Optimization
-
Expected Return Causes Outcome-Level Mode Collapse in Reinforcement Learning and How to Fix It with Inverse Probability Scaling
2026-10-20 03:12:50Abhijeet Sinha, Sundari Elango, Dianbo LiuAbstract
Many reinforcement learning (RL) problems admit multiple terminal solutions of comparable quality, where the goal is not to identify a single optimum but to represent a diverse set of high-quality outcomes. Nevertheless, policies trained by standard expected return maximization routinely collapse onto a small subset of outcomes, a phenomenon commonly attributed to insufficient exploration or weak regularization. We show that this explanation is incomplete: outcome level mode collapse is a structural consequence of the expected-return objective itself. Under idealized learning dynamics, the log-probability ratio between any two outcomes evolves linearly in their reward difference, implying exponential ratio divergence and inevitable collapse independent of the exploration strategy, entropy regularization, or optimization algorithm. We identify the source of this pathology as the probability multiplier inside the expectation and propose a minimal correction: inverse probability scaling, which removes outcome-frequency amplification from the learning signal, fundamentally changes the learning dynamics, and provably yields reward-proportional terminal distributions, preventing collapse in multimodal settings. We instantiate this principle in Group Relative Policy Optimization (GRPO) as a drop-in modification, IPS-GRPO, requiring no auxiliary models or architectural changes. Across different reasoning and molecular generation tasks, IPS-GRPO consistently reduces outcome-level mode collapse while matching or exceeding baseline performance, suggesting that correcting the objective rather than adding exploration heuristics is key to reliable multimodal policy optimization.
Abstract (translated)
URL
https://arxiv.org/abs/2601.21669
PDF
-
TorchTPU - A Hybrid Eager/Compiled Architecture for Native PyTorch Execution on TPU
2026-10-20 02:09:40Claudio Basile, Michael Voznesensky, Gunhyun Park, Ben Wilson, Parker Holloway, Zhanyong Wan, Vlad Belous, Manish Khanna, Paul Ganssle, Kevin Gleason, Andrew Dame, Brendan Folie, Bhavya Bahl, Haifeng Jin, Pedro Goncalves Mokarzel, Arian Arfaian, Luke Boyer, Chi Zhang, Yiling Yuan, Adam Mainz, Aditya Patel, Huangze Tang, Jay Gan, Lizzie Liang, Will Froom, Tushar Agarwal, Karlo Basioli, Theotime Combes, Emilio Cota, Minos Papadopoulos, Tudor Pipernea, Yarong Mu, Qi Zhou, Yaoshiang Ho, Aleksey Vlasenko, Alexander Brauckmann, Ziyin Huang, Jack Oh, Jerry Hsu, Tanvi Bajpai, Jialei Chen, Wenjun Gu, Chintan Shah, Chris Chan, Sayce Falk, Chris Jones, Rob Torres, Kat Ko, Lee Howes, Freya Azad, Andi Gavrilescu, Joe Pamer, Bill JiaAbstract
TorchTPU is an open-source collaboration between Google and Meta that introduces a native and principled PyTorch backend for Google TPUs. It enables seamless out-of-the-box model execution and TPU migration while delivering industry-competitive inference and training performance through eager execution optimizations, this http URL integration, and native hardware acceleration.
Abstract (translated)
URL
https://arxiv.org/abs/2610.24161
PDF
-
How Data Scales in Agentic Reinforcement Learning: Laws and Synthesis Strategies
2026-10-20 01:59:17Bowei He, Yankai Chen, Xiaokun Zhang, Changjiang Han, Ye Yuan, Chong Li, Xue LiuAbstract
Pretraining scaling laws treat training data as a one-dimensional quantity: a token count. Agentic reinforcement learning (RL) inherits this scalar abstraction, but modern synthesis pipelines now generate \textit{tasks}, \textit{environments}, and \textit{trajectories} independently and at very different costs, turning data into a structured design space. This raises a question pretraining never had to answer: given a synthesis budget, \textit{which axis} should be scaled? We address this question end-to-end with three tightly coupled contributions. First, we decompose agentic data into three cardinality axes (task, environment, trajectory), plus a reward-density control variable, and fit per-axis scaling relations on different large language models (LLMs) across math+code, SWE, web, and UI/tool application domains. Across all model sizes, model families, domains, and RL algorithms (PPO and GRPO) we measured, the scaling exponents order as environment $>$ task $>$ trajectory. When environment diversity is decomposed into world and environment-structure diversity, its advantage is mainly driven by the former, while the latter is just comparable to the task diversity. Second, we use the fitted laws as a common ruler to price synthesis methods, capturing each method's efficiency as a per-axis discount factor and identifying critical synthetic-to-real ratios at which collapse begins; these ratios differ by an order of magnitude across axes. Third, we cast budget allocation as a constrained optimization under our laws and discounts, deriving a Chinchilla-style compute-optimal recipe and validating it at held-out scale. The empirical results demonstrate that following the recipe yields up to $1.7\times$ compute efficiency over the dominant ``scale-the-trajectories'' practice.
Abstract (translated)
URL
https://arxiv.org/abs/2610.23946
PDF
-
Explore More, Validate Less: Efficient Reflective Prompt Optimization
Abstract
Reflective prompt optimization improves instructions using task feedback, but coupling candidate generation with full-validation-based selection can make search costly and sequential. We propose a two-stage framework that decouples reflective candidate-pool construction from validation allocation. During construction, we sample multiple parents and reuse each parent's training responses across parallel reflection branches. Each revision is screened against its parent on a small training minibatch, allowing locally improved candidates to support further revisions without full validation. After construction, an exploration policy selects candidates from the frozen pool under a fixed full-validation budget, and the prompt with the highest measured validation score is returned. This separation allows candidate-pool size and validation budget to be set independently. We evaluate random selection and learned score prediction with similarity-based exploration on five benchmarks using Qwen3.5-9B and Llama-3.1-8B-Instruct. With 100 retained candidates and 20 full validations, our framework improves over the initial solver by up to 11.22 percentage points. Learned selection improves over the initial solver in all ten comparisons, although neither selection policy consistently outperforms the other. These results support separating reflective search from budgeted validation while leaving room for stronger candidate-selection policies.
Abstract (translated)
URL
https://arxiv.org/abs/2610.27367
PDF
-
When Does an Added Planning Capability Create Operational Value? An Observable Evidence Standard
2026-10-19 21:21:20Yihua Xu, Yan Lyu, Jize Li, Jiani He, You Lin, Yidan Hu, Rubing QiangAbstract
Industrial artificial intelligence (AI) planning systems combine predictive models, optimization logic, controls, and governance rules, but improvement in the complete policy does not show whether one addition made an independent contribution or whether that contribution justifies its burden. We address this attribution problem with an observable evidence standard that separates opportunity to act, executed-plan change, operational consequence, equivalence, uncertainty, and operating scope. Using forecast-to-plan replay on Favorita, M5, and Walmart data, we apply the standard to heterogeneous planning additions. Three limits of single-indicator evaluation emerge. First, a composite policy can conceal negligible component contribution: one control changed only 15 of 681 executed plans. Second, changing execution does not guarantee stable benefit: the evaluated multi-period capability helped in several long-lead settings but was 7.6% more costly under tight capacity. Third, indirect evidence can prioritize review but cannot complete attribution. Independent runs produced different paths in 19 of 599 cases that a rule-based check predicted would be unchanged. Although a lower-cost diagnostic matched the full-evaluation winner in only 55.0% of cases, it selected an option within the $10^{-6}$ cost-equivalence tolerance in 79.5%; the remaining cases prevented the proxy from authorizing acceptance. These experiments establish capability-level contribution evidence; the organization completes the value decision by weighing it against implementation and sustaining burden. The resulting capability evidence record links the observations to validation, deployment scope, fallback, monitoring, redesign, and retirement decisions.
Abstract (translated)
URL
https://arxiv.org/abs/2610.27305
PDF
-
Vulnerability Analysis for Safe Multi-Agent Reinforcement
2026-10-19 21:09:02Taozheng Zhu, Shixiong Jiang, Fanxin KongAbstract
Safe Multi-Agent Reinforcement Learning (SMARL) aims to learn cooperative control policies under prescribed safety constraints. Al- though existing SMARL methods focus on enforcing safety during policy optimization, their vulnerability remains largely unexplored. This paper investigates inference-time observation attacks against SMARL, where an adversary perturbs the observations of a subset of agents, thereby altering the actions produced by their policies. We propose MinMax Attack, an attack framework that jointly selects adversarial target actions for compromised agents while accounting for the Cooperative Resistance of benign agents. We further derive a bound on the induced safety cost and characterize its relation- ship with the Compromised-Agent Budget and the Perturbation Budget. Experiments across multiple safety-critical multi-agent benchmarks and representative state-of-the-art SMARL algorithms show that MinMax Attack induces substantially higher safety costs than existing attack baselines while preserving task reward, thereby reducing its detectability and achieving both attack effectiveness and stealthiness.
Abstract (translated)
URL
https://arxiv.org/abs/2610.27293
PDF
-
XoNav: An Exoskeleton as a Personalized Haptic Interface for Assisted Navigation
2026-10-19 20:44:35Debadrata Sarkar, Soumen Sen, Aman AroraAbstract
Walking without vision requires timely, unambiguous guidance, yet many wearable aids encode directions as audio or vibrotactile signals that users must translate into movement. We present XoNav, a four-actuator hip exoskeleton that renders guidance cues directly at the hip joints. It renders an Advance cue and two-level Braking cues through sagittal-plane actuators and two-level Steering cues through frontal-plane actuators. A bilateral LSTM-based gait-state classifier uses two thigh-mounted IMUs to identify gait states, enabling haptic cues to be delivered in coordination with the wearer's gait while suppressing them during standing. The classifier generalized to unseen users with recalls of 95.2-97.9%, with mean detection delays of 38.6 and 63.1 ms for stance-to-swing and swing-to-stance transitions, respectively. Constrained human-in-the-loop Bayesian optimization personalized Braking for minimum resistance and Steering for maximum comfort while ensuring detectability and safety. All participants reliably distinguished between grades of their personalized cues and the firmest Braking cue remained below 53% of the safety limit. In blindfolded Wizard-of-Oz navigation trials along three paths, the participants kept 98.8% of the walked distance within a 0.75 m wide passage, with a 4.9% increase in mean progression speed per trial. A body-worn camera, RTAB-Map and Nav2 further enabled a feasibility study of autonomous navigation and detours around previously unmapped obstacles.
Abstract (translated)
URL
https://arxiv.org/abs/2610.27271
PDF
-
Auditing Safety Preference Supervision with Recorded Derivations
Abstract
Preference labels align model training, but the judgments behind them are difficult to inspect or verify. In this paper, we examine how preference construction can be checked when intermediate decisions are recorded, and which of them remain uncertain. We present Symbolic Preference Labeling, a method in which a pretrained language model classifies predefined attributes of prompts and responses. Explicit rules assign response verdicts, and a fixed verdict ranking determines the preference label. Each preference label has a recorded derivation: the response verdicts and preference label can be recomputed from the stored attribute assignments, rules, and verdict ranking without rerunning the language model. On 200 prompts annotated by human raters, the language model produces valid structured outputs for 192 prompts. All eight parsing failures occur in the subset of self-harm prompts marked as harmful. Using 465 preference pairs, we train one policy model with direct preference optimization across five training seeds. Mean measured benign refusal is 40.82 percent on 1,319 benign prompts that resemble unsafe ones, and mean measured toxic refusal is 84.98 percent on 655 toxic prompts. This study distinguishes the reproducibility of preference labels from attribute correctness. The recorded derivations make preference construction open to inspection, while attribute correctness and semantic safety require separate validation.
Abstract (translated)
URL
https://arxiv.org/abs/2610.27266
PDF
-
CEFR-oriented Russian Text Simplification Using Reinforcement Learning
2026-10-19 20:04:47Kristina Mustakova, Alexey SorokinAbstract
We present a study on controlled automatic simplification of Russian texts using preference-based optimization algorithms with a focus on adapting texts to the A1, A2 and B1 proficiency levels. We fine-tune YandexGPT using Rejection Sampling and Direct Preference Optimization (DPO) methods to improve control over the complexity of the generated text. We compare several training configurations that differ in dataset composition, text type and the strategy used to construct preference pairs for DPO. For automatic evaluation we train a regression model based on ruRoBERTa-large to predict continuous complexity scores corresponding to CEFR levels. We also consider semantic similarity, lexical coverage and quantitative readability metrics. Human evaluation assesses simplicity, fluency and adequacy. Our best model achieves 87.15\% target level accuracy, an MAE of 0.35 CEFR levels and a target interval MAE of 0.11. Manual assessment shows that 75% of the evaluated simplifications match the target morphological and syntactic level, and fluency and adequacy scores are especially high for A2 and B1. Although A1 remains the most challenging level, the model significantly reduces the proportion of vocabulary and grammatical constructions beyond the target level while generally preserving meaning. Our model is freely available on $\href{this https URL}{HuggingFace}$.
Abstract (translated)
URL
https://arxiv.org/abs/2610.27237
PDF
-
EDISCO: Equivariant DIScrete Diffusion for Euclidean Combinatorial Optimization
2026-10-19 18:44:19Ruogu Chen, Jie HanAbstract
Euclidean combinatorial optimization problems (ECOPs), such as the Traveling Salesman Problem (TSP) and Capacitated Vehicle Routing Problem (CVRP), possess inherent symmetries under the two-dimensional Euclidean group E(2), including rotations, reflections, and translations. Existing learning-based methods, including recent diffusion-based methods, rely on data augmentation or regularization to approximate E(2)-equivariance. This paper presents EDISCO, the first discrete diffusion model for ECOPs with exact E(2)-invariant generative distributions over node-index solutions. EDISCO introduces an E(2)-equivariant edge-score network coupled with a categorical continuous-time Markov chain over discrete edge variables, and exact posterior sampling provides efficient multi-step inference. This design gives EDISCO a local geometric inductive bias: edge neighborhoods with the same relative geometry and combinatorial context are represented consistently regardless of absolute position or orientation, making learning more efficient and inference more robust than non-equivariant methods. EDISCO outperforms previous learning-based state-of-the-art solvers on synthetic TSP from 100 to 10000 nodes and CVRP from 50 to 2000 customers, while using only 33-50% of the training instances. Trained only on uniform synthetic data, EDISCO also outperforms competing learning-based baselines under spatial distribution shift and CVRP constraint-tightness shift. Code is available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.04953
PDF
-
Teaching AI Agent to Play Poker Strategically Through Preference Optimization
2026-10-19 15:33:17Phurinat Polasa, Thanapon Thanadunpremdet, Phatthanasak Kraiduang, Saksorn RuangtanusakAbstract
A poker agent can make a legal move and still make a strategically poor one. We ask whether compact AI agents can exhibit strategic behavior under hidden information and whether preference optimization can improve beyond static and zero-shot policies. We first evaluate a static baseline and three frozen language models: SmallThinker, Qwen3, and Gemma-3 whose performance varies substantially between exact-game and six-player evaluation. We therefore propose Strategic Preference Optimization with KL-anchored Equilibrium (SPOKE), combining teacher-relative value penalties, policy anchoring, risk constraints, and structured rationale supervision, and compare it with SFT, DPO, GRPO, and GSPO. Across three runs of a SmallThinker-4B-A0.6B policy, SPOKE achieves the lowest mean exploitability in Kuhn (0.213) and Leduc (1.081) and the highest six-max top-1 agreement (0.775), while maintaining teacher KL comparable to SFT (0.522 vs 0.523). These results provide preliminary evidence that strategic preference optimization can improve strategic quality while preserving teacher-consistent behavior.
Abstract (translated)
URL
https://arxiv.org/abs/2610.27028
PDF
-
How to get the best and fairest performance out of your ASR system using activation steering
2026-10-19 07:53:13Felix Herron, Alex Colagrande, Solange Rossato, Fran\c{c}ois Portet, Alexandre Allauzen, Chuyuan LiAbstract
Activation steering for ASR systems is an underexplored technique in improving both ASR performance and demographic fairness. We investigate two forms of activation steering: paired activation subtraction, and supervised optimization. We find the former to be ineffective at ameliorating ASR performance whatsoever, while the latter is quite effective at deriving steering vectors (SVs) that improve at once overall ASR performance and demographic fairness along such diverse axes as gender, age, and dialect. We also explore what types of data---from the most to the least personal---are necessary for calculating or training SVs. First, we show significant performance in performance from training SVs using an eclectic group of speakers from the same dataset. Furthermore, we find that vectors trained using acoustically similar speakers leads to ASR performances that are competitive or superior to vectors trained on the same speaker, or speakers from the same labeled demographic group. This finding shows that we can harness activation steering for few-shot adaptation given access to a speaker's demography, or zero-shot adaptation to unseen speakers without the need for any labels at all.
Abstract (translated)
URL
https://arxiv.org/abs/2610.26994
PDF
-
CoASFL: Coupled Consistency Optimization for Asynchronous Split Federated Learning
2026-10-19 07:37:00Junhao Cheng, Yunming Liao, Hongli Xu, Yang Xu, Zuan XieAbstract
Split Federated Learning (SFL) has emerged as a promising paradigm for training large-scale deep models on resource-constrained edge devices. Although asynchronous SFL alleviates the straggler bottleneck caused by system heterogeneity, it inevitably introduces two coupled dimensions of inconsistency: horizontal inconsistency across clients and vertical inconsistency between clients and the server. These inconsistencies, further exacerbated by statistical heterogeneity, result in biased activation representations, model divergence, and severe client drift. To address these challenges, we propose CoASFL, an inconsistency-aware asynchronous SFL framework that incorporates a dynamic activation merging mechanism and a consistency-aware model aggregation mechanism. Specifically, CoASFL constructs more consistent activation sequences from heterogeneous clients to alleviate horizontal inconsistency, and dynamically adjusts aggregation weights according to client participation frequencies and progress gaps to mitigate vertical inconsistency. By jointly coordinating activation-level and model-level consistency, CoASFL effectively suppresses inconsistency propagation during asynchronous training. Extensive experiments on an 80-device heterogeneous edge testbed demonstrate that CoASFL consistently outperforms state-of-the-art SFL approaches, accelerating training by 1.61$\times$ to 3.74$\times$ and improving final model accuracy by up to 3.26\% under heterogeneous scenarios.
Abstract (translated)
URL
https://arxiv.org/abs/2610.26992
PDF
-
Think Before You Transcribe: Context-Steered Reasoning for Hotword-Aware Speech Recognition
2026-10-19 05:09:28Qihao Wang, Hang Su, Fan Cui, Lichun Fan, Jian Luan, Yue HuAbstract
Contextual automatic speech recognition (ASR) improves rare-entity recognition by supplying candidate hotwords, but retrieved candidates may be incomplete or incorrect. A reliable system must therefore exploit a relevant candidate without allowing unsupported alternatives to override the acoustic evidence. We propose a context-steered reasoning framework that generates a short, audio-grounded caption before transcription. This task-specific representation lets the model assess candidate relevance before making a lexical decision. Supervised fine-tuning establishes the structured caption-to-transcription behavior, while targeted trajectory injection strengthens group relative policy optimization when difficult utterances yield only unsuccessful rollouts. Experiments across diverse context conditions show that the proposed method consistently improves both transcription quality and hotword recognition while remaining robust to irrelevant candidates.
Abstract (translated)
URL
https://arxiv.org/abs/2610.26986
PDF
-
Deep Auxiliary Learning via Bilevel Feature Optimization for Label-Efficient Medical Image Segmentation
2026-10-18 10:30:25Fuping Wu, Xi Jia, Le Zhang, Jiahua Li, Rima Mustafa, Thomas E. Nichols, Bartlomiej W. PapiezAbstract
Auxiliary learning (AL), which can improve the performance of a primary task by leveraging related auxiliary tasks, is commonly regarded as a special case of multi-task learning. Existing AL approaches typically require manually annotated auxiliary labels, increasing annotation costs, particularly in medical imaging where expert labeling is expensive. In this work, we investigate whether segmentation performance can be improved using auxiliary tasks whose labels are automatically derived from existing segmentation annotations, requiring no additional manual labeling. To address this, we propose a novel auxiliary learning framework, formulated as a bilevel optimization problem. Our method enables effective knowledge transfer between tasks via a feature communication module, enhancing the primary segmentation task. We evaluate the method on three medical image segmentation benchmarks under limited-label settings. The results demonstrate consistent improvements over state-of-the-art methods, achieving superior label efficiency. The code is publicly available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.26954
PDF
-
GRaSP: Latency-Aware Optimization for Geo-Distributed Retrieval-Augmented Generation
Abstract
Retrieval-Augmented Generation (RAG) is increasingly served across geo-distributed cloud and edge infrastructures, where wide-area latency, cross-region egress cost, and data-residency rules compete with retrieval quality. Conventional pipelines rank by semantic similarity alone and ignore where content lives, inflating tail latency and cost. We formalize the Geo-RAG Selection Problem (GRaSP), a drop-in re-scoring layer that decays relevance by latency and adds an optional cost penalty under a cardinality budget and per-region caps, with no retraining. We first settle its complexity: the base formulation is a linear objective over a partition matroid, solved optimally by an O(N log N) greedy algorithm, rather than NP-hard as a knapsack framing suggests, with hardness entering only under an additive egress budget. On a five-node CloudLab testbed emulating a four-region WAN over the Natural Questions corpus (2.68M passages), GRaSP's benefit is placement-dependent: under uniform placement it cuts fetch latency at negligible quality cost; under topical (residency-like) placement it traces a recall-latency Pareto frontier that dominates relevance-only, nearest-k, and a latency-budgeted baseline. At a moderate latency setting, GRaSP lowers P90 fetch latency by 41% while retaining 78% of relevance-only's recall, with sub-millisecond selection overhead, stable to 200 queries/s. Findings generalize across four public BEIR benchmarks. In end-to-end question answering, GRaSP reduces context-fetch latency by 43% while remaining within 0.062 absolute EM of relevance-only retrieval.
Abstract (translated)
URL
https://arxiv.org/abs/2610.26938
PDF
-
Expert-Guided Rubrics as Dense Reward Signals for Reinforcement Learning in Non-Verifiable Domains
2026-10-17 19:01:11Ashish Bansal, Hossam Amer, Keerat Kaur Guliani, Parinita Edke, Lesley Matheson, Leizer TeranAbstract
Large language models (LLMs) are increasingly deployed as autonomous agents in non-verifiable domains such as financial services, where they must interpret user intent, rewrite queries, and retrieve information under domain-specific constraints. Although reinforcement learning (RL) can train such agents from the consequences of their actions, recent successes largely rely on verifiable tasks with automatically computed binary rewards. In non-verifiable retrieval tasks, assessing query quality requires scarce subject-matter expert (SME) judgment that is inherently graded rather than binary. With Group Relative Policy Optimization (GRPO), sparse rewards yield no useful learning signal when all sampled outputs within a group receive the same outcome, causing optimization to stall. We introduce a method that converts a small, fixed investment of SME judgment into a dense reward and combines it with retrieval-based metrics to form a hybrid, GRPO-compatible training signal. Experiments show improvements of up to 6\% in Recall@3 and 6.7\% in MRR@3 compared with sparse-reward baselines.
Abstract (translated)
URL
https://arxiv.org/abs/2610.26932
PDF
-
AtomSD: Procedural Atom-Guided Self-Distillation for LLM Agents
2026-10-17 14:12:03Bohan Yu, Pengfei Cao, Kai Tian, Chen Han, Shi-Yang Li, Chenxi Zhou, Junlin Liu, Cao Liu, Ke Zeng, Jun Zhao, Kang LiuAbstract
Large language models (LLMs) possess broad factual knowledge and strong reasoning capabilities, but solving complex tasks also requires procedural knowledge about how to act. Existing approaches often represent procedural knowledge through trajectories or bundled skills, limiting flexible reuse across interaction histories. This paper introduces Procedural Atoms, fine-grained and independently applicable units that represent Intent, When, Do, and Avoid guidance. Building on this representation, we propose AtomSD, an Atom-guided Self-Distillation framework in which the learner continually extracts Procedural Atoms from its own successful trajectories and distills them into the evolving policy through Atom-conditioned supervision and task-reward optimization. Comprehensive experiments show that AtomSD consistently outperforms strong reward-based and distillation methods, exceeding the state-of-the-art baseline by up to 9.31 points in average performance. Further analyses demonstrate that the proposed Procedural Atoms provide effective guidance, highlighting the value of atom-level procedural knowledge for policy learning.
Abstract (translated)
URL
https://arxiv.org/abs/2610.26926
PDF
-
Steerable Pluralistic Policy Alignment from Anonymized Crowd Feedback
2026-10-17 06:14:59Akansha Kalra, Ellen R. Novoseller, Vernon J. Lawhern, Daniel S. BrownAbstract
Pluralistic alignment requires AI systems to learn from diverse populations whose preferences may reflect heterogeneous or even conflicting objectives, and to personalize their behavior to individual users. Existing approaches in pluralistic alignment often assume access to annotator identities and rely on powerful pretrained representations to model individual preference variations. We propose Preference-PCA Pluralistic Learning (P3L), a novel framework for pluralistic alignment that jointly learns trajectory embeddings and the shared basis of preference directions via PCA decomposition from heterogeneous crowd feedback without requiring annotator identities. At test time, our approach enables rapid and accurate personalized reward learning and downstream policy optimization. Experiments across multiple embodied control tasks demonstrate that P3L enables steerable personal alignment that generalizes beyond the offline dataset to out-of-distribution new users and attains competitive RL and test performance, while substantially reducing computational cost and supporting variable adaptation query budgets.
Abstract (translated)
URL
https://arxiv.org/abs/2610.26914
PDF
-
Uncertainty Gates the Gradient: Adaptive Updates for Generalized Few-Shot 3D Segmentation
2026-10-17 04:05:32Yifei Zhao, Fanyu Zhao, Shengtang Wu, Han Zhang, Ruike Cao, Yinsheng LiAbstract
In generalized few-shot 3D point cloud segmentation, support annotations are often compromised by sparsity and occlusion. Blindly adapting to such low quality samples risks learning spurious geometric patterns rather than robust novel concepts. Although 3D Vision-Language Models have been introduced to sup plement supervision, the potential of uncertainty to actively govern adaptation dynamics remains underexplored. Existing signal-level methods like loss reweight ing fail to prevent unreliable episodes from steering the optimization trajectory. To address this, we propose Uncertainty-Gated Adaptive (UGA), to the best of our knowledge the first framework that elevates uncertainty to update-level gating for generalized few-shot 3D point cloud segmentation. The core mod ule, Uncertainty-Gated Update (UG-Update), estimates episodic reliability via intra-class dispersion over merged support and pseudo-labeled novel supervision, and selectively gates residual gradient updates. This encourages the model to learn novel semantics from reliable episodes while attenuating updates during unreliable ones. Complementing this, we introduce an Uncertainty-Based Pseudo Label Filter (UB-Filter) to enforce learning from high-confidence dense signals. On ScanNet200 and ScanNet++ (1-/5-shot), UGA sets new state of the art with an average +3.08 HM IoU over the strongest prior baseline.
Abstract (translated)
URL
https://arxiv.org/abs/2610.26909
PDF
