Optimization
-
Li$^{2}$SAO: Lifting Light Singl-Rollout Asynchronous RL Optimization via Group-Relative Actor-Critic Bootstrapping
2026-10-29 09:02:00Zichen Bai, Yanli Wang, Yang Liu, Junhao Shi, Xueqin Liu, Jinguo Sun, Muye Li, Shaoqiang Zheng, Xuefei Diao, Zhuwei Zeng, Nan Duan, Peng CaoAbstract
Asynchronous reinforcement learning (RL) is becoming a mainstream paradigm for post-training large language models (LLMs), as it improves training efficiency by updating the model as rollouts arrive. Recent work argues that group-wise sampling in GRPO does not naturally fit asynchronous RL training because rollout lengths follow a long-tailed distribution, and proposes Single-rollout Asynchronous Optimization (SAO) to improve stability and off-policy behavior, especially on agentic tasks. However, SAO relies on a closed-source value model obtained from scaling value pretraining. In this paper, we analyze GRPO and LiSAO, our lightweight reproduction of SAO without the pretrained value model, on single-turn reasoning and multi-turn agentic tasks. We find that GRPO outperforms LiSAO on reasoning tasks, while the reverse holds on agentic tasks. Since the value model of LiSAO is not yet well trained , LiSAO lags behind GRPO substantially in early training steps. To address this issue, we propose Li$^{2}$SAO, which Lifts Light Single-rollout Asynchronous Optimization with group-relative actor-critic bootstrapping: while the critic is cold, a group-relative advantage trains the actor and a group-centered return trains the critic, after which training switches to single-rollout GAE. Experiments show that Li$^{2}$SAO catches up with or surpasses GRPO in early training and further raises the later performance ceiling. Code is available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.40726
PDF
-
Sparse Fine-Tuning for Parameter-Efficient Adversarial Training
2026-10-29 06:27:35Zhaoxin Wang, Zihang Ding, Handing WangAbstract
Deep neural networks achieve strong performance across many tasks but remain vulnerable to adversarial perturbations. Adversarial training (AT) is one of the most effective defenses, yet it suffers from high memory cost on large models and the persistent clean-robust accuracy trade-off. Recent work introduces parameter-efficient fine-tuning, such as LoRA, into AT, but this constraint induces a notable robustness gap relative to full-parameter AT. In this work, we revisit parameter-efficient adversarial fine-tuning from a parameter space perspective. Through gradient analysis, we find that adversarial optimization is highly concentrated on a small subset of parameters, which we call Robustness-Critical Parameters (RCPs), suggesting that robustness is encoded in a sparse subspace. Building on this observation, we propose Sparse Adversarial Fine-Tuning (SAFT), which identifies RCPs via adversarial gradient saliency and performs AT by updating only these parameters while freezing the rest. Across architectures and datasets, SAFT outperforms LoRA-based methods in both clean and robust accuracy with fewer trainable parameters, and approaches or surpasses full parameter AT in robustness while updating only about 5\% of parameters.
Abstract (translated)
URL
https://arxiv.org/abs/2610.39455
PDF
-
Timbre: Reliability-aware harmonic consensus for label-free multimodal test-time rexinforcement learning
2026-10-29 06:15:35Haotian Lyu, Haoyu Wang, Ru Wang, Feng Ji, Ziwen Li, Jiaxian Guo, Mingming GongAbstract
Label-free test-time reinforcement learning adapts vision--language models using rewards derived from their own predictions. However, misinterpreted visual evidence and sensitivity to textual formulation can produce a consistent but incorrect consensus that majority voting reinforces. Cross-view aggregation can expose such errors, but image and text transformations may themselves degrade task-relevant evidence to different extents. We introduce \method, a reliability-aware harmonic consensus framework for multimodal test-time reinforcement learning. \method constructs a $2\times2$ grid combining original and perturbed images with direct-solving and reframing modes, selects pseudo-labels through weighted harmonic aggregation, and adapts modality-level weights using batch-averaged, temporally smoothed shifts in pseudo-label confidence as a heuristic reliability signal. We formulate consensus as a weighted $\chi^2$-barycenter problem whose unique minimizer is the normalized weighted harmonic aggregate, without requiring correct-answer invariance. An exact support--dispersion decomposition characterizes when harmonic aggregation preserves or corrects arithmetic rankings, while a sensitivity analysis establishes how modality reweighting changes pairwise ranking margins, informing interpretation of the heuristic updates. Across six benchmarks and three Qwen3-VL model scales, \method outperforms TTRV in 16 of 18 benchmark--model settings on adaptation instances and 17 of 18 settings on held-out instances under matched estimation and optimization rollout counts. These results demonstrate improved target-task adaptation and transfer to unseen instances without further updates.
Abstract (translated)
URL
https://arxiv.org/abs/2610.39439
PDF
-
Learning to Formulate: Agentic Hardware-Software Co-Design of AI Infrastructure through Progressive Optimization and Verification
Abstract
AI infrastructure, the chips and software that train and serve AI models, is designed in layers. Layers make the system manageable, but they also fix which choices an optimizer can consider. Gains that need the chip and its software to change together are excluded before optimization begins: the formulation leaves them out. Language-model agents make it possible to revisit these boundaries, but an agent's proposal cannot show by itself that a design is correct or optimal. We present Formulaic, a flow that divides the work. The agent writes the formulation, a mathematical solver optimizes within it, and verification and measurement check its assumptions and results. The agent refines the problem step by step, from the computation graph toward the chip. Each step is a template: a family of programs with the conditions under which they are correct. Wall relaxation searches the chip by pricing the hardware limits that bind the software. This paper takes a first step, through separate studies rather than one end-to-end run. On NVIDIA GB10, templates the agent wrote run complete Qwen3-8B layers $1.075\times$ faster than SGLang. In a simulated Ventus GPU study, joint design cuts the worst-scenario cycles of Qwen2.5-0.5B to 0.428 of the reference, against 0.934 for software alone. On a five-step GEMM refinement chain, reusing proofs makes checking $2.51\times$ faster.
Abstract (translated)
URL
https://arxiv.org/abs/2610.39380
PDF
-
Semantic Mismapping in Generalized Category Discovery: Evaluation and Mitigation
2026-10-29 04:49:17Zitai Wang, Qianqian Xu, Cong Hua, Xi Chen, Zhiyong Yang, Xiaochun Cao, Qingming HuangAbstract
Generalized Category Discovery (GCD) aims to perform old-category recognition and novel-category discovery in the same unlabeled set. Early methods address both objectives through clustering and evaluate the exchangeable predicted indices with global Hungarian (GH) matching. Recent methods adopt trainable parametric heads, using supervision for old-category recognition and discovery-oriented objectives for new categories. In this way, the output space becomes partially exchangeable: old indices are semantically anchored, while new indices remain exchangeable. However, these methods still apply global matching, which can reassign anchored old indices to other categories and hide some prediction errors. Empirically, we are surprised to observe that this reassignment can induce an overestimation of new-category accuracy by up to 25.63 points. We term this phenomenon semantic mismapping and further derive a permutation-cycle condition that characterizes when semantic mismapping can increase the global matching score. To fix this issue, we propose Semantics-Preserving Evaluation (SPE), which preserves the supervised semantics of old indices and restricts Hungarian matching to the new region. Furthermore, we propose Mismapping-Mitigating Optimization (MMO) to alleviate the prediction errors that can give rise to semantic mismapping, combining a region-balance loss (RBL) for old/new routing with a novel-diversity loss (NDL) that encourages orthogonality among new prototypes. Theoretically, we show that minimizing the two losses can tighten an upper bound on spurious accuracy gains from semantic mismapping, with controllable residual terms. Empirical results under both GH and SPE demonstrate the effectiveness of our proposed method.
Abstract (translated)
URL
https://arxiv.org/abs/2610.39356
PDF
-
MemSteer: Turning User Histories into Supervision for Agent Memory Optimization
2026-10-29 03:28:49Jie Huang, Cuiling Lan, Qize Yu, Xiulian Peng, Yan LuAbstract
An effective agent memory system should adapt to a user's history and task requirements. Human-annotated supervision can guide this adaptation by providing feedback for evaluating and improving memory designs. Yet suitable labels are rarely available for new user histories, and manually constructing QA pairs and supporting evidence is costly. We introduce MemSteer, which turns unlabeled user histories into question-answer (QA) supervision with supporting evidence for memory-design optimization. MemSteer constructs an evidence-grounded answer plan that specifies how facts across related records jointly support a reference answer. The plan guides question generation, while source-grounded auditing and blind answer recovery check the resulting annotations. To study the utility of this supervision, we develop MemSteer-Opt, which combines QA-based design evaluation with evidence-guided failure diagnosis to revise memory store, retrieval, and context construction without updating model parameters. Experiments on ATM-Bench and RHELM show that this supervision is competitive with official-QA-guided optimization. Relative to Base-BM25, MemSteer-Opt improves ATM-Bench Question Score by 13.4 points and RHELM accuracy by 4.1 and 6.6 points without and with external data, respectively. The supervision also benefits two existing optimizers on ATM-Bench, while new-user experiments demonstrate continued design optimization using supervision generated from the target histories.
Abstract (translated)
URL
https://arxiv.org/abs/2610.39276
PDF
-
LLM-Assisted Optimization for Simulation-Based Redesign
2026-10-29 03:28:30Chan-Eui Song, Jeremy J. Michalek, Christopher McCombAbstract
Engineering design frequently involves redesigning a prior solution to adapt to changes in context, such as material prices or system requirements. Redesign often leverages simulators too complex and/or computationally intensive for traditional numerical solvers, which lack engineering knowledge and context, to ensure global solution quality. Large language models (LLMs) can interpret semantic information, including engineering knowledge about the system and changes in context, to help guide search, and recent studies have shown that incorporating this knowledge into optimization can improve search performance. However, how best to combine LLMs with numerical solvers in redesign remains unclear. We develop and test an integrated LLM-optimization approach for redesign and compare three methods for leveraging LLMs (warm start, variable fixing and ReAct) and four black-box solvers (Bayesian optimization, genetic algorithms, mesh adaptive direct search and random search) on an electric vehicle battery pack cost minimization case study across four types of redesign scenarios. We find that the integrated approach can match feasible design discovery of the numerical-solver baselines with 80% fewer simulator evaluations while tripling the fraction of runs that find near-optimal solutions. We find that the best LLM-solver integration approach and the way the prior solution is leveraged depend on the structure of the design requirement change, and that the benefit of increasing LLM model size grows with the extent to which the LLM controls the search. These results show clear potential for LLMs in redesign tasks for guiding or replacing numerical solvers.
Abstract (translated)
URL
https://arxiv.org/abs/2610.39275
PDF
-
Online Dense Reinforcement Learning with Planning-Grounding Agent for UI Navigation on Live IOS Apps
2026-10-29 03:17:33Peitong Duan, Khashayar Gatmiry, Jeffrey Nichols, Samy Bengio, Emmanuel AbbeAbstract
While frontier vision-language models demonstrate strong capabilities in user interface (UI) navigation, their high inference costs prohibit everyday use. Conversely, smaller models trained via supervised fine-tuning (SFT) or offline reinforcement learning often memorize static trajectories, leading to brittle execution and an inability to recover from errors. To address these limitations, we propose a novel online reinforcement learning framework that trains UI navigation agents directly on live iOS applications. Our framework decouples the agent into two specialized components: a Planning agent and a UI Grounding model. We rely on an LLM-as-a-Judge to provide comprehensive reward signals, combining dense, step-level feedback to evaluate intermediate progress with a terminal reward for overall task completion. To effectively leverage this dual reward structure, we optimize the agent using Proximal Policy Optimization (PPO) with step-level advantage estimation. Furthermore, we introduce GiGPPO, a novel neighbor-conditioned value estimation method that improves critic stability by leveraging reward signals from visually similar states across concurrent trajectories. To support this online training paradigm at scale, we introduce an efficient infrastructure that orchestrates hundreds of concurrent virtual iOS devices. Evaluations on rigorous benchmarks, including UINavBench, iOSWorld, and AndroidWorld, demonstrate that our online RL trained agents substantially outperform both SFT and terminal-only RL baselines while accelerating training convergence. Notably, our 8B RL agents achieve nearly a sevenfold improvement over the SFT baseline in out-of-distribution generalization to Android environments, while outperforming Gemini-3.5-Flash and tying with Gemini-3.1-Pro on real iOS applications.
Abstract (translated)
URL
https://arxiv.org/abs/2610.39264
PDF
-
Beyond LLM Wrappers: Context Engineering and the Three-Engine Architecture for Reliable Financial AI
2026-10-29 02:50:00Pradeep Kumar GS, Ravi TanniruAbstract
Financial AI systems must support decision-bearing questions involving payments, collections, cash flow, and compliance while ensuring numerical accuracy, traceability, and auditability. Large language models (LLMs) enable natural language interaction but cannot independently guarantee reproducible calculations or evidence-grounded recommendations. We present a production-grounded architecture integrating three components: a financial knowledge graph for enterprise relationships, temporal context, and provenance; a deterministic optimization engine for financial calculations and decisions; and a multi-agent LLM layer for intent interpretation, task orchestration, explanation, and claim validation. A coordination layer, termed context engineering, selects, ranks, composes, and validates query-specific context across ERP systems, tools, and tenants. Based on observations from 15 deployments over approximately 11 months, single-agent pipelines exhibited hallucination rates of about 7-9% (5-10% across query types), including invented invoices and unsupported financial claims. Graph-backed multi-agent validation reduced observed rates to below 1%, although these findings are observational rather than causal. ERP data quality and tenant-specific schemas remained major implementation challenges. Reusable mappings, graph templates, and validation policies coincided with a reduction in deployment time from roughly six to two weeks. We argue that financial context engineering extends beyond text-centric retrieval-augmented generation by requiring structured relationships, temporal validity, exact numerical evidence, access controls, and validation against source records and solver outputs. We outline research challenges in evaluation, reproducibility, privacy, context quality, and cross-tenant generalization.
Abstract (translated)
URL
https://arxiv.org/abs/2610.39230
PDF
-
Reasoning or Pattern Exploitation? Mechanistic Insights from RL-Trained Language Models
2026-10-29 02:09:38Aishik Nagar, Cheston TanAbstract
Large language models (LLMs) are increasingly described as acquiring "reasoning" skills after reinforcement learning with verifiable rewards. However, whether these gains reflect genuine reasoning or more superficial adaptations remains underexplored. In this work, we probe LLMs trained in a finite and exhaustively analyzable logical domain, namely Tic-Tac-Toe, and trace how internal representations evolve across reinforcement learning (RL) with Group Relative Policy Optimization (GRPO). Quantitatively, RL yields high accuracy and robustness across prompt variations, suggesting the emergence of game reasoning. Mechanistic interpretability, however, paints a different picture: features extracted with sparse autoencoders (SAEs) reveal that models primarily adapt to rely more heavily on information already explicit in the prompt, such as whose turn it is, game progression and board occupancy. In comparison, high-level concepts like board symmetries and strategic forks remain weakly represented, raising concerns that reasoning benchmarks overestimate model abilities. This suggests that RL-driven "reasoning" may conflate task-specific adaptations to input information with reasoning ability. We provide, for the first time: (i) a systematic interpretability pipeline tracing representation dynamics across RL training in small-scale LLMs; (ii) an extension of SAE-based feature discovery to hypothesis-driven testing in a finite logical domain; and (iii) an interpretability-based demonstration that RL post-training neither adds nor removes concept information but reorganizes which features dominate the representation, with causal interventions locating the decision downstream of it, in the late-layer state.
Abstract (translated)
URL
https://arxiv.org/abs/2610.39182
PDF
-
When Does Closed-Loop Curriculum Adaptation Help GRPO? An Analysis on Countdown
2026-10-28 23:35:56Yuv Raj Pant, Moshiur Rahman Tonmoy, Kaiqun FuAbstract
Reinforcement fine-tuning has become an effective post-training paradigm for improving the reasoning ability of large language models, but it remains sample-inefficient when the training distribution is mismatched to the model's current capability. In group-relative objectives such as Group Relative Policy Optimization (GRPO), problems that are too hard often produce uniformly failed rollouts, while problems that are too easy provide little useful advantage signal. In this work, we introduce CLOSEDLOOPGAUSSIAN, a lightweight adaptive curriculum method for efficient reinforcement fine-tuning on ordered reasoning tasks. Instead of following a fixed easy-to-hard schedule, CLOSEDLOOPGAUSSIAN preserves the structure of an Easy-to-Hard Gaussian sampler while dynamically adjusting its pace and spread using periodic probe estimates of per-difficulty accuracy. The controller accelerates training when the model is ahead of a reference curriculum, slows down when the model falls behind, and broadens sampling when earlier difficulties regress. This feedback mechanism keeps training focused on tasks that are challenging but still learnable, without changing the reward function, model architecture, or underlying GRPO optimizer. Experiments on symbolic arithmetic reasoning task show that CLOSEDLOOPGAUSSIAN improves exact-match accuracy over vanilla GRPO and fixed open-loop curricula, with the strongest gains in intermediate-difficulty regimes where group-relative learning signals are most informative. Codes are available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.39056
PDF
-
Rethinking GFlowNet Training: Negative Reinforce Dynamics
2026-10-28 22:18:48Pedro Dall'Antonia, Daniel Csillag, Diego MesquitaAbstract
Generative Flow Networks (GFlowNets) are models that learn to sample discrete objects proportional to a target reward. They are trained by minimizing balance errors over sampled trajectories, which inherently ties correctness to exploration: if a region is not visited, balance cannot be enforced there. Adaptive exploration is therefore crucial, yet standard training relies on a semi-gradient update that treats the sampling distribution as fixed at each optimization step. On the other hand, the full gradient reveals a second mechanism that reduces the observed errors in an anti-exploratory manner. In this paper, we establish a connection between the full gradient and min-min dynamics between the GFlowNet and a decoupled sampler, showing that the sampler reduces observed violations by avoiding high-error regions. This insight motivates a bridge between two-player games and single-policy dynamics, which underpins the design of our proposed method, Negative Reinforce Dynamics (NRD). By tying the parameters of an adversarial game together, we recover a single-policy dynamic that extends exploration while maintaining correctness. In standard benchmarks, our method outperforms existing approaches in both convergence speed and exploration efficiency, without training a separate teacher or auxiliary exploration policy.
Abstract (translated)
URL
https://arxiv.org/abs/2610.38994
PDF
-
On-Policy Self-Distillation with Teacher Evolution
2026-10-28 10:51:52Shufan Shen, Zhongni Hou, Chao Bi, Yufei Zhang, Guojun Yin, Shuhui WangAbstract
On-Policy Self-Distillation (OPSD), which aligns a student model's token-level distribution with that of its self-teacher prompted by privileged context, has emerged as a promising direction to mitigate reward sparsity in sequence-level post-training paradigms. However, OPSD often fails to achieve continuous performance gains during training. In this paper, we find that the model's limited self-teaching capability underlies this phenomenon by generating unreliable guidance on certain samples. To address this problem, we propose Teacher-Evolution Self-Distillation (TESD), which enables the self-teacher to generate rollouts and receive environmental feedback based on task correctness and context independence. First, using these rewards as training objectives, we perform teacher-oriented optimization to continuously enhance the model's self-teaching capability during distillation. Second, by estimating teacher guidance reliability from these rewards, we apply adaptive student-oriented optimization to down-weight samples with unreliable guidance. Extensive experiments across multiple datasets and model scales show that TESD consistently outperforms representative token-level and sequence-level methods, highlighting the necessity of improving supervision quality in OPSD.
Abstract (translated)
URL
https://arxiv.org/abs/2610.38637
PDF
-
Curriculum-Based Deep Reinforcement Learning for Robust Autonomous Rocket Landing under Atmospheric Uncertainty
2026-10-28 10:51:29Atin Chowdhury, Piyush JoshiAbstract
Autonomous far-range rocket landing is a challenging control problem because navigation, descent-rate regulation, attitude management, and fuel use must be coordinated in the presence of uncertain atmospheric and vehicle dynamics. This paper presents a curriculum-driven proximal policy optimization (PPO) framework for three-dimensional rocket landing in a stochastic environment incorporating altitude-dependent aerodynamic drag, wind shear, temporally correlated turbulence, discrete gusts, uncertain propulsion and aerodynamic parameters, thrust-vector misalignment, engine lag, actuator-gain variation, noisy observations, and command latency. Learning is organised into far-range approach, soft descent, and integrated landing stages. Policy transfer, critic warm-up after reward changes, rehearsal of earlier tasks, and progressive expansion of weather severity are used to support learning across these coupled phases. Across 100,000 evaluation episodes, the policy achieves 96,656 successful landings, corresponding to a success rate of 96.656\% and a 95\% Wilson confidence interval of 96.543--96.766\%. Success decreases from 98.84\% under zero-disturbance conditions to 88.07\% in the highest weather-severity band. Successful touchdowns exhibit small miss distances and low residual horizontal and vertical velocities. An exploratory logistic-regression analysis identifies mean wind speed as the strongest measured correlate of success, with initial distance, initial altitude, initial vertical velocity, and wind-shear exponent also associated with outcomes. The results quantify both the effectiveness of the learned policy over the evaluated disturbance distribution and the degradation that occurs under severe compound atmospheric conditions.
Abstract (translated)
URL
https://arxiv.org/abs/2610.38636
PDF
-
Co-Evolving Policy and Rubric via Capability Awareness for Open-Ended Generation
2026-10-28 10:00:30Chengsheng Zhang, Wenshen Xu, Yuchuan Wu, Zhiyuan Yu, Junfeng Fang, Tianlong Wang, Bingxue Qiu, Yongbin Li, Xiang WangAbstract
Rubric-based reinforcement learning (RL) offers interpretable, fine-grained supervision for aligning large language models (LLMs) on open-ended generation tasks. However, static rubric criteria may become saturated or remain beyond the policy's exploration, rendering the reward signals insufficiently discriminative. While recent efforts seek to guide sampling exploration or evolve rubrics, they fail to explicitly factor in how rubric difficulty should track the evolving policy's capabilities, resulting in a persistent mismatch between the optimization objective and the policy's current capability frontier. In this paper, we propose \textbf{Co}-Evolving Policy and Rubric via \textbf{C}apability \textbf{A}wareness (\textbf{CoCA}), a closed-loop framework that jointly adapts the rubric and policy through criterion-level diagnosis. Specifically, CoCA contrasts query-only and rubric-guided rollouts to characterize each criterion relative to the current policy, recalibrates criteria with mismatched difficulty, and identifies guidance-revealed capability gaps for frontier-based learning. Based on this insight, CoCA either expands rubric-conditioned frontiers through guided optimization or transfers qualified gains through on-policy self-distillation. As the policy updates, CoCA improves the capability diagnosis and enables more reliable distillation, thereby completing the co-evolution loop. Extensive experiments across diverse benchmarks and LLMs demonstrate that CoCA consistently improves performance over rubric-based RL methods. Moreover, CoCA promotes the internalization of guidance-revealed capabilities into query-only behavior while mitigating capability degradation.
Abstract (translated)
URL
https://arxiv.org/abs/2610.38566
PDF
-
SparseTokenGS: Gaussian-based Sparse-View CBCT Reconstruction via Reusable Token Pool
2026-10-28 06:57:38Zihan Deng, Zhenzhen Gong, Yujie Liu, Jiaxin Gong, Dingkun Liu, Kexian Xie, Yixing Huang, Peng Jin, Tianye Niu, Zhisheng Wang, Fenglin Liu, Hongming Shan, Lequan YuAbstract
Sparse-view cone-beam computed tomography is widely used for fast, low-dose volumetric imaging and lesion localization in dentistry and image-guided interventions, but it introduces severe artifacts. Feed-forward Gaussian methods transfer the reconstruction cost to training and decode each coordinate from projection-aligned features and neighboring anchors. However, information from the whole scan is written only when the field is constructed and cannot be retrieved during coordinate decoding, leaving the volume inconsistent within the scan. This is a common limitation of such methods: once the field is fixed, decoding uses only local features and reconstruction quality drops substantially. We propose SparseTokenGS, a Gaussian-based method for sparse-view cone-beam computed tomography reconstruction via a reusable token pool. Anchor Token Fusion and Point-Token Attention query this shared pool at the anchors and at the query coordinates, respectively. The local Gaussian prior is retained, and the volume is reconstructed in one forward pass, without per-scan optimization or a projection-domain loss. On three public chest and dental datasets, SparseTokenGS achieves the highest reconstruction performance among the compared methods. On clinical dental scans, reconstruction from only six projections reaches 29.51 decibels in peak signal-to-noise ratio and a structural similarity of 0.842 against the full-view reference, and off-the-shelf tooth-and-pulp segmentation keeps these structures separable for root-canal localization. Experiments on volcanic rock and chicken show performance after same-object adaptation. A reusable token pool gives feed-forward reconstruction a mechanism that keeps scan-wide context available through decoding and can benefit future rapid clinical lesion localization and analysis. The code is available at: this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.38360
PDF
-
Encode Once, Answer Many: Internalizing Context via Parametric Flow
2026-10-28 06:55:19Xinhao Zhong, Mengxuan Wu, Chonghua Liao, Linxi Xie, Ruicheng Feng, Yanpei Guo, Hao Tan, Jiaxin Hong, Jason Li, Panpan Zhang, Ziheng Qin, Qinglin Lu, Yuzhang Shang, Tongliang Liu, Damian Borth, Haggai Maron, Yongxiang Hua, Zhixiang Wei, Jian Hu, Yunfei Wu, Xing Sun, Xin Li, Haoyu Cao, Zhangyang Wang, Kai WangAbstract
Large language models are typically adapted to a task either by placing knowledge in the input context, which keeps it external and re-encodes it on every call, or by fine-tuning, which internalizes it but demands a costly round of optimization per task. We aim to internalize the knowledge carried by a context directly into model parameters, without per-task optimization and without keeping the context in the input at inference time. To this end, we propose CLIK, a context-conditioned parameter generator that maps a given context to LoRA parameters for internalizing knowledge in a single forward pass through a Transformer encoder with cross-attention, requiring no test-time gradient updates. Once the generated LoRA is inserted into a frozen base model, the model answers downstream queries from the query alone, having internalized the context into its weights rather than re-reading it. Because the adapter depends only on the context, it is generated once and reused for every query on the same document, so the cost of reading that document is amortized across queries. On document question answering, CLIK substantially improves over the no-context base model and outperforms existing hypernetwork-based generators, while remaining behind a base model that reads the context directly, and it carries over to multi-turn dialogue without any multi-turn-specific training. Analysis further shows that the context-dependent part of the generated adapter concentrates in a compact and specific substructure.
Abstract (translated)
URL
https://arxiv.org/abs/2610.38356
PDF
-
Aligning Misaligned Benchmarks for Single-View Novel View Synthesis
2026-10-28 06:49:44Yang-Che Tseng, Erqun Dong, Jeong-gi Kwak, Gopal Sharma, Shakiba Kheradmand, Stephan Garbin, Hossam Isack, Kwang Moo YiAbstract
While single-view novel view synthesis methods have been dramatically enhanced recently, how to quantify their performance has yet to be standardized. In this work, we demonstrate that an important factor lies in how each method's own reference frame is aligned with the so-called ground-truth pose. Existing methods often rely on test-time optimization of target poses for this alignment, which inherently is ill-posed due to the complexity of the optimization landscape of camera poses vs render accuracy, leading to methods being evaluated at an arbitrary local optimum. Some recognize this limitation and thus rely on 3D reconstructions of entire scenes using training and test views to align poses, which may be infeasible. In this work, instead, we propose to use traditional 2D-to-2D and 2D-to-3D image matching techniques to align 3D reconstructions. We also evaluate only within co-visible regions, between different views, which is the only place where a `correct' ground truth can be defined. Our evaluations show that with proper coordinate alignment, novel-view synthesis methods that use explicit 3D Gaussians achieve the state of the art, despite being previously reported in the literature to be outperformed by pure image-based 2D methods. We further provide a simple baseline that directly utilizes an off-the-shelf video generator and a feed-forward geometry estimator. Our simple baseline is already comparable to state-of-the-art methods specifically trained for novel-view synthesis, without any task-specific training.
Abstract (translated)
URL
https://arxiv.org/abs/2610.38349
PDF
-
MUSE: Multi-Level Memory with Selective Runtime Injection for Scientific Coding Agents
2026-10-28 06:32:51Haoran Jin, Defu Lian, Xiting Wang, Hong XieAbstract
Historical experience can guide scientific coding at multiple levels, but task-level retrieval may miss implementation lessons that become relevant during execution. We introduce MUSE, a memory management framework combining multi-level memory with selective runtime injection for scientific coding agents. MUSE builds global, approach-level, and implementation-level memory from historical trajectories. At initialization, the agent receives global memory and problem-matched approach-level memory. During execution, current interactions are used to retrieve implementation-level memory. An LLM gate selects relevant experience and generates specific review advice, or skips injection. Across four benchmarks spanning numerical problem solving, kernel optimization, and scientific software repair, MUSE outperforms retrieval and memory baselines on all evaluation metrics. Ablations support the multi-level design and gated injection, while transfer experiments show that memory remains useful across kernel benchmarks. These results show that multi-level memory and selective runtime injection enable effective reuse of experience in scientific coding. Code is available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.38332
PDF
-
A Case Study of a Multi-Agent Kernel-Optimization Loop That Learns a New AI Accelerator
2026-10-28 06:29:02Byonggon ChunAbstract
Frontier large language models write correct CUDA kernels for most single-operator benchmark tasks, yet their accuracy collapses on accelerators whose architecture and programming language they have rarely seen. We present a case study of an LLM agent loop built for such accelerators. Our example is the FuriosaAI RNGD NPU, programmed in a recent, sparsely documented Rust DSL, where the loop optimized three Gemma-4-12B decode kernels over 12 days on a shared, scarce device. The loop builds its own target knowledge: it writes and checks a summary of the DSL documentation, measures the chip with microbenchmarks, records compiler-derived legality rules, and revisits this knowledge when the toolchain changes, rechecking rules and measurements where it can. A role-based search over per-kernel Design Trees explores whole-kernel designs and accumulates small accepted changes in the best version of each kernel. Because device time is scarce and readings are noisy, cheap checks screen new designs before any device run, and late in the run a change was accepted only after repeated comparisons with the best kernel on the device. The final kernels reach a 9.293x geomean speedup over the official baseline, each within 1.25x of the cycles needed just to move its weights. Our prompts, profiler and acceptance test are available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.38329
PDF
