically driven. Although the Universal Budget Aware (UBA) LRD offers a theoretical framework, its reliance on SGD dynamics limits its applicability to modern Adam-type optimizers and leaves its key `optimization difficulty parameter' opaque. To bridge this gap, we propose the Universal Training Adaptive (UTA) LRD. Unlike UBA's fixed offline trajectory anchored solely to iterations, UTA treats LRD as a closed-loop shape-and-scale problem. By continuous-time formulation, UTA is training-state adaptive with segment-wise optimal shape. This adaptivity links external iterations with internal training states, enabling a near-optimal path in both progression shape and learning rate (LR) scale, a property unattainable by other LRDs, including UBA. Empirically, we demonstrate consistent improvements across language models (LLaMA 60M to 1B) and vision models (ResNet50, ViT-B) on ImageNet with plug-and-play ease and negligible overhead. Extensive ablation studies further confirm UTA's robustness across dimensions, including peak LR, final decay ratio, its parameter $\varphi_0$, component efficacy and generalization to fine-tuning scenarios, underscoring its practical implementability.
Abstract (translated)
URL
https://arxiv.org/abs/2610.35113
PDF
-
Joint Dynamic Lane Reversal and Variable Speed Limit Control: A Graphormer-Based Approach
2026-10-26 09:17:05Hamed Araab, Abbas Ahmadi, Mohammad Hossein BasiriAbstract
Connected and automated vehicles could allow reassigning lane directions and speed limits on every road frequently, but the benefit hinges on how traffic reroutes, and one-way roads can isolate junctions. We formulate network-wide dynamic lane allocation and directional speed limit control with one-way roads, lane clearing before each reversal, and a navigable graph kept strongly connected at all times. An action-resolution algorithm grounded in graph orientation theory admits a proposal closing a direction only if the network stays strongly connected and otherwise relaxes or reverts it, keeping all junctions mutually reachable within the model, even during clearing. All road controllers share one transformer whose attention, Global Endpoint Relation-Aware Spatio-Temporal Self-Attention (GERASTSA), spans all road-time tokens, biased by the endpoint relations between roads. It is pretrained on action-conditioned traffic forecasting and fine-tuned with per-road proximal policy optimization under a mixed reward that also credits neighboring roads and the network. On a 6 x 6 grid with commuter rush hours, the learned policy makes about two thirds of roads one-way in the peak direction and cuts daily delay by 90% against the initial configuration held fixed and by 85% against a tuned two-way load balancer. Restricted to two-way roads at test time, it gains only about as much as this balancer, whereas its learned speed limits add delay. Pretraining roughly halves the updates needed to surpass the fixed configuration. Of the ablations, only removing the relational bias or the spatial embedding clearly matters, multiplying delay about fivefold.
Abstract (translated)
URL
https://arxiv.org/abs/2610.35070
PDF
-
Reinforcement Learning Done Right for Explicit Confidence Targets
2026-10-26 09:05:29Zefeng He, Yichen He, Yuan Lin, Yang Yang, Yu ChengAbstract
Large language models (LLMs) can produce plausible but incorrect answers when they are uncertain, which can mislead users and undermine model reliability. Training models to abstain with reinforcement learning is a natural approach, but most existing methods use a fixed error penalty and therefore cannot adapt their behavior to different confidence requirements at inference time. We study \emph{explicit confidence targets}, where a user specifies a target \(p\) in the prompt and the model directly chooses whether to answer or abstain, with the goal of achieving an accuracy of at least \(p\) among answered questions. However, learning to respond to different confidence targets with a single model has proven difficult, and prior attempts have failed to distinguish different targets. We show that this failure comes from two problems: the reward signal that distinguishes different values of $p$ is often lost during optimization, while the original error penalty grows without bound as $p$ approaches one. We propose RLEC (\textbf{R}einforcement \textbf{L}earning with \textbf{E}xplicit \textbf{C}onfidence Targets), which preserves the learning signal needed to distinguish confidence targets, introduces a stable reward family that keeps the desired answer condition unchanged, and uses reward scheduling to balance effective learning and training stability. Together, these designs enable the first successful training of a single LLM for explicit confidence target control. Extensive experiments across diverse benchmarks and model scales show that RLEC largely preserves general capabilities and achieves competitive calibration performance. RLEC also supports fine-grained target control without an external decision system or threshold calibration after training. By making explicit confidence targets part of the trained policy, RLEC opens a new and more direct path toward behavioral calibration.
Abstract (translated)
URL
https://arxiv.org/abs/2610.35059
PDF
-
SimPOTrans: Self-Adaptation for Offline-to-Simultaneous Speech Translation via Reinforcement Learning
2026-10-26 08:54:00Haotian Tan, Hiroki Ouchi, Sakriani SaktiAbstract
Simultaneous speech translation (SimulST) aims to translate ongoing speech with low latency while preserving translation quality. Recent studies have adopted large language models (LLMs) for SimulST and formulated it as a multi-turn dialogue task for efficient generation, achieving state-of-the-art performance. However, existing dialogue-style SimulST approaches typically rely on specialized training schemes to acquire simultaneous translation behavior, either using synthesized read/write trajectories, which may provide suboptimal training signals, or jointly optimizing translation quality and latency, whose competing objectives may cause optimization instability. In this paper, we introduce SimPOTrans, a reinforcement learning (RL) framework that enables offline speech translation models to self-adapt to dialogue-style SimulST. Through a simple two-stage RL post-training pipeline, SimPOTrans rewards an offline model for exhibiting translation behaviors beneficial to simultaneous scenarios when given partial speech inputs, thereby enabling offline-to-simultaneous self-adaptation without synthesized read/write trajectories or explicit latency optimization. At inference time, SimPOTrans requires no external decision policy and enables flexible quality--latency control by simply adjusting the chunk size of speech input. Experiments on the ACL 60/60 and MCIF benchmarks across three language directions (En$\rightarrow$De, En$\rightarrow$Zh, and En$\rightarrow$Ja) demonstrate that SimPOTrans outperforms several strong baselines, achieving a superior quality--latency trade-off.
Abstract (translated)
URL
https://arxiv.org/abs/2610.35042
PDF
-
Context-dependent Runtime Construction of Safe Regions of Attraction from Design-Time Identification for Multi-Candidate Simplex Architectures
2026-10-26 08:53:49Georg J\"ager, Nils-Jonathan Friedrich, Norman Seyffer, Jos\'e Cec\'ilio, Ant\'onio CasimiroAbstract
Deploying safety-critical cyber-physical systems requires control strategies that reconcile high performance and adaptability with rigorous safety guarantees. Classical control strategies offer stability proofs but are conservative, while advanced optimization- and learning-based controllers achieve high performance but lack formal verification. Runtime Assurance (RTA) architectures bridge this gap by introducing their own (dis-)advantages. Safety filters use online optimization and introduce chattering, while Simplex permissiveness is bounded by the tightness of the certified Region of Attraction (RoA) of its High-Assurance (HA) controller. Targeting this conservatism, we present a two-stage RoA certification procedure leveraging contraction-based stability to establish a target set for Taylor-model-based reachability analysis. We then treat the closed-loop system under HA control as a switched system, enabling: (a) runtime construction of context-dependent safe RoA and (b) safe switching between HA candidates to optimize recovery and maximize the operating envelope of the uncertified High-Performance (HP) controller. We validate this framework on a robotic path-tracking task using a three-stage evaluation: Ordinary Differential Equation (ODE) simulation for theoretical verification, high-fidelity Gazebo simulation under unmodeled dynamics, and experiments on physical hardware. Results show that the Simplex architecture maintains safety under fault injection and successfully shields a behavioral-cloning-trained HP controller. Crucially, the decision module introduces negligible overhead (0.028 ms average, 4.13 ms worst-case), consuming under 10 % of the control period.
Abstract (translated)
URL
https://arxiv.org/abs/2610.35040
PDF
-
AutoPRL: Automated Path Representation Learning via a Learnable View Generator
2026-10-26 08:45:38Sean Bin Yang, Ying Sun, Yunyao CHeng, Hao Miao, Jilin Hu, Bin Yang, Torben Bach Pedersen, Christian S. JensenAbstract
Self-supervised learning has emerged as a powerful paradigm for Path Representation Learning (PRL) in road networks. PRL aims to learn informative representations of paths traversed by moving objects, such as vehicles, where each path consists of a sequence of connected road segments. These representations are essential for various downstream tasks, including travel-time estimation and path ranking. However, existing methods often suffer from ineffective view generation and inefficient training procedures due to their reliance on heuristic augmentations and computationally intensive contrastive learning. To address these issues, we propose AutoPRL, an effective and efficient self-supervised PRL framework that integrates three key components. The Automatic Data Augmentation (AutoDA) module employs variational autoencoders to generate semantically meaningful and diverse feature-level path views by modeling each topology-aware node embedding with a Gaussian latent distribution, thereby mitigating information loss and enhancing representation quality. Second, the Cross-View Decoupled Contrastive (CVDC) learning mechanism improves training efficiency by disentangling the optimization of positive and negative pairs across node- and path-level views, enabling more stable contrastive learning. Third, an Adaptive Co-Training (ACT) module dynamically adjusts uncertainty-based loss weights to balance multiple objectives, ensuring efficient and scalable training. Extensive experiments on three real-world datasets demonstrate that AutoPRL achieves strong and consistent performance in both effectiveness and efficiency across multiple downstream tasks.
Abstract (translated)
URL
https://arxiv.org/abs/2610.35032
PDF
-
Multiscale Policy Optimization for Long-Horizon LLM Agents
2026-10-26 08:22:11Tao Ren, Hui Yang, Jia Deng, Tingyang Chen, Qiyao Wang, Shuo Tang, Ziyang Zeng, Yuhua Jiang, Guanghao Li, Huatong Song, Yining Li, Yifan Luo, Kunbao Ding, Feng Chang, Chuan Hao, Ran Tao, Bryan Dai, Jiafeng Liang, Jinyang JiangAbstract
Long-horizon LLM agents interact with their environments through complete assistant turns while generating each action token by token. We observe that critic values vary primarily across turns, motivating optimization anchored at turns with finer credit allocation among tokens. We propose Multiscale Policy Optimization (MPO), combining Multiscale Advantage Estimation (MAE) and turn-level importance sampling to align credit assignment and policy updates at the turn scale. Extensive experiments across search, tool-use, and automation tasks with 4B and 30B-A3B models demonstrate consistent gains over strong RL baselines.
Abstract (translated)
URL
https://arxiv.org/abs/2610.35013
PDF
-
DUET4D: Self-Corrective Coupled Guidance for Sparse-View 4D Reconstruction
2026-10-26 08:14:00Beizhen Zhao, Nanjie Yao, Boyi Fu, Qingsong Yan, Haofan Ren, Kaiyong Zhao, Hao WangAbstract
Sparse-view 4D reconstruction is highly under-constrained when only a few cameras are available. Generative priors can recover missing appearances, but video models are computationally impractical for iterative reconstruction, whereas single-view models produce temporally inconsistent pseudo-observations. Moreover, directly absorbing uncertain generative outputs into Gaussian optimization often creates redundant primitives and floating artifacts. To address these challenges, we propose DUET4D, a self-corrective coupled guidance framework for sparse-view 4D Gaussian splatting. The core of our approach is a self-corrective framework where the 4D representation guides generative prior for spatio-temporal consistent supervisions. We first construct sequential alignments from current Gaussian evidence to steer the latent restoration. Then we use Sequential Fidelity Resampling to lift generative supervisions into Gaussian space to selectively distinguish repeatedly supported primitives. The corrected representation then provides more reliable Gaussian evidence for the next restoration cycle. Extensive experiments demonstrate that DUET4D significantly improves rendering quality and rendering efficiency.
Abstract (translated)
URL
https://arxiv.org/abs/2610.35002
PDF
-
Improving Reasoning with Rollout Agreement for Group-Based Policy Optimization
2026-10-26 07:19:26Sohyun Son, Yerim Hong, Jaeyun Park, Mujeen SungAbstract
Recent methods for Reinforcement Learning with Verifiable Rewards (RLVR) use rollout diversity to improve reasoning performance or training efficiency. Despite these gains, the extent to which differences among correct-answer rollouts reflect useful alternative reasoning or unresolved errors remains poorly understood. We investigate this question through a blind paired analysis of a model trained with Group Relative Policy Optimization (GRPO), comparing the highest- and lowest-agreement correct-answer traces for the same problem. Across 126 pairs, low-agreement traces have more spurious reasoning than high-agreement traces (35% vs. 22%), including internal inconsistencies and misuse of problem conditions. Based on these findings, we propose RoAR (Rollout-Agreement Reward), which adds reasoning-level agreement to GRPO's rewards for verified-correct rollouts. RoAR computes this signal from existing on-policy rollout groups, requiring no human-written rationales, additional sampling, or separately trained process evaluator. On five mathematical reasoning benchmarks, the 1.5B GRPO-RoAR model reaches 55.5% mean pass@1 under zero-shot evaluation using 7K training examples, exceeding GRPO and GRPO-CARE by 6.0 and 5.5 percentage points. A separate blind audit of 100 problems answered correctly by all four 1.5B models finds spurious rates of 18% for GRPO-RoAR, 32% for GRPO, 27% for GRPO-CARE, and 37% for DRA-GRPO. Our findings identify reasoning validity as an important consideration when using rollout differences to guide learning. Our code is available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.34951
PDF
-
Correlation-Aware Exploration for Long-Trace Reasoning
2026-10-26 07:19:00Junxing Hu, Zicheng Zhang, Zhiqian Zhang, Ai Han, Yifeng Zhang, Pengzhang Liu, Ling LiAbstract
Reasoning models trained with Group Relative Policy Optimization (GRPO) now produce chains of thought thousands of tokens long, and recent studies reveal that such training is prone to policy collapse. Entropy regularization is the standard remedy, but it works by pushing probability mass away from the few tokens that carry it and toward the vast low-probability tail of the vocabulary, and it applies this force once for every token of the reasoning trace. This puts the bonus in direct conflict with the entropy decay that convergence requires, and makes its coefficient difficult to set in the long-trace regime. To resolve this tension, we propose Info-GRPO, an information-theoretic framework that cultivates correlation between the policy and a latent prior. Rather than raising entropy everywhere, Info-GRPO builds on the mutual information between a latent variable and the trajectory it induces: augmenting prompts with latent seeds during training lets the model explore diverse policies correlated with the prior, while the seed-conditioned entropy is guided toward convergence. Diversity is therefore carried across whole reasoning traces rather than by per-token randomness, and exploration no longer fights entropy reduction. Experiments show that Info-GRPO outperforms vanilla GRPO and entropy-regularized GRPO across diverse reasoning benchmarks. On AIME24, it improves $Avg@8$ over GRPO by 3.75%, 1.66%, 4.16%, and 4.17% on four backbones, and it reaches the same majority-vote accuracy with $8\times$ fewer samples. Code and models will be publicly released.
Abstract (translated)
URL
https://arxiv.org/abs/2610.34949
PDF
-
Learning Direct Control from Optimization for Dynamic Target Tracking
2026-10-26 06:55:53Bing Sun, Yue Lin, Canhui Wen, Xin Chen, Yang Liu, Dong Wang, Huchuan LuAbstract
Dynamic target tracking requires a mobile robot to maintain target visibility while navigating obstacles with smooth, feasible controls. Online trajectory optimization explicitly encodes these objectives but requires repeated numerical solving, whereas waypoint-based neural policies separate geometric prediction from control execution. We introduce LECO, a framework that transfers tracking optimization into offline learning of a direct-control policy. LECO predicts control sequences and maps them to trajectories through a differentiable kinematic model, allowing trajectory-level objectives to directly guide control learning. A decaying expert-guidance curriculum initializes effective tracking behavior and progressively shifts learning toward these task objectives. At deployment, a predictive safety lock checks short-horizon obstacle clearance and invokes a classical planning and control fallback when the predicted motion is unsafe or the target prediction is invalid. Across three simulated obstacle densities, LECO reduces root-mean-square linear jerk by 14.1--31.2\% relative to the best competing method for this metric and achieves the highest tracking success rate, with per-cycle computation times of 4.25--5.63\,ms. Experiments on two robot platforms further demonstrate tracking through indoor narrow passages and outdoor forest environments. These results demonstrate that learning direct control from optimization enables smooth and robust dynamic target tracking.
Abstract (translated)
URL
https://arxiv.org/abs/2610.34929
PDF
-
Meta Inverse Prompting for Video Generative Models
2026-10-26 06:42:28Yerin Jung, Jongheon JeongAbstract
Recent advances in video generative models have enabled flexible conditional generation and editing pipelines, such as text-to-video, image-to-video, and reference-guided synthesis. However, these workflows are inherently one-directional, mapping conditions to videos without providing a mechanism to invert this process. In complex media production settings, the ability to recover controllable textual conditions from a given video is highly desirable, yet remains underexplored. While inverse prompting has been studied for image generation, directly extending these approaches to video is challenging due to the need to capture temporal dynamics beyond static appearance, as well as the substantial computational cost of per-video optimization. In this work, we propose the first practical framework for efficient video inverse prompting. Specifically, we formulate inverse prompting as a meta-learning problem and introduce Meta Video Inverse Prompting (MVIP), which learns a meta-prompt optimized over temporal dimensions to enable fast and effective prompt inference. Our approach amortizes the inversion process, eliminating the need for costly iterative optimization at test time. Experimental results demonstrate that MVIP recovers higher-quality inverse prompts with significantly improved reconstruction fidelity compared to image-based inversion and iterative captioning baselines, while achieving substantial gains in efficiency. Code and video results are available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.24965
PDF
-
Waypoint: Analytically Modeling GEMM for CPUs
2026-10-26 06:40:51Srinidhi N, Sunanda Somwase, Hemant Kumar Chandrakant Chandavar, Eashan Dash, Varun Agrawal, Prakash RaghavendraAbstract
High-performance General Matrix Multiplication (GEMM) implementations rely on breaking down GEMM into carefully selected tiles to achieve peak performance. We present Waypoint, a new analytical model to automatically identify the optimal tile size for a given hardware and input matrix shape. Waypoint estimates the achievable performance for a given tile choice by capturing all the architectural effects including interactions with memory hierarchy, instruction-level parallelism, and multi-threading. Using the estimates as a proxy, we search through the space of possible tile sizes to identify the one that maximizes performance. Additionally, Waypoint helps hardware designers estimate realistic performance on experimental hardware optimizations. We build a new GEMM library OPAM for CPUs that uses Waypoint to automatically break down GEMM and achieve performance that matches state-of-the-art GEMM libraries and helps accelerate LLM inference on CPUs achieving $1.36 - 1.98 \times$ speedups.
Abstract (translated)
URL
https://arxiv.org/abs/2610.34914
PDF
-
Quantile-Based Command-Reserve MPC for Current-Command Gimbal Pointing
2026-10-26 06:37:46Yaokun Lu, Yifan Li, Enmei Wang, Hao Teng, Pengwei Hu, Jianzhong QiaoAbstract
This paper studies the constrained line-of-sight pointing control problem of a two-axis current-command gimbal subjected to mismatched mechanical disturbances. Friction, torque ripple, current-response lag, model mismatch, and limited drive commands make accurate pointing more challenging yet interesting. A quantile disturbance predictor is proposed to compensate predictable acceleration residuals and describe their remaining uncertainty, with the aid of learning median and outer quantiles from state, command, and reference histories. A tube-inspired model predictive controller is proposed to reserve command headroom according to prediction uncertainty, with the aid of calibrated radii and stage-dependent input margins. The median improves the nominal prediction, while the radius adapts the command reserve without adding optimization variables. A finite-horizon tube inclusion result is established for a specified linear error model and fixed ancillary policy; its conditions are distinguished from the approximations used in hardware. Native command bounds and a final command limiter keep the sent commands within their prescribed range. On a two-axis embedded platform, four controller configurations are compared in seven repeated trials each. The proposed configuration achieves a two-dimensional pointing root-mean-square error of $3.94 \pm 0.10$ mrad with no command-limit samples in the evaluation windows. Its performance supports the combined use of residual prediction and adaptive command margins in the tested operating condition. Solver times and logged loop overruns are reported separately for the nominal 100 Hz implementation.
Abstract (translated)
URL
https://arxiv.org/abs/2610.34913
PDF
-
LE-GDA: Layer Ensemble Gaussian Discriminant Analysis for Training-Free Few-Shot CLIP Adaptation
2026-10-26 06:21:21Pavodi Maniamfu, Vogel Kiketa, Zezina Kuzulu Kuetu, Jakin Mukendi Kabongo, David Mayoya Kutangila, Pescie Honore Kinkani Mampuya, L'Or Wolo SiekaAbstract
Training-free few-shot adaptation of vision-language models typically constructs a classifier from CLIP's final projected image representation, leaving intermediate representations unused. Combining them is not straightforward. Class scores from different representations differ in spread by orders of magnitude for the same query. Direct averaging allocates influence by numerical scale rather than predictive information. We propose Layer-Ensemble Gaussian Discriminant Analysis (LE-GDA), a method that fits regularized shared-covariance discriminants independently to intermediate and final representations extracted in one frozen encoder pass. LE-GDA standardizes each discriminant's class scores per query, selects a small representation multiset on labeled validation data, and combines it with the zero-shot classifier without gradient-based optimization. Across 11 datasets, four CLIP backbones, five shot counts, and three independently sampled K-shot training sets, LE-GDA achieves the highest mean accuracy of all evaluated methods on every backbone, raising the 16-shot mean from 78.01% to 79.54% over a matched GDA-CLIP (final-feature) baseline and improving that baseline for every backbone-shot combination. At one shot, LE-GDA retains a 1.20-point mean gain over the baseline even though covariance estimation is unavailable and per-query standardization is neutral. A control with richer per-class structure inside a single representation does not reproduce the gain, indicating that the benefit comes from combining representations. These results show that intermediate CLIP representations carry information beyond the final embedding and can improve training-free few-shot adaptation. Code and evaluation resources are available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.34901
PDF
-
AIPC: Evidence-Guided Agent Workflows for PyTorch Project Deployment
2026-10-26 06:10:01Zhanwei Wu (Qualcomm Technologies, Inc), Jianhao Su (Qualcomm Technologies, Inc), Brian Zhang (Qualcomm Technologies, Inc), ShengTing Huang (Qualcomm Technologies, Inc), Weidong Feng (Qualcomm Technologies, Inc), Gang Liu (Qualcomm Technologies, Inc), Zhiguo Li (Qualcomm Technologies, Inc), Wei Huang (Qualcomm Technologies, Inc), Li Tan (Qualcomm Technologies, Inc), Wanjun Li (Qualcomm Technologies, Inc), Zejiong Teng (Qualcomm Technologies, Inc), Changgeng Zhong (Qualcomm Technologies, Inc), Guanping Wu (Qualcomm Technologies, Inc), Han-Hsing Tu (Independent Researcher)Abstract
Deploying and optimizing a PyTorch inference project on a dedicated accelerator, especially an on-device NPU, requires more than using a framework device API as with CUDA. The project must be converted into hardware-runtime artifacts and adapted to the target toolchain and application interface. Differences in model formats, quantization flows, operator support, and toolchain interfaces create cross-layer problems spanning PyTorch/ONNX models, toolchains, runtimes, and device execution. LLM Agents can invoke tools to address such problems, but may drift because of insufficient domain knowledge, large configuration and repair spaces, or poor path selection. We propose AIPC (AI Porting & Conversion), an Agent method for deploying PyTorch inference projects. AIPC combines structured Skills with toolchain, runtime, and profiling evidence to drive cross-layer diagnosis and repair, organizing model adaptation, artifact building, on-target execution, accuracy repair, and performance optimization into one deployment flow. We evaluate AIPC on the on-device NPU path of Qualcomm QAIRT across vision, speech, LLM, and diffusion models. In two optimization cases, 6/7 Skill-arm Whisper-Small runs meet predefined accuracy and latency criteria versus 0/6 control runs, while YOLO-Worldv2 achieves a 2.791x end-to-end speedup versus 1.599x for the control. Under a pre-fixed deployment contract, Skill, tool permissions, and check conditions, some runs achieve zero in-run human intervention (0-HITL). We also examine Qwen3.8-27B as a locally deployable model for driving an Agent workflow.
Abstract (translated)
URL
https://arxiv.org/abs/2610.34890
PDF
-
LayoutMamba: Efficient Document Layout Analysis with Hybrid Mamba-Transformer Networks
2026-10-26 05:51:46Kehua Chen, Mingyi ChenAbstract
Document layout analysis (DLA) faces a long-standing dilemma: existing methods either capture rich local details at prohibitive computational cost (e.g., Transformer-based approaches with quadratic complexity) or achieve efficient global modeling but sacrifice fine-grained structural awareness (e.g., state-space models). This trade-off hinders deployment in resource-constrained scenarios. We present LayoutMamba, a hybrid framework that synergistically combines sparse-attention Transformers for localized feature extraction (O(n) complexity) and the recurrent state-space mechanism of Mamba for dilution-free long-range dependency modeling across the 2D document canvas. Our two-stage paradigm first uses a lightweight Transformer encoder to hierarchically distill salient local semantics; second, Mamba propagates global contextual dependencies without attention dilution, enabling precise, efficient, and scalable layout understanding. We further identify a critical yet overlooked issue in existing DLA benchmarks: the inherent low-overlap distribution bias of layout objects, which we mitigate through a novel Geometric Overlap Penalty Loss (GOPLoss), a dual-optimization objective that minimizes redundant bounding box overlaps and encourages geometrically consistent predictions. Extensive experiments on PubLayNet and D4LA show that LayoutMamba surpasses state-of-the-art AP on PubLayNet with 8x fewer parameters and maintains leading performance on D4LA, ranking top-3 among published methods, achieving an unprecedented balance between accuracy, efficiency, and scalability. Our work pioneers a cognition-inspired paradigm for document analysis, advancing practical deployment in edge computing and resource-limited environments.
Abstract (translated)
URL
https://arxiv.org/abs/2610.34879
PDF
-
Priority-Preserving Advantage Aggregation for Multi-Reward Policy Optimization
2026-10-26 05:26:04Xiangchi Yuan, Xingjian Diao, Zheyuan Liu, Chunhui Zhang, Dachuan Shi, Kejing Xia, Zhenbang Du, Shenglong Yao, Soroush Vosoughi, Wenke LeeAbstract
LLM post-training often optimizes multiple objectives with unequal priorities, such as correctness and reasoning efficiency. However, the widely used additive reward aggregation in RL can violate these priorities: when primary and auxiliary advantages conflict, a sufficiently large auxiliary signal can overturn the preference induced by the primary objective. We term this problem \emph{advantage priority inversion}. Although reward-level gating can protect priority more directly, it still removes potentially useful auxiliary distinctions among rollout samples that fail the primary condition. To address this trade-off, we propose Primary Advantage Policy Optimization (\textbf{PAPO}), a priority-aware advantage aggregation method that asymmetrically combines a primary and an auxiliary advantage. PAPO retains auxiliary supervision when the objectives agree and smoothly attenuates conflicting auxiliary signals, preserving the primary sign during fusion for \emph{primary-consistent ordering}. This rule applies to sequence-level advantages in GRPO and token-level advantages in actor--critic PPO. Experiments on multiple domains and models show that PAPO consistently achieves the best primary objective while maintaining best or competitive auxiliary-objective performance across both GRPO- and PPO-based optimization. \href{this https URL}{Code is available}.
Abstract (translated)
URL
https://arxiv.org/abs/2610.34862
PDF
