arXiv subjects

Meng Li

Publications and source records attributed to Meng Li.

At least 37 records · Page 2Linked to original sources

arXiv2026

All Work And No Play Makes Jack a Dull Boy: Understanding and Preventing Catastrophic Strategy Collapse in RLVR

During post-training of large language models (LLMs) with Reinforcement Learning with Verifiable Rewards (RLVR), GRPO-style algorithms can exhibit severe late-stage collapse. Prompt-based probing reveals that this is not benign strategic pruning, but a harmful contraction of effective strategy capacity that makes distinct reasoning strategies increasingly inaccessible. To characterize this phenomenon, we define strategies through trajectory-level policy-update interactions and develop a unified theoretical framework combining optimization dynamics and information theory. We prove that major RLVR objectives progressively concentrate probability mass onto a single strategy, while sustaining nontrivial task accuracy requires a minimum strategy capacity. The conflict between these two results provides a mechanistic explanation for catastrophic collapse. We further derive the {Mirrored Entanglement Index (MEI)} as a lightweight online warning signal. To prevent collapse, we propose \textbf{Mesh Learning}, which exposes multiple reasoning strategies and prevents any single strategy from dominating optimization. Across AIME26, AIME25, MATH-500, GPQA, and LiveCodeBench, Mesh Learning consistently outperforms strong baselines across Qwen and Phi model families, with gains of up to 13.4 pp and 11.5 pp, respectively. These results establish strategy preservation as a key principle for stable RLVR. Code is available at https://github.com/Ayanami-0123/Open-Mesh-Learning.

cs.LG↗

arXiv2026

WholeBodyWAM: Learning Whole-Body World Action Models with Scalable Motion Priors

Humanoid whole-body manipulation requires coordinated whole-body dynamics, yet large-scale trajectories from a target robot are expensive to collect and difficult to scale. In contrast, whole-body motion from human and humanoid sources is abundantly available, although such data cannot be directly used as embodiment-specific robot actions. This work asks whether these scalable motion resources can instead provide a transferable predictive prior for humanoid world-action modeling. We introduce WholeBodyWAM, a humanoid world-action model that learns whole-body dynamics from large-scale heterogeneous motion before target-robot training. We curate UniMotion-4K, a motion corpus spanning more than 4K hours from human videos, native 3D motion datasets, and heterogeneous humanoid platforms, and canonicalize these diverse sources into a unified motion space. A language-conditioned Motion Expert is then pretrained to predict future whole-body motion without target-robot action supervision. During robot post-training, the pretrained Motion Expert is integrated with Video and Action Experts through asymmetric Mixture-of-Transformers (MoT) attention, enabling predictive scene dynamics and whole-body motion to jointly inform embodiment-specific action generation. Experiments show that WholeBodyWAM consistently benefits from increased motion-pretraining scale, improves future-motion prediction and downstream task performance, and transfers effectively to real-world humanoid manipulation. Moreover, the pretrained motion prior substantially improves data efficiency under limited target-robot demonstrations.

cs.RO↗

arXiv2026

Scaling Parameter and Context in Attention: Native Sparse Attention from Mixture-of-Head

Scaling attention parameters can improve language model quality, but retaining full token histories makes additional heads costly at long contexts. Furthermore, since attention retrieves and combines contextual information, parameter scaling should also support longer contexts. We therefore ask whether attention parameter scaling can directly enable efficient and effective context scaling. We introduce NAMOH, an architecture-native sparse attention mechanism that activates $K$ of $H$ heads per token. Each head retains only its assigned tokens and performs causal attention within this subsequence. Head selection thus jointly determines active parameters and available context without scanning the full history. Under balanced assignments, increasing $H$ at fixed $K$ shortens head histories and reduces per-token key-value (KV) access without increasing total KV storage. We further support head-relative rotary position embeddings to shorten positional spans within routed subsequences, aiming to mitigate position-induced attention noise. Experiments show that NAMOH can outperform fully activated models with the same total parameters, while enabling more efficient long-context inference than smaller dense models with matched active parameter counts. It remains compatible with GQA and existing sparse attention mechanisms. We hope this work offers a new path for scaling attention, with parameter scaling directly enabling context scaling.

cs.CL↗

arXiv2026

BrainWave: A Brain Signal Foundation Model for Clinical Applications

Neural electrical activity is fundamental to brain function, underlying a range of cognitive and behavioral processes, including movement, perception, decision-making, and consciousness. Abnormal patterns of neural signaling often indicate the presence of underlying brain diseases. The variability among individuals, the diverse array of clinical symptoms from various brain disorders, and the limited availability of diagnostic classifications, have posed significant barriers to formulating reliable model of neural signals for diverse application contexts. Here, we present BrainWave, the first foundation model for both invasive and non-invasive neural recordings, pretrained on more than 40,000 hours of electrical brain recordings (13.79 TB of data) from approximately 16,000 individuals. Our analysis show that BrainWave outperforms all other competing models and consistently achieves state-of-the-art performance in the diagnosis and identification of neurological disorders. We also demonstrate robust capabilities of BrainWave in enabling zero-shot transfer learning across varying recording conditions and brain diseases, as well as few-shot classification without fine-tuning, suggesting that BrainWave learns highly generalizable representations of neural signals. We hence believe that open-sourcing BrainWave will facilitate a wide range of clinical applications in medicine, paving the way for AI-driven approaches to investigate brain disorders and advance neuroscience research.

q-bio.NC↗

arXiv2026

Reciprocal Guidance: Orchestrating Draft and Verify Budgets for Advancing the Diffusion-AR Self-Speculation Frontier

Diffusion drafting with autoregressive (AR) verification has emerged as a promising paradigm for efficient speculative decoding. Recent self-speculation models, represented by Nemotron-Labs-Diffusion, further simplify the speculative pipeline by unifying drafting and verification within a shared backbone, while enabling longer acceptance lengths. However, the Pareto frontier between aggregate and per-request throughput remains underexplored. At low concurrency, sequential draft-verify execution requires two model forward passes per round, limiting the effective tokens per forward (TPF). By contrast, at high concurrency, longer drafts incur increasingly expensive computation, forcing individual requests to operate under constrained speculation budgets and preventing full exploitation of the full-backbone drafter. Our key observation indicates that drafting and verification exhibit reciprocal predictability. Draft logits can anticipate likely verification mismatches, while recent verification outcomes predict future drafting utility and suitable block sizes. Building on this observation, we introduce Reciprocal Guidance (RecGuide), a runtime draft-verify orchestration framework that adapts speculative decoding to varying serving loads. RecGuide exploits spare compute capacity through verification-overlapped drafting at low concurrency, while dynamically allocating request-specific draft block sizes as the workload becomes increasingly compute-intensive. Experiments across a wide range of concurrency levels demonstrate consistent throughput improvements over vanilla self-speculation, achieving up to $1.8\times$ speedup.

cs.CL↗

arXiv2026

HBFlex: A Flexible Memory System for Bridging Fine-Grained LLM States and Coarse-Grained HBF Parallel Execution

Large language models (LLMs) require increasing memory capacity to accommodate growing model weights and KV caches. High-Bandwidth Flash (HBF) offers high memory density and aggregate read bandwidth through massive plane-level parallelism, making it an attractive option for LLM serving. However, serving LLMs entirely from HBF introduces three challenges: fine-grained KV reads create placement and access imbalance, incremental writes interfere with foreground reads, and mixed KV lifetimes amplify garbage collection. Hybrid HBM/HBF designs retain HBM to support dynamic KV management, but this allocation reduces the HBF resources available under a fixed packaging budget, limiting aggregate HBF bandwidth. We present HBFlex, a full-HBF memory system with coordinated optimizations for KV reads, writes, and reclamation. HBFlex balances KV placement and attention accesses to improve plane utilization. It aggregates incremental updates and schedules writeback within sufficiently long compute windows to reduce write--read interference. It also combines lifetime-guided block packing with deferred reclamation to reduce valid-page migration. We evaluate HBFlex through trace-driven simulation across different configurations. HBFlex achieves average throughput speedups of up to 1.58$\times$ over FlashAccel and 3.30$\times$ over H3, benefiting from higher HBF bandwidth and more efficient management of dynamic KV-cache reads, writes, and erases.

cs.AR↗

arXiv2026

OptiPrime: Optimizing Private Inference through Protocol-Hardware Co-design

Private deep neural network (DNN) inference based on hybrid homomorphic encryption (HE) and multi-party computation (MPC) can protect user data with a formal guarantee, but at the cost of significant latency overhead due to HE. Customized HE accelerators have been proposed and have achieved orders-of-magnitude speedup for individual HE operations. However, when directly applying a commercial HE accelerator to state-of-the-art HE-MPC frameworks, we observe only limited end-to-end performance gain. This is because HE-MPC frameworks often require wireless transmission of input and output ciphertexts for each HE operation, leading to a severe network communication bottleneck. To overcome this challenge, we introduce OptiPrime, a protocol-hardware co-optimization framework for efficient private DNN inference. OptiPrime features a novel HE protocol for convolutions that substantially reduces the number of transmitted output ciphertexts and mitigates the network communication bottleneck. Meanwhile, as the new protocol introduces complex computation for fewer output ciphertext, we observe new memory access challenges due to a high volume of weight plaintexts and intermediate ciphertexts. Hence, we further propose a lightweight compression system for the weight plaintexts, reducing memory traffic by 10 times, as well as a specialized dataflow to maximize on-chip data reuse of intermediate ciphertexts. Extensive experiments show that our framework outperforms the Cheetah baseline by at most 5.7 times on CPUs and 4.2 times with an accelerator.

cs.AR↗

arXiv2026

ROSETTA: Efficient and Accurate Privacy-Preserving LLM Decoding via Hybrid CKKS/TFHE Evaluation

Generative large language models (LLMs) have achieved state-of-the-art performance on many real-world tasks such as code generation and question answering. These models predominantly rely on an autoregressive decoding strategy that generates output tokens sequentially. However, their pervasive deployment raises serious privacy concerns, motivating private inference frameworks based on fully homomorphic encryption (FHE). A major limitation of existing FHE frameworks is their inefficiency in evaluating nonlinear operations, which incur substantial overhead and dominate the decode stage. In this paper, we propose ROSETTA, a hybrid CKKS/TFHE framework that overcomes this limitation. We first observe that nonlinear operations in the decode stage exhibit heterogeneous workload patterns, which can be handled effectively via a hybrid approach. We then realize this with two key contributions: 1) an adaptive segmented lookup-table protocol based on TFHE that enables efficient and accurate evaluation of nonlinear operations; and 2) a scheme-aware operator-selection framework that automatically assigns each nonlinear operator to CKKS or TFHE to minimize end-to-end decoding latency. We demonstrate that ROSETTA achieves up to $4.8\times$ Softmax speedup and $1.5$--$2.1\times$ end-to-end speedup over the SOTA framework CacheMir.

cs.CR↗

arXiv2026

Adaptive moving mesh methods for isotropic/anisotropic mean curvature flow with axisymmetric geometry

This paper introduces adaptive moving mesh methods for the numerical simulation of axisymmetric mean curvature flow, addressing both isotropic and anisotropic cases. The methods are developed within the framework of the mesh equidistribution principle, where a carefully designed tangential velocity is employed to dynamically redistribute mesh points during the evolution. To accurately capture the key geometric features of the evolving interfaces, we select monitor functions based on the curvature $κ$, its arc-length derivative $κ_s$, and the squared curvature $κ^2$. These monitor functions can be flexibly tailored to suit different problem settings and play a vital role in determining the resulting mesh quality and numerical accuracy. Spatial discretization is performed using central finite differences, while temporal integration is handled with first- and second-order time-stepping schemes, including the BDFk ($k=1,2$) and Crank-Nicolson methods. Additionally, a Lagrange multiplier approach is incorporated into the adaptive system to enforce the underlying geometric constraint, resulting in energy-stable numerical schemes. Numerical experiments confirm the convergence and energy stability of the proposed methods. More importantly, the results clearly show that the proposed methods offer significant advantages in complex geometric evolutions: by utilizing appropriately designed monitor functions, the adaptive methods achieve dynamic redistribution of mesh points, efficiently capturing localized geometric features, significantly improving numerical accuracy, and effectively preventing mesh degeneration, particularly in anisotropic cases.

math.NA↗

arXiv2026

MotionQ: Operator-Conditioned Motion Quotients for Cross-Observation WiFi Gesture Recognition

WiFi gesture recognition is accurate in fixed deployments but often degrades when user orientation, available links, or transceiver placement changes. Unlike ordinary domain shifts, these changes alter the wireless observation operator, so the same motion is expected to produce different measurements. Existing methods nevertheless pursue domain-invariant features and largely overlook changing layouts and observation configurations. Yet changing the observation operator also changes which task-relevant motion cues are physically observable, rather than merely altering the appearance of a fixed set of cues. Under a local linearization of the WiFi forward process, we derive a common task-observability condition under which a strict common linear representation is recoverable from every geometry-induced operator while preserving the gesture task. When the condition fails, enforcing stronger alignment across additional heterogeneous source operators may discard task-relevant cues still observable under individual operators. We therefore present MotionQ, which generates an operator-conditioned two-support motion measure for each candidate geometry. A motion quotient removes only the arbitrary ordering of its unlabeled supports and is represented by permutation-invariant central moments. Rather than matching quotients across operators, single-link-retention interventions encourage each view to retain information sufficient for gesture recognition. Extensive evaluations show that MotionQ is robust to extrapolative observation operators.

cs.HC↗

arXiv2026

Should I Be Polite to My LLM Relevance Judge? Tone as a Severity Operating-Point Shift

Large language models are increasingly used as relevance judges, yet their labels can shift with prompt surface form. We study one such feature -- tone -- on 3,498 TREC DL19/DL20 query-passage pairs, across eight judge models, five classifier-calibrated politeness levels, and three paraphrases per level. Effects are strongly model-dependent: one judge shows a structured U-shaped response, whereas most show only small changes. Where tone changes agreement, the results are more consistent with a shift in the judge's severity operating point -- its overall scoring leniency -- than with improved judgment. Agreement rises or falls as this shift moves the judge toward or away from human annotators' strictness. A query-disjoint cross-fit retains the expected association (Spearman $ρ= -0.683$; exact model-block permutation $p = 0.019$). Tone affects calibration-based agreement more than ranking outcomes: across 32 model-tone contrasts, the largest absolute mean change in NDCG@10 is 0.011, although Kendall's $τ$ as low as 0.743 shows that reordering is reduced, not absent. The account reconciles prior contradictory findings and identifies prompt tone as a potential validity threat when absolute relevance labels matter.

cs.IR↗

arXiv2026

HDA-MoE: Hybrid Parallelism and Dynamic, Adaptive Scheduling for Mixture-of-Experts with 3D Near-Memory Processing

Mixture-of-Experts (MoE) architectures have become a key technique for scaling Large Language Models (LLMs), enabling high model capacity with reduced computational cost. However, this efficiency comes at the expense of increased memory capacity and bandwidth demands. Recent 3D Near-Memory Processing (NMP) architectures, which vertically integrate memory and compute through hybrid bonding, provide high internal bandwidth and energy efficiency, making them attractive for accelerating MoE inference. Nevertheless, the distributed memory and compute organization of NMP systems introduces new challenges for mapping MoE workloads. Existing parallelization strategies, such as Tensor Parallelism (TP) and Expert Parallelism (EP), suffer from either high communication costs or unbalanced computation utilization, leading to inferior efficiency. In addition, the dynamic routing behavior of MoE models further complicates efficient deployment. To address these challenges, we present HDA-MoE, a framework that optimizes MoE execution on NMP architectures through hybrid parallel deployment and runtime scheduling. HDA-MoE integrates an offline hybrid parallel mapping algorithm with an online dynamic and adaptive scheduling mechanism to reduce communication overhead while improving computation utilization. Experimental results show that HDA-MoE achieves a speedup of 1.1x--3.4x over TP, 1.1x--1.5x over EP, 1.1x--3.7x over the Hybrid TP-EP compute-balanced baseline, and 1.1x--1.3x over HD-MoE. Source code is available at https://github.com/PKU-SEC-Lab/HDA-MoE-TCAD26.

cs.AR↗

arXiv2026

EnvPilot: Systematic Design and Evaluation of an Experience-Augmented Agent for Software Environment Setup

Environment Setup is a critical yet complex task in software engineering that relies heavily on expert knowledge. Existing automated environment setup methods lack the ability to accumulate experience from past execution trajectories and to evolve over time. As a result, their performance is limited because they often perform redundant exploration, ignore useful past solutions, and fail to generalize across diverse software ecosystems. We present the systematic design and empirical validation of EnvPilot, an experience-augmented agent that operationalizes trajectory-derived experience reuse for software environment setup. EnvPilot maintains an expandable Trajectory-Derived Memory (TDM), initialized with 667 high-quality experiences. It systematically transforms implicit knowledge from historical execution trajectories into structured experience and retrieves the most relevant guidance during task execution through the Context-aware Retrieval mechanism. This enables EnvPilot to combine multiple validated setup strategies, providing more precise and detailed guidance than methods that rely solely on static project files or web retrieval. To evaluate EnvPilot, we construct AES-Bench, a multilingual benchmark of 112 real-world GitHub instances across 9 programming languages. Experiments show that EnvPilot achieves a new state-of-the-art (SOTA) with a 75.00% Pass@1 success rate while reducing reasoning costs. Our empirical study shows that both the structured experience representation and the Context-aware Retrieval mechanism are essential.

cs.SE↗

arXiv2026

Valid inference for regression with best subset selection

Best subset selection is widely implemented in statistical software and is routinely used by practitioners in scientific fields for variable selection. However, classical confidence intervals and $p$-values constructed after model selection generally fail to achieve their nominal frequentist guarantees, which can invalidate subsequent findings. In this article, we characterize the altered conditional sampling distributions of pivotal quantities after best subset selection. Building on selective-inference techniques developed in related settings, our finite-sample characterization of the AIC selection event reveals that its geometry is a union of finitely many intervals on the real line. This geometry enables exact conditioning and avoids the excessive conditioning common in other post-selection methods. We use this characterization to develop valid inference procedures that provide $p$-values with nominal Type~I error and confidence intervals with finite-sample coverage guarantees. The proposed methods are easy to implement, computationally efficient, and broadly applicable to other commonly used best subset selection criteria. We also study inference with unknown noise level using a Monte Carlo selective test conditional on the AIC-selected model, which controls finite-sample Type~I error at the nominal level under the selected model null. In an application to a classical U.S. consumption dataset, the proposed confidence intervals lead to different conclusions from the conventional intervals, even when the selected model is the full model, producing interpretable findings that better align with empirical observations.

stat.ME↗

arXiv2026

Brain4FMs: A Benchmark of Foundation Models for Electrical Brain Signal

Brain foundation models (BFMs) are advancing neurotechnology by learning transferable representations from neural signals, with broad potential in clinical diagnosis and neuroscience research. Their development relies on large-scale pretraining corpora of electrical brain signals, including scalp electroencephalography (EEG) and intracranial EEG (iEEG). However, existing BFM benchmarks primarily focus on EEG, cover only a limited subset of models, and provide limited analysis beyond downstream performance. We introduce Brain4FMs, the first unified benchmark, to our knowledge, for jointly evaluating BFMs on EEG and iEEG. It integrates 17 representative models and 21 public datasets across clinical diagnosis, sleep staging, communication, and affective computing. Brain4FMs is open and plug-and-play, with dataset-aware preprocessing, cross-subject evaluation, heterogeneous multichannel handling, and standardized downstream adaptation workflows. The benchmark reveals performance variation across tasks, signal modalities, and adaptation protocols, with no single BFM consistently dominating all evaluation scenarios. % To better understand these heterogeneous transfer behaviors, we further conduct exploratory analyses of model-specific properties. To better understand these behaviors, we further conduct exploratory analyses of model-specific properties of spatial, spectral, and discrete representations. The code is available at https://github.com/wajtsq/Brain4FMs.

cs.LG↗

arXiv2026

Understanding Graphene-Perovskite Interactions: From Flake Chemistry to Crystallisation and Solar Cell Performance

Graphene-derivatives are widely employed materials to improve bulk and interface properties of metal-halide perovskite devices. Yet the implications of their flake chemistry and interactions with the perovskite precursors remain unclear. Here, we show that pristine graphene flakes (GF) and more conventional graphene oxide flakes (GOF) are not interchangeable. Density functional theory calculations reveal that GOF interacts more strongly with the perovskite lattice but induces larger structural distortions, stronger interfacial polarisation, and localised gap states. In contrast, GF forms comparatively non-disruptive contacts, a response retained across a wide compositional range. Machine-learning atomistic simulations further show that GF contacts both Pb- and I-containing regions of solvated perovskite nanocrystals, with a strong solvent dependency. Solution spectroscopic characterization indicates that GF additives serve as scaffold for preorganised Pb/I-containing precursors, favouring film crystallisation. In this sense, GF enhances solar cell performance across perovskite compositions, but particularly those facing a more challenging crystallisation. In mixed Sn-Pb perovskite solar cells, GF raises the champion power-conversion efficiency from 21.5\% to 23.7\% with improved storage stability. These results establish pristine GF as a chemically defined additive and connect its atomic-scale interactions with precursor organisation, crystallisation, device performance, and stability.

cond-mat.mtrl-sci↗

arXiv2026

MeRoPE: Metric Rotary Position Embedding for Camera-Controlled Video Generation

In camera-controlled video generation, geometry-aware positional encodings condition tokens on camera extrinsics and per-token viewing rays. Existing schemes, however, have a scale-dependent failure mode on real-world metric camera trajectories: homogeneous projective encodings cause attention logits and feature norms to grow unbounded with physical translation baselines. We propose MeRoPE (Metric Rotary Position Embedding), a norm-preserving relative camera encoding for attention. MeRoPE encodes relative orientations between calibrated viewing rays with orthogonal rotation blocks, maps raw metric displacements into multi-frequency rotary phases, and adds a disparity-anchored correspondence prior along the epipolar arc. This design strictly preserves feature norms, bounds pre-softmax attention logits regardless of the physical translation scale, and maintains exact invariance to global rigid coordinate changes. Across nuScenes and PanShot, which cover large-baseline trajectories and diverse camera optics, respectively, MeRoPE achieves stronger camera control than prior encodings, with the best consistency between generated camera motion and conditioning poses in both rotation and translation. Code will be made publicly available.

cs.CV↗

arXiv2026

Checkerboard: Closed-Form and Data-Independent Trigger Design for Clean-Label Backdoor Attacks

Backdoor attacks threaten the deep-learning supply chain by poisoning a small fraction of the training data so that a model behaves normally on clean inputs but maps triggered inputs to an attacker-chosen class. Clean-label backdoor attacks are especially difficult to audit because poisoned examples preserve their semantic labels. Yet existing clean-label attacks often require surrogate model training, auxiliary data access, or iterative optimization. In this paper, we present \emph{Checkerboard}, a clean-label backdoor attack with closed-form, data-independent trigger design. We formulate trigger design through an input-space Fisher-separability objective and, under a ridge four-neighbor local-smoothness prior for natural images, obtain the pixel-wise checkerboard as a closed-form maximizer of the resulting design proxy without data access, model training, or optimization. Across four benchmark datasets, Checkerboard outperforms the evaluated norm-bounded clean-label attacks and achieves state-of-the-art performance under low global poisoning rates. For example, on CIFAR-10, under a trigger perturbation of $10/255$, poisoning 20 training samples achieves $95.72\%$ Attack Success Rate (ASR). On IN-100, a global poisoning rate of only $0.4\%$ yields over $83\%$ ASR without degrading clean accuracy. The proposed attack also remains effective against state-of-the-art backdoor defenses and shows resistance to adaptive defenses under simple modification.

cs.CR↗