Search arXiv⌕ Search

arXiv subjects

Xi Chen

Publications and source records attributed to Xi Chen.

At least 343 records · Page 19Linked to original sources
arXiv2025

Conan-Embedding-v2: Training an LLM from Scratch for Text Embeddings

Large language models (LLMs) have recently demonstrated excellent performance in text embedding tasks. Previous work usually use LoRA to fine-tune existing LLMs, which are limited by the data and training gap between LLMs and embedding models. In this work, we introduce Conan-embedding-v2, a new 1.4B-parameter LLM trained from scratch and fine-tuned as a text embedder. First, we add news data and multilingual pairs for LLM pretraining to bridge the data gap. Based on this, we propose a cross-lingual retrieval dataset that enables the LLM to better integrate embeddings across different languages. Second, whereas LLMs use a causal mask with token-level loss, embedding models use a bidirectional mask with sentence-level loss. This training gap makes full fine-tuning less effective than LoRA. We introduce a soft-masking mechanism to gradually transition between these two types of masks, enabling the model to learn more comprehensive representations. Based on this, we propose a dynamic hard negative mining method that exposes the model to more difficult negative examples throughout the training process. Being intuitive and effective, with only approximately 1.4B parameters, Conan-embedding-v2 achieves SOTA performance on both the Massive Text Embedding Benchmark (MTEB) and Chinese MTEB (May 19, 2025).

cs.CL↗
arXiv2025

FastMTP: Accelerating LLM Inference with Enhanced Multi-Token Prediction

As large language models (LLMs) become increasingly powerful, the sequential nature of autoregressive generation creates a fundamental throughput bottleneck that limits the practical deployment. While Multi-Token Prediction (MTP) has demonstrated remarkable benefits for model training efficiency and performance, its inherent potential for inference acceleration remains largely unexplored. This paper introduces FastMTP, a simple yet effective method that improves multi-step draft quality by aligning MTP training with its inference pattern, significantly enhancing speculative decoding performance. Our approach fine-tunes a single MTP head with position-shared weights on self-distilled data, enabling it to capture dependencies among consecutive future tokens and maintain high acceptance rates across multiple recursive draft steps. By integrating language-aware dynamic vocabulary compression into the MTP head, we further reduce computational overhead in the drafting process. Experimental results across seven diverse benchmarks demonstrate that FastMTP achieves an average of 2.03x speedup compared to standard next token prediction with lossless output quality, outperforming vanilla MTP by 82%. FastMTP requires only lightweight training and seamlessly integrates with existing inference frameworks, offering a practical and rapidly deployable solution for accelerating LLM inference.

cs.LG↗
arXiv2025

Stencil-Lifting: Hierarchical Recursive Lifting System for Extracting Summary of Stencil Kernel in Legacy Codes

We introduce Stencil-Lifting, a novel system for automatically converting stencil kernels written in low-level languages in legacy code into semantically equivalent Domain-Specific Language (DSL) implementations. Targeting the efficiency bottlenecks of existing verified lifting systems, Stencil-Lifting achieves scalable stencil kernel abstraction through two key innovations. First, we propose a hierarchical recursive lifting theory that represents stencil kernels, structured as nested loops, using invariant subgraphs, which are customized data dependency graphs that capture loop-carried computation and structural invariants. Each vertex in the invariant subgraph is associated with a predicate-based summary, encoding its computational semantics. By enforcing self-consistency across these summaries, Stencil-Lifting ensures the derivation of correct loop invariants and postconditions for nested loops, eliminating the need for external verification. Second, we develop a hierarchical recursive lifting algorithm that guarantees termination through a convergent recursive process, avoiding the inefficiencies of search-based synthesis. The algorithm efficiently derives the valid summaries of stencil kernels, and its completeness is formally proven. We evaluate Stencil-Lifting on diverse stencil benchmarks from two different suites and on four real-world applications. Experimental results demonstrate that Stencil-Lifting achieves 31.62$\times$ and 5.8$\times$ speedups compared to the state-of-the-art verified lifting systems STNG and Dexter, respectively, while maintaining full semantic equivalence. Our work significantly enhances the translation efficiency of low-level stencil kernels to DSL implementations, effectively bridging the gap between legacy optimization techniques and modern DSL-based paradigms.

cs.SE↗
arXiv2025

Quantum control of exciton motion in electric field

We study quantum control of classical motion of a two-dimensional exciton by optimizing the time-dependent electric field of a stripe-like gate acting on the exciton and inducing its time-dependent quantum dipole moment. We propose a search method that significantly reduces computational requirements while efficiently identifying optimal control parameters. By leveraging this method, one can precisely manipulate the exciton's final position and velocity over a specified evolution time. These results can be applied for control of exciton fluxes and population, and for spatially resolved light emission in two-dimensional semiconducting structures.

cond-mat.mes-hall↗
arXiv2025

Analysis of the Dick Effect for AI-based Dynamic Gravimeter

Atom interferometer (AI)-based dynamic gravimeter enable high-precision absolute gravity measurements, crucial for applications in geophysics, navigation, resource exploration, and metrology. Understanding their underlying mechanisms and minimizing measurement noise are essential for enhancing performance. This work investigates the gravity measurement noise in AI-based systems induced by the dead time of the classical accelerometer. Using actual dynamic gravity measurement data, we demonstrate that a dead time of 0.12 s introduces significant gravity measurement noise of 8 mGal. To elucidate the mechanism of this noise, we derive a formula for this noise in frequency domain, identifying high-frequency aliasing as its source. Analysis of the derived expressions indicates that reducing the dead time duration and suppressing the high-frequency noise of the acceleration are effective strategies for mitigating this noise. This work provides significant insights for noise analysis and future scheme design of AI-based dynamic gravimeters.

physics.atom-ph↗
arXiv2025

Relative-error monotonicity testing

The standard model of Boolean function property testing is not well suited for testing $\textit{sparse}$ functions which have few satisfying assignments, since every such function is close (in the usual Hamming distance metric) to the constant-0 function. In this work we propose and investigate a new model for property testing of Boolean functions, called $\textit{relative-error testing}$, which provides a natural framework for testing sparse functions. This new model defines the distance between two functions $f, g: \{0,1\}^n \to \{0,1\}$ to be $$\textsf{reldist}(f,g) := { \frac{|f^{-1}(1) \triangle g^{-1}(1)|} {|f^{-1}(1)|}}.$$ This is a more demanding distance measure than the usual Hamming distance ${ {|f^{-1}(1) \triangle g^{-1}(1)|}/{2^n}}$ when $|f^{-1}(1)| \ll 2^n$; to compensate for this, algorithms in the new model have access both to a black-box oracle for the function $f$ being tested and to a source of independent uniform satisfying assignments of $f$. In this paper we first give a few general results about the relative-error testing model; then, as our main technical contribution, we give a detailed study of algorithms and lower bounds for relative-error testing of $\textit{monotone}$ Boolean functions. We give upper and lower bounds which are parameterized by $N=|f^{-1}(1)|$, the sparsity of the function $f$ being tested. Our results show that there are interesting differences between relative-error monotonicity testing of sparse Boolean functions, and monotonicity testing in the standard model. These results motivate further study of the testability of Boolean function properties in the relative-error model.

cs.CC↗
arXiv2025

Fully-chromophoric ferroelectric nematics for electronic electro-optics

Electronic electro-optic (EEO) phase modulation is a key emerging technology for the chip-scale inter-conversion of signals between the electronic and photonic domains. The recent discovery of the ferroelectric nematic ($N_F$) liquid crystal phase, a three dimensional fluid of rod-shaped organic molecules having near-perfect equilibrium polar molecular orientational order, offers attractive opportunities for the creation of second-order nonlinear optical materials for EEO. Here we propose and realize a design motif for $N_F$ EEO molecules in which few-nanometer-long molecular rods are functionalized both for electro-static end-to-end association, facilitating $N_F$ phase formation, and for chromophoric optical nonlinearity, enabling high EEO efficiency, a combination enabling an active second-order nonlinear EEO medium that is 100% chromophoric.

cond-mat.soft↗
arXiv2025

ROSE: Remove Objects with Side Effects in Videos

Video object removal has achieved advanced performance due to the recent success of video generative models. However, when addressing the side effects of objects, e.g., their shadows and reflections, existing works struggle to eliminate these effects for the scarcity of paired video data as supervision. This paper presents ROSE, termed Remove Objects with Side Effects, a framework that systematically studies the object's effects on environment, which can be categorized into five common cases: shadows, reflections, light, translucency and mirror. Given the challenges of curating paired videos exhibiting the aforementioned effects, we leverage a 3D rendering engine for synthetic data generation. We carefully construct a fully-automatic pipeline for data preparation, which simulates a large-scale paired dataset with diverse scenes, objects, shooting angles, and camera trajectories. ROSE is implemented as an video inpainting model built on diffusion transformer. To localize all object-correlated areas, the entire video is fed into the model for reference-based erasing. Moreover, additional supervision is introduced to explicitly predict the areas affected by side effects, which can be revealed through the differential mask between the paired videos. To fully investigate the model performance on various side effect removal, we presents a new benchmark, dubbed ROSE-Bench, incorporating both common scenarios and the five special side effects for comprehensive evaluation. Experimental results demonstrate that ROSE achieves superior performance compared to existing video object erasing models and generalizes well to real-world video scenarios. The project page is https://rose2025-inpaint.github.io/.

cs.CV↗
arXiv2025

Mimicking the Physicist's Eye:A VLM-centric Approach for Physics Formula Discovery

Automated discovery of physical laws from observational data in the real world is a grand challenge in AI. Current methods, relying on symbolic regression or LLMs, are limited to uni-modal data and overlook the rich, visual phenomenological representations of motion that are indispensable to physicists. This "sensory deprivation" severely weakens their ability to interpret the inherent spatio-temporal patterns within dynamic phenomena. To address this gap, we propose VIPER-R1, a multimodal model that performs Visual Induction for Physics-based Equation Reasoning to discover fundamental symbolic formulas. It integrates visual perception, trajectory data, and symbolic reasoning to emulate the scientific discovery process. The model is trained via a curriculum of Motion Structure Induction (MSI), using supervised fine-tuning to interpret kinematic phase portraits and to construct hypotheses guided by a Causal Chain of Thought (C-CoT), followed by Reward-Guided Symbolic Calibration (RGSC) to refine the formula structure with reinforcement learning. During inference, the trained VIPER-R1 acts as an agent: it first posits a high-confidence symbolic ansatz, then proactively invokes an external symbolic regression tool to perform Symbolic Residual Realignment (SR^2). This final step, analogous to a physicist's perturbation analysis, reconciles the theoretical model with empirical data. To support this research, we introduce PhysSymbol, a new 5,000-instance multimodal corpus. Experiments show that VIPER-R1 consistently outperforms state-of-the-art VLM baselines in accuracy and interpretability, enabling more precise discovery of physical laws. Project page: https://jiaaqiliu.github.io/VIPER-R1/

cs.AI↗
arXiv2025

Pragmatic Inference Chain (PIC) Improving LLMs' Reasoning of Authentic Implicit Toxic Language

The rapid development of large language models (LLMs) gives rise to ethical concerns about their performance, while opening new avenues for developing toxic language detection techniques. However, LLMs' unethical output and their capability of detecting toxicity have primarily been tested on language data that do not demand complex meaning inference, such as the biased associations of 'he' with programmer and 'she' with household. Nowadays toxic language adopts a much more creative range of implicit forms, thanks to advanced censorship. In this study, we collect authentic toxic interactions that evade online censorship and that are verified by human annotators as inference-intensive. To evaluate and improve LLMs' reasoning of the authentic implicit toxic language, we propose a new prompting method, Pragmatic Inference Chain (PIC), drawn on interdisciplinary findings from cognitive science and linguistics. The PIC prompting significantly improves the success rate of GPT-4o, Llama-3.1-70B-Instruct, DeepSeek-v2.5, and DeepSeek-v3 in identifying implicit toxic language, compared to five baseline prompts, such as CoT and rule-based baselines. In addition, it also facilitates the models to produce more explicit and coherent reasoning processes, hence can potentially be generalized to other inference-intensive tasks, e.g., understanding humour and metaphors.

cs.CL↗
arXiv2025

It Takes Two: A Peer-Prediction Solution for Blockchain Verifier's Dilemma

The security of blockchain systems is fundamentally based on the decentralized consensus in which the majority of parties behave honestly, and the content verification process is essential to maintaining the robustness of blockchain systems. However, the phenomenon that a rational verifier may not have the incentive to honestly perform the costly verification, referred to as the Verifier's Dilemma, could incentivize lazy reporting and undermine the fundamental security of blockchain systems, particularly for verification-expensive decentralized AI applications. In this paper, we initiate the research with the development of a Byzantine-robust peer prediction framework towards the design of one-phase Bayesian truthful mechanisms for the decentralized verification games among multiple verifiers, incentivizing all verifiers to perform honest verification without access to the ground truth even in the presence of noisy observations, malicious players and inaccurate priors in the verification process, proposing the compactness criteria that ensures such robustness guarantees. With robust incentive guarantees and budget efficiency, our study provides a framework of incentive design for decentralized verification protocols that enhances the security and robustness of the blockchain, decentralized AI, and potentially other decentralized systems.

cs.CR↗
arXiv2025

TBAC-UniImage: Unified Understanding and Generation by Ladder-Side Diffusion Tuning

This paper introduces TBAC-UniImage, a novel unified model for multimodal understanding and generation. We achieve this by deeply integrating a pre-trained Diffusion Model, acting as a generative ladder, with a Multimodal Large Language Model (MLLM). Previous diffusion-based unified models face two primary limitations. One approach uses only the MLLM's final hidden state as the generative condition. This creates a shallow connection, as the generator is isolated from the rich, hierarchical representations within the MLLM's intermediate layers. The other approach, pretraining a unified generative architecture from scratch, is computationally expensive and prohibitive for many researchers. To overcome these issues, our work explores a new paradigm. Instead of relying on a single output, we use representations from multiple, diverse layers of the MLLM as generative conditions for the diffusion model. This method treats the pre-trained generator as a ladder, receiving guidance from various depths of the MLLM's understanding process. Consequently, TBAC-UniImage achieves a much deeper and more fine-grained unification of understanding and generation.

cs.CV↗
arXiv2025

ChineseHarm-Bench: A Chinese Harmful Content Detection Benchmark

Large language models (LLMs) have been increasingly applied to automated harmful content detection tasks, assisting moderators in identifying policy violations and improving the overall efficiency and accuracy of content review. However, existing resources for harmful content detection are predominantly focused on English, with Chinese datasets remaining scarce and often limited in scope. We present a comprehensive, professionally annotated benchmark for Chinese content harm detection, which covers six representative categories and is constructed entirely from real-world data. Our annotation process further yields a knowledge rule base that provides explicit expert knowledge to assist LLMs in Chinese harmful content detection. In addition, we propose a knowledge-augmented baseline that integrates both human-annotated knowledge rules and implicit knowledge from large language models, enabling smaller models to achieve performance comparable to state-of-the-art LLMs. Code and data are available at https://github.com/zjunlp/ChineseHarm-bench.

cs.CL↗
arXiv2025

Animate-X++: Universal Character Image Animation with Dynamic Backgrounds

Character image animation, which generates high-quality videos from a reference image and target pose sequence, has seen significant progress in recent years. However, most existing methods only apply to human figures, which usually do not generalize well on anthropomorphic characters commonly used in industries like gaming and entertainment. Furthermore, previous methods could only generate videos with static backgrounds, which limits the realism of the videos. For the first challenge, our in-depth analysis suggests to attribute this limitation to their insufficient modeling of motion, which is unable to comprehend the movement pattern of the driving video, thus imposing a pose sequence rigidly onto the target character. To this end, this paper proposes Animate-X++, a universal animation framework based on DiT for various character types, including anthropomorphic characters. To enhance motion representation, we introduce the Pose Indicator, which captures comprehensive motion pattern from the driving video through both implicit and explicit manner. The former leverages CLIP visual features of a driving video to extract its gist of motion, like the overall movement pattern and temporal relations among motions, while the latter strengthens the generalization of DiT by simulating possible inputs in advance that may arise during inference. For the second challenge, we introduce a multi-task training strategy that jointly trains the animation and TI2V tasks. Combined with the proposed partial parameter training, this approach achieves not only character animation but also text-driven background dynamics, making the videos more realistic. Moreover, we introduce a new Animated Anthropomorphic Benchmark (A2Bench) to evaluate the performance of Animate-X++ on universal and widely applicable animation images. Extensive experiments demonstrate the superiority and effectiveness of Animate-X++.

cs.CV↗
arXiv2025

SASST: Leveraging Syntax-Aware Chunking and LLMs for Simultaneous Speech Translation

This work proposes a grammar-based chunking strategy that segments input streams into semantically complete units by parsing dependency relations (e.g., noun phrase boundaries, verb-object structures) and punctuation features. The method ensures chunk coherence and minimizes semantic fragmentation. Building on this mechanism, we present SASST (Syntax-Aware Simultaneous Speech Translation), an end-to-end framework integrating frozen Whisper encoder and decoder-only LLM. The unified architecture dynamically outputs translation tokens or symbols to jointly optimize translation timing and content, with target-side reordering addressing word-order divergence. Experiments on CoVoST2 multilingual corpus En-{De, Zh, Ja} demonstrate significant translation quality improvements across languages and validate the effectiveness of syntactic structures in LLM-driven SimulST systems.

cs.CL↗
arXiv2025

Universal Magnetocaloric Effect near Quantum Critical Point of Magnon Bose-Einstein Condensation

Bose-Einstein condensation (BEC), a macroscopic quantum phenomenon arising from phase coherence and bosonic statistics, has been realized in quantum magnets. Here, we report the observation of a universal magnetocaloric effect (MCE) near a BEC quantum critical point (QCP) in copper sulfate crystal ($CuSO_4 \cdot 5H_2O$). By conducting magnetocaloric and nuclear magnetic resonance measurements, we uncover a field-driven BEC QCP, evidenced by the universal scaling law $T_c \propto (B_c - B)^{2/3}$ and the perfect data collapse of the magnetic Grüneisen ratio. Thermal excitation triggers a dimensional crossover to a 1D quantum-critical regime, where the MCE scaling strictly matches the universality class of 1D Fermi gases. Notably, the quantum-critical MCE enables cooling down to 12.8 mK without helium-3, with very fast thermal relaxation rate that is critical for high cooling power. This work demonstrates the universal MCE in magnon BEC systems, using a common copper sulfate compound as a paradigmatic example, and paves the way for next-generation sub-Kelvin cooling.

cond-mat.str-el↗
arXiv2025

Dynamic Knowledge Exchange and Dual-diversity Review: Concisely Unleashing the Potential of a Multi-Agent Research Team

Scientific progress increasingly relies on effective collaboration among researchers, a dynamic that large language models (LLMs) have only begun to emulate. While recent LLM-based scientist agents show promise in autonomous scientific discovery, they often lack the interactive reasoning and evaluation mechanisms essential to real-world research. We propose IDVSCI (Internal Discussion and Vote SCIentists), a multi-agent framework built on LLMs that incorporates two key innovations: a Dynamic Knowledge Exchange mechanism enabling iterative feedback among agents, and a Dual-Diversity Review paradigm that simulates heterogeneous expert evaluation. These components jointly promote deeper reasoning and the generation of more creative and impactful scientific ideas. To evaluate the effectiveness and generalizability of our approach, we conduct experiments on two datasets: a widely used benchmark in computer science and a new dataset we introduce in the health sciences domain. Results show that IDVSCI consistently achieves the best performance across both datasets, outperforming existing systems such as AI Scientist and VIRSCI. These findings highlight the value of modeling interaction and peer review dynamics in LLM-based autonomous research.

cs.AI↗