arXiv subjects

Zhi Zheng

Publications and source records attributed to Zhi Zheng.

At least 55 records · Page 3Linked to original sources

arXiv2026

GLEAM: Learning to Match and Explain in Cross-View Geo-Localization

Cross-View Geo-Localization (CVGL) focuses on identifying correspondences between images captured from distinct perspectives of the same geographical location. However, existing CVGL approaches are typically restricted to a single view or modality, and their direct visual matching strategy lacks interpretability: they only determine whether two images correspond, without explaining the rationale behind the match. In this paper, we present GLEAM-C, a foundational CVGL model that unifies multiple views and modalities by aligning them exclusively with satellite imagery. Our framework improves training efficiency through optimized implementation and achieves accuracy comparable to prior modality-specific CVGL models via a novel two-phase training strategy. To address interpretability, we further propose GLEAM-X, a novel task that combines cross-view correspondence prediction with explainable reasoning enabled by multimodal large language models (MLLMs). We construct a bilingual benchmark using commercial MLLMs to generate training and testing data, and refine the test set through rigorous human revision for systematic evaluation of explainable cross-view reasoning. Together, GLEAM-C and GLEAM-X form a comprehensive CVGL pipeline that integrates multi-modal, multi-view alignment with interpretable correspondence analysis, unifying accurate cross-view matching with explainable reasoning and advancing Geo-Localization by enabling models to better Explain And Match. Code and datasets used in this work will be made publicly accessible at https://github.com/Lucky-Lance/GLEAM.

cs.CV↗

arXiv2026

SofT-GRPO: Surpassing Discrete-Token LLM Reinforcement Learning via Gumbel-Reparameterized Soft-Thinking Policy Optimization

The soft-thinking paradigm for Large Language Model (LLM) reasoning can outperform the conventional discrete-token Chain-of-Thought (CoT) reasoning in some scenarios, underscoring its research and application value. However, while the discrete-token CoT reasoning pattern can be reinforced through policy optimization algorithms such as group relative policy optimization (GRPO), extending the soft-thinking pattern with Reinforcement Learning (RL) remains challenging. This difficulty stems from the complexities of injecting stochasticity into soft-thinking tokens and updating soft-thinking policies accordingly. As a result, previous attempts to combine soft-thinking with GRPO typically underperform their discrete-token GRPO counterparts. To fully unlock the potential of soft-thinking, this paper presents a novel policy optimization algorithm, SofT-GRPO, to reinforce LLMs under the soft-thinking reasoning pattern. SofT-GRPO injects the Gumbel noise into logits, employs the Gumbel-Softmax technique to avoid soft-thinking tokens outside the pre-trained embedding space, and leverages the reparameterization trick in policy gradient. We conduct experiments across base LLMs ranging from 1.5B to 7B parameters, and results demonstrate that SofT-GRPO enables soft-thinking LLMs to slightly outperform discrete-token GRPO on Pass@1 (+0.13% on average accuracy), while exhibiting a substantial uplift on Pass@32 (+2.19% on average accuracy). Codes and weights are available on https://github.com/zz1358m/SofT-GRPO-master

cs.AI↗

arXiv2026

Beyond Imitation: Reinforcement Learning for Active Latent Planning

Aiming at efficient and dense chain-of-thought (CoT) reasoning, latent reasoning methods fine-tune Large Language Models (LLMs) to substitute discrete language tokens with continuous latent tokens. These methods consume fewer tokens compared to the conventional language CoT reasoning and have the potential to plan in a dense latent space. However, current latent tokens are generally supervised based on imitating language labels. Considering that there can be multiple equivalent but diverse CoT labels for a question, passively imitating an arbitrary one may lead to inferior latent token representations and latent reasoning policies, undermining the potential planning ability and resulting in clear gaps between training and testing. In this work, we emphasize the importance of active planning over the representation space of latent tokens in achieving the optimal latent reasoning policy. So, we propose the \underline{A}c\underline{t}ive Latent \underline{P}lanning method (ATP-Latent), which models the supervision process of latent tokens as a conditional variational auto-encoder (VAE) to obtain a smoother latent space. Moreover, to facilitate the most reasonable latent reasoning policy, ATP-Latent conducts reinforcement learning (RL) with an auxiliary coherence reward, which is calculated based on the consistency between VAE-decoded contents of latent tokens, enabling a guided RL process. In experiments on LLaMA-1B, ATP-Latent demonstrates +4.1\% accuracy and -3.3\% tokens on four benchmarks compared to advanced baselines. Codes are available on https://github.com/zz1358m/ATP-Latent-master.

cs.AI↗

arXiv2026

Token-level Collaborative Alignment for LLM-based Generative Recommendation

Large Language Models (LLMs) have demonstrated strong potential for generative recommendation by leveraging rich semantic knowledge. However, existing LLM-based recommender systems struggle to effectively incorporate collaborative filtering (CF) signals, due to a fundamental mismatch between item-level preference modeling in CF and token-level next-token prediction (NTP) optimization in LLMs. Prior approaches typically treat CF as contextual hints or representation bias, and resort to multi-stage training to reduce behavioral semantic space discrepancies, leaving CF unable to explicitly regulate LLM generation. In this work, we propose Token-level Collaborative Alignment for Recommendation (TCA4Rec), a model-agnostic and plug-and-play framework that establishes an explicit optimization-level interface between CF supervision and LLM generation. TCA4Rec consists of (i) Collaborative Tokenizer, which projects raw item-level CF logits into token-level distributions aligned with the LLM token space, and (ii) Soft Label Alignment, which integrates these CF-informed distributions with one-hot supervision to optimize a soft NTP objective. This design preserves the generative nature of LLM training while enabling collaborative alignment with essential user preference of CF models. We highlight TCA4Rec is compatible with arbitrary traditional CF models and generalizes across a wide range of decoder-based LLM recommender architectures. Moreover, it provides an explicit mechanism to balance behavioral alignment and semantic fluency, yielding generative recommendations that are both accurate and controllable. Extensive experiments demonstrate that TCA4Rec consistently improves recommendation performance across a broad spectrum of CF models and LLM-based recommender systems.

cs.IR↗

arXiv2026

Self-Manager: Parallel Agent Loop for Long-form Deep Research

Long-form deep research requires multi-faceted investigations over extended horizons to get a comprehensive report. When handling such complex tasks, existing agents manage context at the subtask level to overcome linear context accumulation and information loss. However, they still adhere to a single context window and sequential execution paradigm, which results in mutual interference and blocking behavior, restricting scalability and adaptability. To address this issue, this paper introduces Self-Manager, a parallel agent loop that enables asynchronous and concurrent execution. The main thread can create multiple subthreads, each with its own isolated context, and manage them iteratively through Thread Control Blocks, allowing for more focused and flexible parallel agent execution. To assess its effectiveness, we benchmark Self-Manager on DeepResearch Bench, where it consistently outperforms existing single-agent loop baselines across all metrics. Furthermore, we conduct extensive analytical experiments to demonstrate the necessity of Self-Manager's design choices, as well as its advantages in contextual capacity, efficiency, and generalization.

cs.CL↗

arXiv2026

VIGIL: Defending LLM Agents Against Tool Stream Injection via Verify-Before-Commit

LLM agents operating in open environments face escalating risks from indirect prompt injection, particularly within the tool stream where manipulated metadata and runtime feedback hijack execution flow. Existing defenses encounter a critical dilemma as advanced models prioritize injected rules due to strict alignment while static protection mechanisms sever the feedback loop required for adaptive reasoning. To reconcile this conflict, we propose \textbf{VIGIL}, a framework that shifts the paradigm from restrictive isolation to a verify-before-commit protocol. By facilitating speculative hypothesis generation and enforcing safety through intent-grounded verification, \textbf{VIGIL} preserves reasoning flexibility while ensuring robust control. We further introduce \textbf{SIREN}, a benchmark comprising 959 tool stream injection cases designed to simulate pervasive threats characterized by dynamic dependencies. Extensive experiments demonstrate that \textbf{VIGIL} outperforms state-of-the-art dynamic defenses by reducing the attack success rate by over 22\% while more than doubling the utility under attack compared to static baselines, thereby achieving an optimal balance between security and utility.

cs.CR↗

arXiv2025

ViewBridge:Revisiting Cross-View Localization from Image Matching

Cross-view localization aims to estimate the 3-DoF pose of a ground-view image by aligning it with aerial or satellite imagery. Existing methods typically address this task through direct regression or feature alignment in a shared bird's-eye view (BEV) space. Although effective for coarse alignment, these methods fail to establish fine-grained and geometrically reliable correspondences under large viewpoint variations, thereby limiting both the accuracy and interpretability of localization results. Consequently, we revisit cross-view localization from the perspective of image matching and propose a unified framework that enhances both matching and localization. Specifically, we introduce a Surface Model that constrains BEV feature projection to physically valid regions for geometric consistency, and a SimRefiner that adaptively refines similarity distributions to enhance match reliability. To further support research in this area, we present CVFM, the first benchmark with 32,509 cross-view image pairs annotated with pixel-level correspondences. Extensive experiments demonstrate that our approach achieves geometry-consistent and fine-grained correspondences across extreme viewpoints and further improves the accuracy and stability of cross-view localization.

cs.CV↗

arXiv2025

Rethinking Neural Combinatorial Optimization for Vehicle Routing Problems with Different Constraint Tightness Degrees

Recent neural combinatorial optimization (NCO) methods have shown promising problem-solving ability without requiring domain-specific expertise. Most existing NCO methods use training and testing data with a fixed constraint value and lack research on the effect of constraint tightness on the performance of NCO methods. This paper takes the capacity-constrained vehicle routing problem (CVRP) as an example to empirically analyze the NCO performance under different tightness degrees of the capacity constraint. Our analysis reveals that existing NCO methods overfit the capacity constraint, and they can only perform satisfactorily on a small range of the constraint values but poorly on other values. To tackle this drawback of existing NCO methods, we develop an efficient training scheme that explicitly considers varying degrees of constraint tightness and proposes a multi-expert module to learn a generally adaptable solving strategy. Experimental results show that the proposed method can effectively overcome the overfitting issue, demonstrating superior performances on the CVRP and CVRP with time windows (CVRPTW) with various constraint tightness degrees.

cs.LG↗

arXiv2025

An Efficient Rubric-based Generative Verifier for Search-Augmented LLMs

Search augmentation empowers Large Language Models with retrieval capabilities to overcome the limitations imposed by static parameters. Recently, Reinforcement Learning leverages tailored reward signals as a viable technique to enhance LLMs performing tasks involving search. However, existing reward modeling for search-augmented LLMs faces several limitations. Rule-based rewards, such as Exact Match, are verifiable but fragile to variations in expression and cannot be applied to long-form workloads. In contrast, generative rewards improve robustness, but designing verifiable and stable rewards for long-form workloads in dynamic corpora remains challenging and also incurs high computational costs. In this paper, we propose a unified and verifiable paradigm, "nugget-as-rubric", which treats atomic information points as structured evaluation criteria for different search-augmentation workloads. Short-form tasks correspond to a single rubric, whereas long-form tasks expand to multiple rubrics aligned with the question's information needs. To support long-form settings, we design an automatic rubric construction pipeline based on query rewriting, which can automatically retrieve passages relevant to each question and extract rubrics from them, both from static corpora and from dynamic online web content. Furthermore, we introduce \textbf{Search-Gen-V}, a 4B-parameter efficient generative verifier under our proposed verifiable paradigm, which is trained via the idea of distillation and a two-stage strategy. Experimental results show that Search-Gen-V achieves strong verification accuracy across different workloads, making it a scalable, robust, and efficient verifiable reward constructor for search-augmented LLMs.

cs.CL↗

arXiv2025

A Survey on Parallel Reasoning

With the increasing capabilities of Large Language Models (LLMs), parallel reasoning has emerged as a new inference paradigm that enhances reasoning robustness by concurrently exploring multiple lines of thought before converging on a final answer. It has become a significant trend to explore parallel reasoning to overcome the fragility of standard sequential methods and improve practical performance. In this paper, we aim to survey and summarize the progress and challenges of parallel reasoning. We first present a formal definition of parallel reasoning and clarify its distinction from related concepts like Chain-of-Thought. Then, we organize and discuss advanced techniques based on a novel taxonomy, including non-interactive reasoning, interactive reasoning, and efficiency-focused decoding strategies. Additionally, we explore various application scenarios, such as solving complex problems and enhancing the reliability of LLM outputs.Finally, we highlight the core challenges of parallel reasoning and suggest potential directions for future research. We hope that our work can provide a useful roadmap for beginners and encourage more research on improving parallel reasoning methods. Related source can be avaliable in https://github.com/PPPP-kaqiu/Awesome-Parallel-Reasoning.

cs.CL↗

arXiv2025

Diffusion-Based mmWave Radar Point Cloud Enhancement Driven by Range Images

Millimeter-wave (mmWave) radar has attracted significant attention in robotics and autonomous driving. However, despite the perception stability in harsh environments, the point cloud generated by mmWave radar is relatively sparse while containing significant noise, which limits its further development. Traditional mmWave radar enhancement approaches often struggle to leverage the effectiveness of diffusion models in super-resolution, largely due to the unnatural range-azimuth heatmap (RAH) or bird's eye view (BEV) representation. To overcome this limitation, we propose a novel method that pioneers the application of fusing range images with image diffusion models, achieving accurate and dense mmWave radar point clouds that are similar to LiDAR. Benefitting from the projection that aligns with human observation, the range image representation of mmWave radar is close to natural images, allowing the knowledge from pre-trained image diffusion models to be effectively transferred, significantly improving the overall performance. Extensive evaluations on both public datasets and self-constructed datasets demonstrate that our approach provides substantial improvements, establishing a new state-of-the-art performance in generating truly three-dimensional LiDAR-like point clouds via mmWave radar. Code will be released after publication.

cs.RO↗

arXiv2025

A2R: An Asymmetric Two-Stage Reasoning Framework for Parallel Reasoning

Recent Large Reasoning Models have achieved significant improvements in complex task-solving capabilities by allocating more computation at the inference stage with a "thinking longer" paradigm. Even as the foundational reasoning capabilities of models advance rapidly, the persistent gap between a model's performance in a single attempt and its latent potential, often revealed only across multiple solution paths, starkly highlights the disparity between its realized and inherent capabilities. To address this, we present A2R, an Asymmetric Two-Stage Reasoning framework designed to explicitly bridge the gap between a model's potential and its actual performance. In this framework, an "explorer" model first generates potential solutions in parallel through repeated sampling. Subsequently,a "synthesizer" model integrates these references for a more refined, second stage of reasoning. This two-stage process allows computation to be scaled orthogonally to existing sequential methods. Our work makes two key innovations: First, we present A2R as a plug-and-play parallel reasoning framework that explicitly enhances a model's capabilities on complex questions. For example, using our framework, the Qwen3-8B-distill model achieves a 75% performance improvement compared to its self-consistency baseline. Second, through a systematic analysis of the explorer and synthesizer roles, we identify an effective asymmetric scaling paradigm. This insight leads to A2R-Efficient, a "small-to-big" variant that combines a Qwen3-4B explorer with a Qwen3-8B synthesizer. This configuration surpasses the average performance of a monolithic Qwen3-32B model at a nearly 30% lower cost. Collectively, these results show that A2R is not only a performance-boosting framework but also an efficient and practical solution for real-world applications.

cs.AI↗

arXiv2025

PerchMobi^3: A Multi-Modal Robot with Power-Reuse Quad-Fan Mechanism for Air-Ground-Wall Locomotion

Achieving seamless integration of aerial flight, ground driving, and wall climbing within a single robotic platform remains a major challenge, as existing designs often rely on additional adhesion actuators that increase complexity, reduce efficiency, and compromise reliability. To address these limitations, we present PerchMobi^3, a quad-fan, negative-pressure, air-ground-wall robot that implements a propulsion-adhesion power-reuse mechanism. By repurposing four ducted fans to simultaneously provide aerial thrust and negative-pressure adhesion, and integrating them with four actively driven wheels, PerchMobi^3 eliminates dedicated pumps while maintaining a lightweight and compact design. To the best of our knowledge, this is the first quad-fan prototype to demonstrate functional power reuse for multi-modal locomotion. A modeling and control framework enables coordinated operation across ground, wall, and aerial domains with fan-assisted transitions. The feasibility of the design is validated through a comprehensive set of experiments covering ground driving, payload-assisted wall climbing, aerial flight, and cross-mode transitions, demonstrating robust adaptability across locomotion scenarios. These results highlight the potential of PerchMobi^3 as a novel design paradigm for multi-modal robotic mobility, paving the way for future extensions toward autonomous and application-oriented deployment.

cs.RO↗

arXiv2025

Don't Trade Off Safety: Diffusion Regularization for Constrained Offline RL

Constrained reinforcement learning (RL) seeks high-performance policies under safety constraints. We focus on an offline setting where the agent has only a fixed dataset -- common in realistic tasks to prevent unsafe exploration. To address this, we propose Diffusion-Regularized Constrained Offline Reinforcement Learning (DRCORL), which first uses a diffusion model to capture the behavioral policy from offline data and then extracts a simplified policy to enable efficient inference. We further apply gradient manipulation for safety adaptation, balancing the reward objective and constraint satisfaction. This approach leverages high-quality offline data while incorporating safety requirements. Empirical results show that DRCORL achieves reliable safety performance, fast inference, and strong reward outcomes across robot learning tasks. Compared to existing safe offline RL methods, it consistently meets cost limits and performs well with the same hyperparameters, indicating practical applicability in real-world scenarios.

cs.LG↗

arXiv2025

MiniCPM4: Ultra-Efficient LLMs on End Devices

This paper introduces MiniCPM4, a highly efficient large language model (LLM) designed explicitly for end-side devices. We achieve this efficiency through systematic innovation in four key dimensions: model architecture, training data, training algorithms, and inference systems. Specifically, in terms of model architecture, we propose InfLLM v2, a trainable sparse attention mechanism that accelerates both prefilling and decoding phases for long-context processing. Regarding training data, we propose UltraClean, an efficient and accurate pre-training data filtering and generation strategy, and UltraChat v2, a comprehensive supervised fine-tuning dataset. These datasets enable satisfactory model performance to be achieved using just 8 trillion training tokens. Regarding training algorithms, we propose ModelTunnel v2 for efficient pre-training strategy search, and improve existing post-training methods by introducing chunk-wise rollout for load-balanced reinforcement learning and data-efficient tenary LLM, BitCPM. Regarding inference systems, we propose CPM.cu that integrates sparse attention, model quantization, and speculative sampling to achieve efficient prefilling and decoding. To meet diverse on-device requirements, MiniCPM4 is available in two versions, with 0.5B and 8B parameters, respectively. Furthermore, we construct a hybrid reasoning model, MiniCPM4.1, which can be used in both deep reasoning mode and non-reasoning mode. Evaluation results demonstrate that MiniCPM4 and MiniCPM4.1 outperform similar-sized open-source models across benchmarks, with the 8B variants showing significant speed improvements on long sequence understanding and generation.

cs.CL↗

arXiv2025

TextOnly: A Unified Function Portal for Text-Related Functions on Smartphones

Text boxes serve as portals to diverse functionalities in today's smartphone applications. However, when it comes to specific functionalities, users always need to navigate through multiple steps to access particular text boxes for input. We propose TextOnly, a unified function portal that enables users to access text-related functions from various applications by simply inputting text into a sole text box. For instance, entering a restaurant name could trigger a Google Maps search, while a greeting could initiate a conversation in WhatsApp. Despite their brevity, TextOnly maximizes the utilization of these raw text inputs, which contain rich information, to interpret user intentions effectively. TextOnly integrates large language models(LLM) and a BERT model. The LLM consistently provides general knowledge, while the BERT model can continuously learn user-specific preferences and enable quicker predictions. Real-world user studies demonstrated TextOnly's effectiveness with a top-1 accuracy of 71.35%, and its ability to continuously improve both its accuracy and inference speed. Participants perceived TextOnly as having satisfactory usability and expressed a preference for TextOnly over manual executions. Compared with voice assistants, TextOnly supports a greater range of text-related functions and allows for more concise inputs.

cs.HC↗

arXiv2025

RAVine: Reality-Aligned Evaluation for Agentic Search

Agentic search, as a more autonomous and adaptive paradigm of retrieval augmentation, is driving the evolution of intelligent search systems. However, existing evaluation frameworks fail to align well with the goals of agentic search. First, the complex queries commonly used in current benchmarks often deviate from realistic user search scenarios. Second, prior approaches tend to introduce noise when extracting ground truth for end-to-end evaluations, leading to distorted assessments at a fine-grained level. Third, most current frameworks focus solely on the quality of final answers, neglecting the evaluation of the iterative process inherent to agentic search. To address these limitations, we propose RAVine -- a Reality-Aligned eValuation framework for agentic LLMs with search. RAVine targets multi-point queries and long-form answers that better reflect user intents, and introduces an attributable ground truth construction strategy to enhance the accuracy of fine-grained evaluation. Moreover, RAVine examines model's interaction with search tools throughout the iterative process, and accounts for factors of efficiency. We benchmark a series of models using RAVine and derive several insights, which we hope will contribute to advancing the development of agentic search systems. The code and datasets are available at https://github.com/SwordFaith/RAVine.

cs.CL↗

arXiv2025

Temporal-Aware Graph Attention Network for Cryptocurrency Transaction Fraud Detection

Cryptocurrency transaction fraud detection faces the dual challenges of increasingly complex transaction patterns and severe class imbalance. Traditional methods rely on manual feature engineering and struggle to capture temporal and structural dependencies in transaction networks. This paper proposes an Augmented Temporal-aware Graph Attention Network (ATGAT) that enhances detection performance through three modules: (1) designing an advanced temporal embedding module that fuses multi-scale time difference features with periodic position encoding; (2) constructing a temporal-aware triple attention mechanism that jointly optimizes structural, temporal, and global context attention; (3) employing weighted BCE loss to address class imbalance. Experiments on the Elliptic++ cryptocurrency dataset demonstrate that ATGAT achieves an AUC of 0.9130, representing a 9.2% improvement over the best traditional method XGBoost, 12.0% over GCN, and 10.0% over standard GAT. This method not only validates the enhancement effect of temporal awareness and triple attention mechanisms on graph neural networks, but also provides financial institutions with more reliable fraud detection tools, with its design principles generalizable to other temporal graph anomaly detection tasks.

cs.LG↗