Searcharxiv⌕ Search

arXiv subjects

Di Zhang

Publications and source records attributed to Di Zhang.

At least 73 records · Page 4Linked to original sources
arXiv2026

Faithful-First Reasoning, Planning, and Acting for Multimodal LLMs

Multimodal Large Language Models (MLLMs) frequently suffer from unfaithfulness, generating reasoning chains that drift from visual evidence or contradict final predictions. We propose Faithful-First Reasoning, Planning, and Acting (RPA) framework in which FaithEvi provides step-wise and chain-level supervision by evaluating the faithfulness of intermediate reasoning, and FaithAct uses these signals to plan and execute faithfulness-aware actions during inference. Experiments across multiple multimodal reasoning benchmarks show that faithful-first RPA improves perceptual faithfulness by up to 24% over prompt-based and tool-augmented reasoning frameworks, without degrading task accuracy. Our analysis shows that treating faithfulness as a guiding principle perceptually faithful reasoning trajectories and mitigates hallucination behavior. This work thereby establishes a unified framework for both evaluating and enforcing faithfulness in multimodal reasoning. Code is at https://github.com/lijunxian111/Faithful-First-RPA.

cs.AI↗
arXiv2026

Constructive Instrumental Variable Identification and Inference with Many Weak Interaction Moments

Instrumental variable methods are widely used for causal inference, but identification becomes especially challenging when instruments are weak and potentially invalid. These challenges are particularly pronounced in Mendelian randomization, where genetic variants serve as instruments and violations of exclusion restriction or independence assumptions are common. We propose MAGIC, a constructive and assumption-lean framework that achieves identification even when all candidate instruments may be invalid. The method exploits pairwise and higher-order interactions among mutually independent instruments to construct moment conditions orthogonal to both unmeasured confounding and direct effects under a linear structural model. The resulting estimation problem involves many potentially weak interaction moments with unknown nuisance parameters. We develop a semiparametric generalized method of moments estimator and introduce a global Neyman orthogonality condition to ensure robustness of both the moment function and its derivative to nuisance estimation under many weak moment asymptotics. We establish consistency and asymptotic normality when the number of moments diverges with sample size and characterize the semiparametric efficiency bound under fixed dimension. Simulations and an application to UK Biobank data illustrate the method.

stat.ME↗
arXiv2026

Point What You Mean: Visually Grounded Instruction Policy

Vision-Language-Action (VLA) models align vision and language with embodied control, but their object referring ability remains limited when relying solely on text prompt, especially in cluttered or out-of-distribution (OOD) scenes. In this study, we introduce the Point-VLA, a plug-and-play policy that augments language instructions with explicit visual cues (e.g., bounding boxes) to resolve referential ambiguity and enable precise object-level grounding. To efficiently scale visually grounded datasets, we further develop an automatic data annotation pipeline requiring minimal human effort. We evaluate Point-VLA on diverse real-world referring tasks and observe consistently stronger performance than text-only instruction VLAs, particularly in cluttered or unseen-object scenarios, with robust generalization. These results demonstrate that Point-VLA effectively resolves object referring ambiguity through pixel-level visual grounding, achieving more generalizable embodied control.

cs.CV↗
arXiv2026

IAG: Input-aware Backdoor Attack on VLM-based Visual Grounding

Recent advances in vision-language models (VLMs) have significantly enhanced the visual grounding task, which involves locating objects in an image based on natural language queries. Despite these advancements, the security of VLM-based grounding systems has not been thoroughly investigated. This paper reveals a novel and realistic vulnerability: the first multi-target backdoor attack on VLM-based visual grounding. Unlike prior attacks that rely on static triggers or fixed targets, we propose IAG, a method that dynamically generates input-aware, text-guided triggers conditioned on any specified target object description to execute the attack. This is achieved through a text-conditioned UNet that embeds imperceptible target semantic cues into visual inputs while preserving normal grounding performance on benign samples. We further develop a joint training objective that balances language capability with perceptual reconstruction to ensure imperceptibility, effectiveness, and stealth. Extensive experiments on multiple VLMs (e.g., LLaVA, InternVL, Ferret) and benchmarks (RefCOCO, RefCOCO+, RefCOCOg, Flickr30k Entities, and ShowUI) demonstrate that IAG achieves the best ASRs compared with other baselines on almost all settings without compromising clean accuracy, maintaining robustness against existing defenses, and exhibiting transferability across datasets and models. These findings underscore critical security risks in grounding-capable VLMs and highlight the need for further research on trustworthy multimodal understanding.

cs.CV↗
arXiv2026

The Efficiency Attenuation Phenomenon: A Computational Challenge to the Language of Thought Hypothesis

This paper computationally investigates whether thought requires a language-like format, as posited by the Language of Thought (LoT) hypothesis. We introduce the ``AI Private Language'' thought experiment: if two artificial agents develop an efficient, inscrutable communication protocol via multi-agent reinforcement learning (MARL), and their performance declines when forced to use a human-comprehensible language, this Efficiency Attenuation Phenomenon (EAP) challenges the LoT. We formalize this in a cooperative navigation task under partial observability. Results show that agents with an emergent protocol achieve 50.5\% higher efficiency than those using a pre-defined, human-like symbolic protocol, confirming the EAP. This suggests optimal collaborative cognition in these systems is not mediated by symbolic structures but is naturally coupled with sub-symbolic computations. The work bridges philosophy, cognitive science, and AI, arguing for pluralism in cognitive architectures and highlighting implications for AI ethics.

cs.AI↗
arXiv2026

CARE: A Molecular-Guided Foundation Model with Adaptive Region Modeling for Whole Slide Image Analysis

Foundation models have recently achieved impressive success in computational pathology, demonstrating strong generalization across diverse histopathology tasks. However, existing models overlook the heterogeneous and non-uniform organization of pathological regions of interest (ROIs) because they rely on natural image backbones not tailored for tissue morphology. Consequently, they often fail to capture the coherent tissue architecture beyond isolated patches, limiting interpretability and clinical relevance. To address these challenges, we present Cross-modal Adaptive Region Encoder (CARE), a foundation model for pathology that automatically partitions WSIs into several morphologically relevant regions. Specifically, CARE employs a two-stage pretraining strategy: (1) a self-supervised unimodal pretraining stage that learns morphological representations from 34,277 whole-slide images (WSIs) without segmentation annotations, and (2) a cross-modal alignment stage that leverages RNA and protein profiles to refine the construction and representation of adaptive regions. This molecular guidance enables CARE to identify biologically relevant patterns and generate irregular yet coherent tissue regions, selecting the most representative area as ROI. CARE supports a broad range of pathology-related tasks, using either the ROI feature or the slide-level feature obtained by aggregating adaptive regions. Based on only one-tenth of the pretraining data typically used by mainstream foundation models, CARE achieves superior average performance across 33 downstream benchmarks, including morphological classification, molecular prediction, and survival analysis, and outperforms other foundation model baselines overall.

cs.CV↗
arXiv2026

Cyclic Delay-Doppler Shift: A Simple Transmit Diversity Technique for Ultra-Reliable Communications in Doubly Selective Channels

Affine frequency division multiplexing (AFDM) and orthogonal time frequency space (OTFS) are two promising advanced waveforms proposed for reliable communications in high-mobility scenarios. In this paper, we introduce a simple transmit diversity technique, termed cyclic delay-Doppler shift (CDDS), for these two advanced waveforms to achieve ultra-reliable communications in doubly selective channels (DSCs). Two simple CDDS schemes, named modulation-domain CDDS (MD-CDDS) and time-domain CDDS (TD-CDDS), are proposed, which perform CDDS in advance at the transmitter before and after the modulation, respectively. We demonstrate that both of the two proposed CDDS schemes can be implemented efficiently and flexibly by multiplying the transmit vector with a well-designed precoding matrix, which is nothing but a sparse phase-compensated permutation matrix. Moreover, we theoretically and numerically prove that CDDS can provide MIMO-AFDM and MIMO-OTFS with optimal transmit diversity gain when a proper CDDS step is adopted. Compared to the conventional transmit diversity techniques, the proposed CDDS scheme enjoys the advantages of lower channel estimation overhead, implementation complexity, and signal processing latency, making it particularly suitable for ultra-reliable communications in high-mobility scenarios.

eess.SP↗
arXiv2026

Digital Hydrogen Platform (DigHyd): A Rigorously Curated Database for Hydrogen Storage Materials Empowered by AI-Assisted Literature Mining

Solid-state hydrogen storage materials are promising candidates for safe and compact hydrogen storage; however, data-driven discovery in this field remains limited by the availability of large-scale, well-curated datasets. Here, we present the Digital Hydrogen Platform (DigHyd: www.dighyd.org), a rigorously curated database comprising $>4,000$ experimental literature sources and $>30,000$ data entries on hydrogen storage materials, constructed through AI-assisted literature mining combined with human-in-the-loop validation. In addition to gravimetric hydrogen storage density ($w$), DigHyd also covers thermodynamic parameters, specifically the enthalpy ($ΔH$) and entropy ($ΔS$) changes associated with hydrogenation reactions, primarily defined as $M + \frac{1}{2} {\rm H}_2 \rightleftarrows M{\rm H}$. These parameters were obtained by manually analyzing multi-temperature pressure-composition-temperature (PCT) data using van't Hoff analysis. By focusing on $ΔH$ and $ΔS$ rather than fixing equilibrium pressure at a single temperature, DigHyd enables flexible evaluation of equilibrium behavior under application-specific operating conditions. Statistical analyses reveal distinct distributions of thermodynamic parameters across material classes, together with broad compositional variability within representative hydride systems. Furthermore, both physically interpretable symbolic regression and black-box XGBoost models achieve comparable predictive performance for $w$ and equilibrium pressure at room temperature ($P_{\rm eq,RT}$), demonstrating internal consistency and learnable composition-property relationships within the curated dataset. Overall, DigHyd provides a rigorously curated thermodynamic dataset that serves as a reliable basis for data-driven analyses of hydrogen storage materials and supports systematic exploration of structure-property relationships.

cond-mat.mtrl-sci↗
arXiv2026

A Survey on Wi-Fi Sensing Generalizability: Taxonomy, Techniques, Datasets, and Future Research Prospects

Wi-Fi sensing has emerged as a powerful non-intrusive technology for recognizing human activities, monitoring vital signs, and enabling context-aware applications using commercial wireless devices. However, the performance of Wi-Fi sensing often degrades when applied to new users, devices, or environments due to significant domain shifts. To address this challenge, researchers have proposed a wide range of generalization techniques aimed at enhancing the robustness and adaptability of Wi-Fi sensing systems. In this survey, we provide a comprehensive and structured review of over 200 papers published since 2015, categorizing them according to the Wi-Fi sensing pipeline: experimental setup, signal preprocessing, feature learning, and model deployment. We analyze key techniques, including signal preprocessing, domain adaptation, meta-learning, metric learning, data augmentation, cross-modal alignment, federated learning, and continual learning. Furthermore, we summarize publicly available datasets across various tasks, such as activity recognition, user identification, indoor localization, and pose estimation, and provide insights into their domain diversity. We also discuss emerging trends and future directions, including large-scale pretraining, integration with multimodal foundation models, and continual deployment. To foster community collaboration, we introduce the Sensing Dataset Platform (SDP) for sharing datasets and models. This survey aims to serve as a valuable reference and practical guide for researchers and practitioners dedicated to improving the generalizability of Wi-Fi sensing systems. Survey papge: https://github.com/aiotgroup/awesome-wireless-sensing-generalization.

cs.CV↗
arXiv2026

Competing Hydrogenation Pathways to Metastable CaH$_6$ Revealed by Machine-Learning-Potential Molecular Dynamics

The synthesis of the high-$T_c$ superhydride CaH$_6$ has stimulated significant interest in understanding synthesis pathways for metastable hydrides. However, the microscopic mechanisms governing such hydrogenation reactions remain poorly understood. Here, we show that machine-learning potential molecular dynamics (MLP-MD) simulations can reproduce and distinguish competing reaction pathways leading to metastable and stable hydrides. By simulating hydrogenation reactions at CaH$_2$/H$_2$ and CaH$_4$/H$_2$ interfaces, we identify two distinct pathways that produce clathrate-type CaH$_6$ and A15-type CaH$_{5.75}$, respectively. CaH$_{5.75}$ lies on the convex hull but requires extensive Ca sublattice rearrangement and therefore forms only at elevated temperatures. In contrast, CaH$_6$ becomes kinetically accessible when CaH$_2$ is used as the precursor. The crystallographic compatibility between the Ca sublattice of CaH$_2$ and the bcc framework of CaH$_6$ enables a martensitic-like topotactic transformation that bypasses the reconstructive pathway leading to CaH$_{5.75}$. These results reveal how precursor structure and thermodynamic stability compete to determine superhydride formation pathways and demonstrate that machine-learning molecular dynamics can directly capture the kinetic selection of metastable phases in reactive materials systems.

cond-mat.mtrl-sci↗
arXiv2026

Trimaximal Mixing Patterns Meet the First JUNO Result

The JUNO experiment has recently released its first measurement results based on 59.1 days of data, achieving unprecedented precision in measuring the lepton mixing angle $θ_{12}$. This significant improvement places stringent constraints on certain neutrino mass models and flavor mixing patterns. In this work, we examine the impact of the latest JUNO results on the two trimaximal (i.e., TM1 and TM2) mixing patterns. They are two well-motivated variants of the tri-bimaximal mixing pattern and predict specific correlations between $θ_{12}$ and $θ_{13}$. After taking into account the first JUNO results, the TM1 mixing pattern sits on the edge of the experimentally allowed $1σ$ region, while the TM2 mixing pattern lies outside the $3σ$ region. To reconcile these TM mixing patterns with the latest experimental data, we further investigate the renormalization group (RG) running effects on them in the both Majorana and Dirac neutrino cases. Our analytical and numerical results show that RG corrections can bring the two TM mixing patterns into excellent agreement with the latest JUNO data if neutrino masses are quasi-degenerate. However, the Majorana case faces severe constraints from neutrinoless double beta decay limits, and particularly, the TM2 mixing pattern with Majorana neutrinos has been essentially ruled out. In the Dirac case, the TM1 mixing pattern is fully consistent with current data including beta decay results, whereas the TM2 pattern is strongly constrained by the KATRIN limit and even could be largely ruled out if the KATRIN experiment reaches its final sensitivity without any discovery. Future high-precision measurements of lepton mixing parameters and absolute neutrino masses in both oscillation and non-oscillation experiments will provide decisive tests of these mixing patterns.

hep-ph↗
arXiv2026

Sparse-BitNet: 1.58-bit LLMs are Naturally Friendly to Semi-Structured Sparsity

Semi-structured N:M sparsity and low-bit quantization (e.g., 1.58-bit BitNet) are two promising approaches for improving the efficiency of large language models (LLMs), yet they have largely been studied in isolation. In this work, we investigate their interaction and show that 1.58-bit BitNet is naturally more compatible with N:M sparsity than full-precision models. To study this effect, we propose Sparse-BitNet, a unified framework that jointly applies 1.58-bit quantization and dynamic N:M sparsification while ensuring stable training for the first time. Across multiple model scales and training regimes (sparse pretraining and dense-to-sparse schedules), 1.58-bit BitNet consistently exhibits smaller performance degradation than full-precision baselines at the same sparsity levels and can tolerate higher structured sparsity before accuracy collapse. Moreover, using our custom sparse tensor core, Sparse-BitNet achieves substantial speedups in both training and inference, reaching up to 1.30X. These results highlight that combining extremely low-bit quantization with semi-structured N:M sparsity is a promising direction for efficient LLMs. Code available at https://github.com/AAzdi/Sparse-BitNet

cs.CL↗
arXiv2026

SlideSparse: Fast and Flexible (2N-2):2N Structured Sparsity

NVIDIA's 2:4 Sparse Tensor Cores deliver 2x throughput but demand strict 50% pruning -- a ratio that collapses LLM reasoning accuracy (Qwen3: 54% to 15%). Milder $(2N-2):2N$ patterns (e.g., 6:8, 25% pruning) preserve accuracy yet receive no hardware support, falling back to dense execution without any benefit from sparsity. We present SlideSparse, the first system to unlock Sparse Tensor Core acceleration for the $(2N-2):2N$ model family on commodity GPUs. Our Sliding Window Decomposition reconstructs any $(2N-2):2N$ weight block into $N-1$ overlapping 2:4-compliant windows without any accuracy loss; Activation Lifting fuses the corresponding activation rearrangement into per-token quantization at near-zero cost. Integrated into vLLM, SlideSparse is evaluated across various GPUs (A100, H100, B200, RTX 4090, RTX 5080, DGX-spark), precisions (FP4, INT8, FP8, BF16, FP16), and model families (Llama, Qwen, BitNet). On compute-bound workloads, the measured speedup ratio (1.33x) approaches the theoretical upper-bound $N/(N-1)=4/3$ at 6:8 weight sparsity in Qwen2.5-7B, establishing $(2N-2):2N$ as a practical path to accuracy-preserving LLM acceleration. Code available at https://github.com/bcacdwk/vllmbench.

cs.LG↗
arXiv2026

Mod-Adapter: Tuning-Free and Versatile Multi-concept Personalization via Modulation Adapter

Personalized text-to-image generation aims to synthesize images of user-provided concepts in diverse contexts. Despite recent progress in multi-concept personalization, most are limited to object concepts and struggle to customize abstract concepts (e.g., pose, lighting). Some methods have begun exploring multi-concept personalization supporting abstract concepts, but they require test-time fine-tuning for each new concept, which is time-consuming and prone to overfitting on limited training images. In this work, we propose a novel tuning-free method for multi-concept personalization that can effectively customize both object and abstract concepts without test-time fine-tuning. Our method builds upon the modulation mechanism in pre-trained Diffusion Transformers (DiTs) model, leveraging the localized and semantically meaningful properties of the modulation space. Specifically, we propose a novel module, Mod-Adapter, to predict concept-specific modulation direction for the modulation process of concept-related text tokens. It introduces vision-language cross-attention for extracting concept visual features, and Mixture-of-Experts (MoE) layers that adaptively map the concept features into the modulation space. Furthermore, to mitigate the training difficulty caused by the large gap between the concept image space and the modulation space, we introduce a VLM-guided pre-training strategy that leverages the strong image understanding capabilities of vision-language models to provide semantic supervision signals. For a comprehensive comparison, we extend a standard benchmark by incorporating abstract concepts. Our method achieves state-of-the-art performance in multi-concept personalization, supported by quantitative, qualitative, and human evaluations.

cs.CV↗
arXiv2026

Bayesian Optimality of In-Context Learning with Selective State Spaces

We propose Bayesian optimal sequential prediction as a new principle for understanding in-context learning (ICL). Unlike interpretations framing Transformers as performing implicit gradient descent, we formalize ICL as meta-learning over latent sequence tasks. For tasks governed by Linear Gaussian State Space Models (LG-SSMs), we prove a meta-trained selective SSM asymptotically implements the Bayes-optimal predictor, converging to the posterior predictive mean. We further establish a statistical separation from gradient descent, constructing tasks with temporally correlated noise where the optimal Bayesian predictor strictly outperforms any empirical risk minimization (ERM) estimator. Since Transformers can be seen as performing implicit ERM, this demonstrates selective SSMs achieve lower asymptotic risk due to superior statistical efficiency. Experiments on synthetic LG-SSM tasks and a character-level Markov benchmark confirm selective SSMs converge faster to Bayes-optimal risk, show superior sample efficiency with longer contexts in structured-noise settings, and track latent states more robustly than linear Transformers. This reframes ICL from "implicit optimization" to "optimal inference," explaining the efficiency of selective SSMs and offering a principled basis for architecture design.

cs.LG↗
arXiv2026

Golden Goose: A Simple Trick to Synthesize Unlimited RLVR Tasks from Unverifiable Internet Text

Reinforcement Learning with Verifiable Rewards (RLVR) has become a cornerstone for unlocking complex reasoning in Large Language Models (LLMs). Yet, scaling up RL is bottlenecked by limited existing verifiable data, where improvements increasingly saturate over prolonged training. To overcome this, we propose Golden Goose, a simple trick to synthesize unlimited RLVR tasks from unverifiable internet text by constructing a multiple-choice question-answering version of the fill-in-the-middle task. Given a source text, we prompt an LLM to identify and mask key reasoning steps, then generate a set of diverse, plausible distractors. This enables us to leverage reasoning-rich unverifiable corpora typically excluded from prior RLVR data construction (e.g., science textbooks) to synthesize GooseReason-0.7M, a large-scale RLVR dataset with over 0.7 million tasks spanning mathematics, programming, and general scientific domains. Empirically, GooseReason effectively revives models saturated on existing RLVR data, yielding robust, sustained gains under continuous RL and achieving new state-of-the-art results for 1.5B and 4B-Instruct models across 15 diverse benchmarks. Finally, we deploy Golden Goose in a real-world setting, synthesizing RLVR tasks from raw FineWeb scrapes for the cybersecurity domain, where no prior RLVR data exists. Training Qwen3-4B-Instruct on the resulting data GooseReason-Cyber sets a new state-of-the-art in cybersecurity, surpassing a 7B domain-specialized model with extensive domain-specific pre-training and post-training. This highlights the potential of automatically scaling up RLVR data by exploiting abundant, reasoning-rich, unverifiable internet text.

cs.AI↗
arXiv2026

Learning Video Generation for Robotic Manipulation with Collaborative Trajectory Control

Recent advances in video diffusion models shows promise for generating robotic decision-making data, with trajectory conditions further enabling fine-grained control. However, existing methods primarily focus on individual object motion and struggle to capture multi-object interaction crucial in complex manipulation. This limitation arises from entangled features in overlapping regions, leading to degraded visual fidelity. To address this, we present RoboMaster, a novel framework that models inter-object dynamics via a collaborative trajectory formulation. Unlike prior methods that decompose objects, our core is to decompose the interaction process into three sub-stages: pre-interaction, interaction, and post-interaction, and models each phase using the dominant object, specifically the robotic arm in the pre- and post-interaction phases and the manipulated object during interaction. This design effectively alleviates the multi-object feature fusion issue in prior work. To further ensure subject semantic consistency across the video, we incorporate appearance- and shape-aware latent representations for objects. Extensive experiments on the challenging Bridge dataset, as well as RLBench and SIMPLER benchmarks, demonstrate that our method establishs new state-of-the-art performance in trajectory-controlled video generation for robotic manipulation. Project Page: https://fuxiao0719.github.io/projects/robomaster/

cs.CV↗
arXiv2026

Unleashing the Potential of Sparse Attention on Long-term Behaviors for CTR Prediction

In recent years, the success of large language models (LLMs) has driven the exploration of scaling laws in recommender systems. However, models that demonstrate scaling laws are actually challenging to deploy in industrial settings for modeling long sequences of user behaviors, due to the high computational complexity of the standard self-attention mechanism. Despite various sparse self-attention mechanisms proposed in other fields, they are not fully suited for recommendation scenarios. This is because user behaviors exhibit personalization and temporal characteristics: different users have distinct behavior patterns, and these patterns change over time, with data from these users differing significantly from data in other fields in terms of distribution. To address these challenges, we propose SparseCTR, an efficient and effective model specifically designed for long-term behaviors of users. To be precise, we first segment behavior sequences into chunks in a personalized manner to avoid separating continuous behaviors and enable parallel processing of sequences. Based on these chunks, we propose a three-branch sparse self-attention mechanism to jointly identify users' global interests, interest transitions, and short-term interests. Furthermore, we design a composite relative temporal encoding via learnable, head-specific bias coefficients, better capturing sequential and periodic relationships among user behaviors. Extensive experimental results show that SparseCTR not only improves efficiency but also outperforms state-of-the-art methods. More importantly, it exhibits an obvious scaling law phenomenon, maintaining performance improvements across three orders of magnitude in FLOPs. In online A/B testing, SparseCTR increased CTR by 1.72\% and CPM by 1.41\%. Our source code is available at https://github.com/laiweijiang/SparseCTR.

cs.IR↗