Searcharxiv⌕ Search

arXiv subjects

Gao Huang

Publications and source records attributed to Gao Huang.

At least 19 recordsLinked to original sources
arXiv2026

Position Anchor Tuning: Towards Efficient Adaptation of Pre-Trained Point Cloud Transformers

Parameter-efficient fine-tuning (PEFT) has recently emerged as a pivotal research direction for adapting pre-trained point cloud transformers to diverse downstream tasks. Although existing methods achieve excellent fine-tuning performance with high parameter efficiency, they ignore inference efficiency. To tackle this problem, a novel PEFT method termed position anchor tuning (PAT) is proposed in this paper. As multi-head attention (MHA) and feed-forward network (FFN) are computation-heavy blocks in pre-trained transformers, PAT decreases their computational cost through token aggregation-expansion pairs. Each pair comprises a token aggregation module (TAM) and a token expansion module (TEM). For MHA and FFN blocks, TAMs extract representative tokens from their input tokens based on position anchors in 3D space. These extracted tokens, rather than the original input tokens, are processed by the blocks, thereby reducing the number of tokens involved in computation. Then, TEMs propagate the learned representations back to the original input tokens. Since TAMs are solely responsible for capturing task-specific representations, base-sharing low-rank adaptation (BSLoRA) is further introduced to enable them to learn such representations effectively with only a small number of trainable parameters. Extensive experiments on widely used benchmarks demonstrate that PAT performs comparably to state-of-the-art methods while incurring significantly lower computational overhead and fewer trainable parameters.

cs.CV↗
arXiv2026

World Models for Embodied Intelligence: From Plausible to Controllable to Actionable

World models connect perception and decision-making in embodied intelligence by maintaining hidden state, anticipating consequences, comparing interventions, and adapting when execution departs from expectations. Although progress is often measured by visual fidelity, their value lies in improving behavior. Before reaching for a cup, a person anticipates its weight and resistance to grasping, shaping the hand before contact. Such anticipation is coarse and rarely pictorial, yet it guides action. This raises a central question: which predictive capabilities improve behavior? Existing surveys, organized by architecture, output modality, or application domain, leave this question implicit. We introduce three progressively stronger capability levels: Plausible models preserve task-relevant temporal, geometric, or physical structure; Controllable models additionally predict how interventions alter that structure; and Actionable models translate predictions into measurable gains in planning, action, learning, evaluation, verification, recovery, or data selection. We complement this hierarchy with a 3 x 4 matrix crossing geometry, physics, and action grounding with improvement loops centered on data, rewards, policies, and the model itself. Using this framework, we survey manipulation, navigation, locomotion, autonomous driving, and general embodied learning, tracing technical progressions, clarifying capability requirements, and examining datasets, benchmarks, and evaluation protocols. We identify challenges in long-horizon consistency, uncertainty calibration, causal intervention testing, latency, verification and recovery, and cross-embodiment transfer. This perspective shifts evaluation from visual plausibility toward whether predictions capture task-relevant state, reflect intervention effects, and improve the closed-loop behavior of embodied agents.

cs.RO↗
arXiv2026

Boosting LLM Reasoning via Human-Inspired Reward Shaping

Reinforcement Learning with Verifiable Rewards (RLVR) has emerged as a promising paradigm for enhancing reasoning in Large Language Models (LLMs). However, existing reward formulations typically treat exploration and consolidation as a monolithic process, resulting in entangled stage-wise learning dynamics. This contradicts the natural learning behavior of human learners. In human learning, individuals adopt distinct behavioral patterns toward mastered versus unfamiliar problems. When confronting unmastered challenges, humans prioritize broad exploration to seek viable solutions. By contrast, for well-mastered problems, they focus instead on reasoning condensation and knowledge abstraction to distill concise underlying principles. Motivated by this gap, we introduce T2T(Thickening-to-Thinning), a dynamic reward framework inspired by human learning processes. Specifically, it implements a dual-phase mechanism: (1) On incorrect attempts, T2T incentivizes "thickening" to broaden the search space and explore novel solution paths; (2) Upon achieving correctness, it shifts to "thinning", imposing length penalties to discourage redundancy, thereby fostering model confidence and crystallizing reasoning capabilities. Extensive experiments on mathematical benchmarks (MATH-500, AIME, AMC) across 5 mainstream LLMs demonstrate that T2T significantly outperforms standard GRPO and recent baselines, achieving superior performance.

cs.LG↗
arXiv2026

On-policy Distillation with Verifiable Reward

Reinforcement Learning with Verifiable Rewards (RLVR) and on-policy distillation (OPD) have become two widely adopted paradigms for post-training large language models. However, RLVR suffers from sparse task-level feedback, while OPD provides dense token-level guidance but ignores trajectory correctness, limiting its performance to that of the teacher. Combining them is a promising direction: OPD supplies dense supervisory signals, while RLVR provides task-level correctness. Nevertheless, existing integrations often rely on weighted combination or heuristic switching, introducing extra hyperparameters and trade-offs. We propose On-policy Distillation with Verifiable Reward (OPDVR), a simple yet effective method that seamlessly combines OPD and RLVR without adding any hyperparameters. We first reformulate the implicit reward of sampled-token OPD based on trajectory correctness, then apply a ReLU gating mechanism to ensure that correct trajectories receive non-negative rewards and incorrect ones receive non-positive rewards---thereby aligning the distillation signal with task success while preserving the teacher's distributional guidance. Furthermore, our modification transforms sampled-token OPD into a proper RLVR method, making it readily combinable with any policy gradient algorithm, such as GRPO. Experiments on six reasoning benchmarks show that OPDVR consistently outperforms standard OPD. Our code is available at https://github.com/LeapLabTHU/OPDVR.

cs.LG↗
arXiv2026

Towards World Models in Biomedical Research

A central goal of biomedicine is to understand, predict and ultimately control the dynamic mechanisms by which biological systems respond to perturbations, disease progression and therapeutic intervention. Although foundation models and large language models have accelerated biomedical data interpretation, most current systems remain focused on static pattern recognition rather than prospective simulation of biological futures. Here we propose biomedical world models as a paradigm for AI-driven discovery. These models learn latent representations of molecular, cellular, tissue and clinical states, together with intervention-conditioned dynamics that allow future trajectories to be simulated before actions are taken. We discuss how biomedical world models could function as data engines, environment simulators and scientific planning substrates across applications including virtual cells, organoids, virtual patients and surgical simulation. We outline the data infrastructure, evaluation benchmarks, safety constraints and governance frameworks required. Biomedical world models may provide a foundation for simulation-guided, closed-loop and experimentally actionable biomedical discovery.

cs.AI↗
arXiv2026

PhaseLift for Coded Diffraction Patterns: Optimal Sampling Rate

Recovering a complex-valued signal from coded diffraction patterns, namely the Fourier intensities obtained after modulating the signal with a collection of masks, is a fundamental structured phase retrieval problem arising in diffraction imaging and related applications. Despite its practical importance, the theoretical analysis of this structured framework remains scarce. In the standard random mask model, the optimal sampling rate achievable by computationally tractable recovery methods has remained open. In this paper, we establish the optimal sampling rate for the PhaseLift feasibility program. More precisely, PhaseLift achieves exact recovery of an unknown signal $\pmb{x}_0\in\mathbb{C}^n$, up to a global phase, from $\mathcal{O}(\log n)$ random masks, with polynomially decaying failure probability. Since $Ω(\log n)$ masks are necessary to identify certain signals under the erasure mask ensemble, our result thereby achieves the optimal mask complexity. Equivalently, PhaseLift attains the optimal total sampling rate of $m=\mathcal{O}( n\log n)$ scalar intensity measurements. The proof is based on an approximate dual certificate construction via a refined golfing scheme that combines adaptive mask allocation with a dimension-independent truncation threshold.

cs.IT↗
arXiv2026

Low-Rank Matrix Recovery via Heavy-Tailed Quadratic Sampling

The problem of recovering an (approximately) low-rank Hermitian matrix $\pmb{M}_0 \in \mathbb{C}^{n \times n}$ of rank $r$ from quadratic sampling matrices of the form $\{\pmb{a}_k \pmb{a}_k^*\}_{k=1}^m$ arises in a variety of applications, including phase retrieval. To obtain rigorous recovery guarantees, the sampling vectors $\{\pmb{a}_k\}_{k=1}^m$ are typically modeled probabilistically. However, most existing theoretical results rely on Gaussian or sub-Gaussian assumptions, which may not accurately capture practical data models. In many applications, sampling vectors exhibit heavier tails, while theoretical understanding in such regimes remains scarce. In this paper, we bridge this gap. We show that two widely used convex approaches, nuclear norm minimization and semidefinite-constrained empirical risk minimization, achieve uniform, stable, and robust recovery under the mild assumption that the entries of the sampling vectors have only finite $4+δ$ moments, with the optimal sample complexity $m = \mathcal{O}(rn)$ up to moment-dependent constants. The two main ingredients of our analysis are moment estimates for quadratic forms established via decoupling, together with recent advances in covariance estimation in heavy-tailed settings. As byproducts, we also establish the optimal sample complexity for low-rank matrix recovery under complex projective $4$-design sampling, thereby improving upon previous results, and obtain stability guarantees for phase retrieval under similarly weak moment assumptions.

math.ST↗
arXiv2026

Focused Width in Adversarial Fake Detection: A Separation

We study the adversarial fake detection model introduced by Mendelson, Paouris and Vershynin. In this model, a genuine sample is $\pmb{X}\sim N(0,\pmb{I}_n)$, while a fake sample is produced as $\pmb{X}+r\pmb{t}({\pmb{X}})$, where the adversary first observes $\pmb{X}$ and then chooses an admissible perturbation $\pmb{t}({\pmb{X}})$ from a prescribed set $\mathscr{T}\subset\mathbb{R}^n$. The central quantity is the detectability radius $r(\mathscr{T})$, which formalizes the transition scale at which fake samples become reliably distinguishable from genuine ones. Mendelson, Paouris and Vershynin introduced the focused width $\widetilde{w}(\mathscr{T})$ as a geometric parameter for this radius and conjectured that, for every origin-symmetric set $\mathscr{T}$, it characterizes $r(\mathscr{T})$ up to universal constants. In this note, we disprove this conjecture for a broad class of discrete sets. More precisely, we consider any origin-symmetric set $\mathscr{T}_n$ lying between the hypercube and the odd integer grid: \begin{equation*} \{-1,1\}^n\subset\mathscr{T}_n\subset ( 2\mathbb{Z}+1)^n. \end{equation*} For every such $\mathscr{T}_n$, we prove that $\frac{\widetilde{w}(\mathscr{T}_n)}{r(\mathscr{T}_n) }\gtrsim \sqrt{\log n}$. Thus, in the Gaussian model, the focused width can overestimate the detectability radius by a $\sqrt{\log n}$ factor and therefore does not characterize it in general. We further show that this logarithmic scale is not intrinsic: in the corresponding non-Gaussian model with product Laplace data, the focused width benchmark can even exceed the detectability radius by at least a polynomial factor of order $n^{1/4}$.

math.ST↗
arXiv2026

UltraStar: Semantic-Aware Star Graph Modeling for Echocardiography Navigation

Echocardiography is critical for diagnosing cardiovascular diseases, yet the shortage of skilled sonographers hinders timely patient care, due to high operational difficulties. Consequently, research on automated probe navigation has significant clinical potential. To achieve robust navigation, it is essential to leverage historical scanning information, mimicking how experts rely on past feedback to adjust subsequent maneuvers. Practical scanning data collected from sonographers typically consists of noisy trajectories inherently generated through trial-and-error exploration. However, existing methods typically model this history as a sequential chain, forcing models to overfit these noisy paths, leading to performance degradation on long sequences. In this paper, we propose UltraStar, which reformulates probe navigation from path regression to anchor-based global localization. By establishing a Star Graph, UltraStar treats historical keyframes as spatial anchors connected directly to the current view, explicitly modeling geometric constraints for precise positioning. We further enhance the Star Graph with a semantic-aware sampling strategy that actively selects the representative landmarks from massive history logs, reducing redundancy for accurate anchoring. Extensive experiments on a dataset with over 1.31 million samples demonstrate that UltraStar outperforms baselines and scales better with longer input lengths, revealing a more effective topology for history modeling under noisy exploration. Code is available at https://github.com/LeapLabTHU/UltraStar.

cs.CV↗
arXiv2026

Deeper is Not Always Better: Mitigating the Alignment Tax via Confident Layer Decoding

Autoregressive generation in large language models (LLMs) conventionally decodes from the final layer, assuming that deeper representations yield more reliable next-token predictions. We revisit this assumption by revealing a recurring Guess-Refine-Perturb dynamic: early layers form coarse guesses, intermediate layers refine reasoning-relevant semantics, and final layers can perturb these refined predictions toward generic or alignment-preferred tokens. We introduce Confident Decoding, a training-free decoding strategy that dynamically selects the most reliable near-final layer through entropy-guided conservative backward search. We further provide a theoretical formulation of layer selection as an optimal stopping problem, showing that under bounded projection noise and dominant late-stage alignment perturbation, our search rule filters perturbation while bounding the loss relative to the oracle refinement layer. Experiments across dense and Mixture-of-Experts LLMs demonstrate consistent gains on challenging reasoning benchmarks, including GPQA-Diamond, Omni-MATH, and HLE, with zero memory overhead and less than 2% latency increase. These results suggest dynamically bypassing final-layer perturbations can unlock stronger reasoning behavior from aligned LLMs.

cs.CL↗
arXiv2026

Action with Visual Primitives

Vision-Language-Action (VLA) models have emerged as a promising paradigm for generalist robotic manipulation. A common design in current architectures maps language instructions and visual observations to actions in a single forward pass. While conceptually simple, this formulation entangles instruction comprehension, spatial scene understanding, and motor control within a single learning objective. As a result, the action expert must implicitly relearn cognitive and perceptual capabilities already present in the pretrained VLM, which can limit both learning efficiency and generalization. We introduce AVP (Action with Visual Primitives), an end-to-end architecture that implements this visual-primitive-centric interface: the VLM infers the next-stage target and emits visual-primitive tokens that condition a flow-matching action expert, with supervision derived from end-effector kinematics. Real-robot experiments on general pick-and-place tasks show that AVP improves the success rate by 37.04% over pi_0.5 and outperforms other recent methods, with consistent gains in data efficiency, spatial-compositional generalization, and object-level transfer.

cs.RO↗
arXiv2026

The Flexibility Trap: Rethinking the Value of Arbitrary Order in Diffusion Language Models

Diffusion Large Language Models (dLLMs) break the rigid left-to-right constraint of traditional LLMs, enabling token generation in arbitrary orders. Intuitively, this flexibility implies a solution space that strictly supersets the fixed autoregressive trajectory, theoretically unlocking superior reasoning potential. However, in this paper, we find that for general reasoning tasks (e.g., mathematics and coding), arbitrary order generation may in fact limit the reasoning potential of dLLMs. We observe that dLLMs tend to exploit this order flexibility to bypass high-uncertainty tokens that are crucial for exploration, which can lead to a premature collapse of solution coverage. This observation motivates a rethink of RL approaches for dLLMs, where considerable complexities, such as handling combinatorial trajectories and intractable likelihoods, are often devoted to preserving this flexibility. We show that effective reasoning can be elicited by simply forgoing arbitrary order and applying standard Group Relative Policy Optimization (GRPO) instead. Our approach, JustGRPO, is minimalist yet surprisingly effective (e.g., 89.1% accuracy on GSM8K) while fully retaining the parallel decoding ability of dLLMs. Project page: https://nzl-thu.github.io/the-flexibility-trap

cs.CL↗
arXiv2026

MemoryVLA++: Temporal Modeling via Memory and Imagination in Vision-Language-Action Models

Temporal modeling is essential for robotic manipulation, as effective control requires both memory of past interactions and imagination of future states. However, most VLA models rely primarily on the current observation and therefore struggle with long-horizon, temporally dependent tasks. Cognitive science suggests that humans rely on working memory to buffer short-lived context, the hippocampal system to preserve episodic memory of past experience, and internal models to imagine possible future state evolution. Inspired by these mechanisms, we propose MemoryVLA++, a full temporal modeling framework that equips VLA models with memory and imagination for robotic manipulation. A pretrained VLM encodes the current observation into perceptual and cognitive tokens, forming working memory. These tokens query a Perceptual-Cognitive Memory Bank to retrieve relevant historical context. This bank stores low-level details and high-level semantics from past interactions, and is updated through redundancy-aware consolidation. A world model imagines future states in a denoising latent space, and the imagined latents are integrated under memory guidance to form full temporal-aware tokens. The resulting tokens condition a diffusion action expert to predict temporally consistent action sequences. We conduct extensive experiments on 5 simulation benchmarks and 3 categories of real-robot tasks across 3 robots, covering general manipulation, long-horizon temporal tasks, robustness, and generalization. Our method achieves strong performance across Libero, SimplerEnv, Mikasa-Robo, Calvin, Libero-Plus, and diverse real-robot tasks, validating the effectiveness of full temporal modeling with memory and imagination. For example, on real robots, it achieves +9%, +26%, +28% gains on general, memory-dependent, and imagination-dependent tasks. Project Page: https://shihao1895.github.io/MemoryVLA-PP-Web

cs.RO↗
arXiv2026

Potential-Guided Flow Matching for Vision-Language-Action Policy Improvement

Large vision-language-action (VLA) policies are increasingly trained as conditional generative models over action chunks. Yet deployment produces mixed-quality experience-successful demonstrations, partial completions, recoverable mistakes, and failures-that is difficult to use with standard imitation. Full behavior cloning (BC) imitates failures, filtered BC discards useful sub-trajectories, and offline reinforcement learning adds a large critic. We introduce ForesightFlow, a self-guided flow-matching policy that augments each generated action chunk with a learned success-potential trajectory. The same flow proposes and scores candidate actions, enabling best-of-$K$ inference without an external critic. The key issue is that policy improvement and value calibration require different supervision: advantage weighting should emphasize high-quality actions, but applying the same weights to potential coordinates suppresses failure gradients and creates overconfident scores. We address this with decoupled advantage-weighted flow matching, applying exponentiated advantage weights only to action velocities while training potential velocities uniformly. We further derive a one-step boundary estimator for conditional flow matching, allowing advantage computation with a single stop-gradient forward pass. Across five BEHAVIOR-1K simulation tasks and five real-world bimanual tasks, ForesightFlow improves over imitation baselines, matches the strongest separate-critic baseline in simulation success, improves real-world success, and reduces training compute by $38\%$. Ablations show that decoupling prevents value hallucination, the one-step estimator preserves candidate-ranking fidelity, and self-guided sampling improves long-horizon execution.

cs.RO↗
arXiv2026

Linearizing Vision Transformer with Test-Time Training

While linear-complexity attention mechanisms offer a promising alternative to Softmax attention for overcoming the quadratic bottleneck, training such models from scratch remains prohibitively expensive. Inheriting weights from pretrained Transformers provides an appealing shortcut, yet the fundamental representational gap between Softmax and linear attention prevents effective weight transfer. In this work, we address this conversion challenge from two perspectives: architectural alignment and representational alignment. We identify Test-Time Training (TTT) as a linear-complexity architecture whose two-layer dynamic formulation is structurally aligned with Softmax attention, enabling direct inheritance of pretrained attention weights. To further align representational properties, including key shift-invariance and locality, we introduce key instance normalization and a lightweight locality enhancement module. We validate our approach by linearizing Stable Diffusion 3.5 and introduce SD3.5-T$^5$ (Transformer To Test Time Training). With only 1 hour of fine-tuning on 4$\times$H20 GPUs, SD3.5-T$^5$ achieves comparable text-to-image quality to the fine-tuned Softmax model, while accelerating inference by 1.32$\times$ and 1.47$\times$ at 1K and 2K resolutions. Code is available at https://github.com/LeapLabTHU/Transformer-to-TTT.

cs.CV↗
arXiv2026

SiameseNorm: Breaking the Barrier to Reconciling Pre/Post-Norm

The long-standing tension between Pre- and Post-Norm remains an open problem in Transformer architecture, reflecting a fundamental trade-off between training stability and representational capacity. Prior attempts to combine their strengths have made progress, but often show limited robustness across training settings, restricting their broader applicability. We revisit this dilemma, showing that single-stream architectures struggle to reconcile Pre-Norm's stable identity-gradient propagation with Post-Norm's normalization of the main residual path. To address this structural tension, we propose SiameseNorm, a simple yet effective two-stream architecture that remains compatible with Pre-Norm training recipes. SiameseNorm couples Pre-Norm-like and Post-Norm-like streams through shared residual blocks, allowing each residual block to receive optimization signals from both pathways with negligible overhead. Extensive experiments on 400M and 1.3B dense language models, 15B MoE models, Vision Transformers, and Diffusion Transformers show that SiameseNorm consistently improves performance while maintaining strong training stability across architectures and modalities. Code is available at https://github.com/Qwen-Applications/SiameseNorm.

cs.LG↗
arXiv2026

From Reasoning Chains to Verifiable Subproblems: Curriculum Reinforcement Learning Enables Credit Assignment for LLM Reasoning

Reinforcement learning from verifiable rewards (RLVR) has shown strong promise for LLM reasoning, but outcome-based RLVR remains inefficient on hard problems because correct final-answer rollouts are rare and sample-level credit assignment cannot use partial progress in failed attempts. We introduce SCRL (Subproblem Curriculum Reinforcement Learning), a curriculum RL framework that derives verifiable subproblems from reference reasoning chains and fixes the final subproblem as the original problem. This turns partial progress on hard problems into verifiable learning signals. Algorithmically, SCRL uses subproblem-level normalization, which normalizes rewards independently at each subproblem position and assigns the resulting advantages to the corresponding answer spans, enabling finer-grained credit assignment without external rubrics or reward models. Our analysis shows that subproblem curricula lift hard problems out of gradient dead zones, with larger relative gains as the original problem becomes harder. Across seven mathematical reasoning benchmarks, SCRL outperforms strong curriculum-learning baselines, improving average accuracy over GRPO by +4.1 points on Qwen3-4B-Base and +1.9 points on Qwen3-14B-Base. On AIME24, AIME25, and IMO-Bench, SCRL further improves pass@1 by +3.7 points and pass@64 by +4.6 points on Qwen3-4B-Base, indicating better exploration on hard reasoning problems.

cs.LG↗
arXiv2026

TwinRL: Digital Twin-Driven Reinforcement Learning for Real-World Robotic Manipulation

Despite strong generalization capabilities, Vision-Language-Action (VLA) models remain constrained by the high cost of expert demonstrations and limited real-world interaction. While online reinforcement learning (RL) has shown promise, its application to real-world VLA manipulation is hindered by low exploration efficiency and restricted exploration coverage. Through systematic real-world experiments, we observe that the effective exploration space of online RL is largely constrained by the trajectory distribution induced during supervised fine-tuning (SFT). Motivated by this observation, we propose TwinRL, a digital twin-real-world collaborative post-training framework that expands and guides RL exploration for VLA models through three stages: SFT warm-up, twin RL warm-up, and real-world RL. TwinRL first reconstructs a high-fidelity digital twin from smartphone-captured scenes. During the SFT stage, we introduce an exploration space expansion strategy that expands the support of the trajectory distribution beyond real demonstrations, reshaping the exploration space for more effective RL. Rather than treating the twin as a data augmentation tool, we propose a twin RL warm-up strategy that enables it to act as an exploration guide for real-world RL. Specifically, TwinRL performs efficient parallel RL in the digital twin to generate interactive trajectories that populate the replay buffer and stabilize subsequent real-world RL learning. This process also identifies failure-prone yet informative configurations, enabling targeted human-in-the-loop rollouts to further improve on-robot efficiency. Across four tasks, TwinRL achieves near-100% success in both in-distribution and out-of-distribution regions, delivering over 30% faster convergence than prior real-world RL methods with only 20 minutes of on-robot interaction.

cs.RO↗