arXiv subjects

Zhiqi Huang

Publications and source records attributed to Zhiqi Huang.

At least 37 records · Page 2Linked to original sources

arXiv2026

Testing the cosmological principle on gigaparsec scales

Recent observational analyses have suggested possible evidence of hemisphere asymmetry in cosmological datasets. Parameterizations of this kind place observers in a privileged position-specifically on the plane that divides the two hemispheres. To quantify potential deviations from the cosmological principle without presuming a special location, we develop a stochastic framework that parametrizes departures from statistical homogeneity and isotropy. The near-uniform temperature of the cosmic microwave background indicates that anisotropy is negligible (at the $\lesssim 10^{-5}$ level) on the last scattering surface. This serves as a zero boundary condition, enabling the construction of an orthogonal basis of functions below the recombination redshift. Within this basis, we expand the relative deviation from the Hubble diagram of isotropic models (such as $Λ$CDM or $w_0w_a$CDM) in a hierarchy of increasing resolution. Applying this approach, we test the cosmological principle using Type Ia supernovae, strong lensing time delays, and gravitational-wave standard sirens. For the class of large-scale anisotropies and low-order radial variations described by this framework, the current datasets are found to be consistent with statistical homogeneity and isotropy on gigaparsec scales.

astro-ph.CO↗

arXiv2026

Kimi K2: Open Agentic Intelligence

We introduce Kimi K2, a Mixture-of-Experts (MoE) large language model with 32 billion activated parameters and 1 trillion total parameters. We propose the MuonClip optimizer, which improves upon Muon with a novel QK-clip technique to address training instability while enjoying the advanced token efficiency of Muon. Based on MuonClip, K2 was pre-trained on 15.5 trillion tokens with zero loss spike. During post-training, K2 undergoes a multi-stage post-training process, highlighted by a large-scale agentic data synthesis pipeline and a joint reinforcement learning (RL) stage, where the model improves its capabilities through interactions with real and synthetic environments. Kimi K2 achieves state-of-the-art performance among open-source non-thinking models, with strengths in agentic capabilities. Notably, K2 obtains 66.1 on Tau2-Bench, 76.5 on ACEBench (En), 65.8 on SWE-Bench Verified, and 47.3 on SWE-Bench Multilingual -- surpassing most open and closed-sourced baselines in non-thinking settings. It also exhibits strong capabilities in coding, mathematics, and reasoning tasks, with a score of 53.7 on LiveCodeBench v6, 49.5 on AIME 2025, 75.1 on GPQA-Diamond, and 27.1 on OJBench, all without extended thinking. These results position Kimi K2 as one of the most capable open-source large language models to date, particularly in software engineering and agentic tasks. We release our base and post-trained model checkpoints to facilitate future research and applications of agentic intelligence.

cs.LG↗

arXiv2026

WorldVQA: Measuring Atomic World Knowledge in Multimodal Large Language Models

We introduce WorldVQA, a benchmark designed to evaluate the atomic visual world knowledge of Multimodal Large Language Models (MLLMs). Unlike current evaluations, which often conflate visual knowledge retrieval with reasoning, WorldVQA decouples these capabilities to strictly measure "what the model memorizes." The benchmark assesses the atomic capability of grounding and naming visual entities across a stratified taxonomy, spanning from common head-class objects to long-tail rarities. We expect WorldVQA to serve as a rigorous test for visual factuality, thereby establishing a standard for assessing the encyclopedic breadth and hallucination rates of current and next-generation frontier models.

cs.CV↗

arXiv2026

Towards Pixel-Level VLM Perception via Simple Points Prediction

We present SimpleSeg, a strikingly simple yet highly effective approach to endow Multimodal Large Language Models (MLLMs) with native pixel-level perception. Our method reframes segmentation as a simple sequence generation problem: the model directly predicts sequences of points (textual coordinates) delineating object boundaries, entirely within its language space. To achieve high fidelity, we introduce a two-stage SF$\to$RL training pipeline, where Reinforcement Learning with an IoU-based reward refines the point sequences to accurately match ground-truth contours. We find that the standard MLLM architecture possesses a strong, inherent capacity for low-level perception that can be unlocked without any specialized architecture. On segmentation benchmarks, SimpleSeg achieves performance that is comparable to, and often surpasses, methods relying on complex, task-specific designs. This work lays out that precise spatial understanding can emerge from simple point prediction, challenging the prevailing need for auxiliary components and paving the way for more unified and capable VLMs. Homepage: https://simpleseg.github.io/

cs.CV↗

arXiv2025

Extensible Multi-Granularity Fusion Network and Transferable Curriculum Learning for Aspect-based Sentiment Analysis

Aspect-based Sentiment Analysis (ABSA) aims to determine sentiment polarity toward specific aspects in text. Existing methods enrich semantic and syntactic representations through external knowledge or GNNs, but the growing diversity of linguistic features increases model complexity and lacks a unified, extensible framework. We propose an Extensible Multi-Granularity Fusion Network (EMGF) that integrates dependency syntax, constituent syntax, attention-based semantics, and external knowledge graphs. EMGF employs multi-anchor triplet learning and orthogonal projection to effectively fuse multi-granularity features and strengthen their interactions without additional computational overhead. Furthermore, we introduce the first task-specific curriculum learning framework for text-only ABSA, which assigns difficulty scores using five indicators and trains the model from easy to hard to mimic human learning and improve generalization. Experiments on SemEval 2014, Twitter, and MAMS datasets show that EMGF+CL consistently outperforms state-of-the-art ABSA models.

cs.AI↗

arXiv2025

Reionization optical depth and CMB-BAO tension in punctuated inflation

Within the standard six-parameter Lambda cold dark matter ($Λ$CDM) model, a $2$-$3σ$ tension persists between baryon acoustic oscillation (BAO) measurements from the Dark Energy Spectroscopic Instrument (DESI) and observations of the cosmic microwave background (CMB). Although this tension has often been interpreted as evidence for dynamical dark energy or a sum of neutrino masses below the established minimum, recent studies suggest it may instead originate from an underestimation of the reionization optical depth, particularly when inferred from large-scale CMB polarization. Jhaveri et al. propose that a suppression of large-scale primordial curvature power could partially cancel the contribution of $τ$ to the CMB low-$\ell$ polarization power spectrum, leading to a biased low $τ$ measurement in standard analyses. In this work, we investigate whether punctuated inflation - which generates a suppression of primordial power on large scales through a transient fast-roll phase - can raise the inferred $τ$ value and thereby reconcile the consistency between CMB and BAO. For simple models with step-like features in the inflaton potential, we find that the constraint on $τ$ and the CMB-BAO tension remain nearly identical to those in the standard six-parameter $Λ$CDM model. We provide a physical explanation for this negative result.

astro-ph.CO↗

arXiv2025

Distillation versus Contrastive Learning: How to Train Your Rerankers

Training effective text rerankers is crucial for information retrieval. Two strategies are widely used: contrastive learning (optimizing directly on ground-truth labels) and knowledge distillation (transferring knowledge from a larger reranker). While both have been studied extensively, a clear comparison of their effectiveness for training cross-encoder rerankers under practical conditions is needed. This paper empirically compares these strategies by training rerankers of different sizes (0.5B, 1.5B, 3B, 7B) and architectures (Transformer, Recurrent) using both methods on the same data, with a strong contrastive learning model acting as the distillation teacher. Our results show that knowledge distillation generally yields better in-domain and out-of-domain ranking performance than contrastive learning when distilling from a more performant teacher model. This finding is consistent across student model sizes and architectures. However, distilling from a teacher of the same capacity does not provide the same advantage, particularly for out-of-domain tasks. These findings offer practical guidance for choosing a training strategy based on available teacher models. We recommend using knowledge distillation to train smaller rerankers if a larger, more performant teacher is accessible; in its absence, contrastive learning remains a robust baseline. Our code implementation is made available to facilitate reproducbility.

cs.CL↗

arXiv2025

Uncertainty as Feature Gaps: Epistemic Uncertainty Quantification of LLMs in Contextual Question-Answering

Uncertainty Quantification (UQ) research has primarily focused on closed-book factual question answering (QA), while contextual QA remains unexplored, despite its importance in real-world applications. In this work, we focus on UQ for the contextual QA task and propose a theoretically grounded approach to quantify epistemic uncertainty. We begin by introducing a task-agnostic, token-level uncertainty measure defined as the cross-entropy between the predictive distribution of the given model and the unknown true distribution. By decomposing this measure, we isolate the epistemic component and approximate the true distribution by a perfectly prompted, idealized model. We then derive an upper bound for epistemic uncertainty and show that it can be interpreted as semantic feature gaps in the given model's hidden representations relative to the ideal model. We further apply this generic framework to the contextual QA task and hypothesize that three features approximate this gap: context-reliance (using the provided context rather than parametric knowledge), context comprehension (extracting relevant information from context), and honesty (avoiding intentional lies). Using a top-down interpretability approach, we extract these features by using only a small number of labeled samples and ensemble them to form a robust uncertainty score. Experiments on multiple QA benchmarks in both in-distribution and out-of-distribution settings show that our method substantially outperforms state-of-the-art unsupervised (sampling-free and sampling-based) and supervised UQ methods, achieving up to a 13-point PRR improvement while incurring a negligible inference overhead.

cs.CL↗

arXiv2025

Confidence-Based Response Abstinence: Improving LLM Trustworthiness via Activation-Based Uncertainty Estimation

We propose a method for confidence estimation in retrieval-augmented generation (RAG) systems that aligns closely with the correctness of large language model (LLM) outputs. Confidence estimation is especially critical in high-stakes domains such as finance and healthcare, where the cost of an incorrect answer outweighs that of not answering the question. Our approach extends prior uncertainty quantification methods by leveraging raw feed-forward network (FFN) activations as auto-regressive signals, avoiding the information loss inherent in token logits and probabilities after projection and softmax normalization. We model confidence prediction as a sequence classification task, and regularize training with a Huber loss term to improve robustness against noisy supervision. Applied in a real-world financial industry customer-support setting with complex knowledge bases, our method outperforms strong baselines and maintains high accuracy under strict latency constraints. Experiments on Llama 3.1 8B model show that using activations from only the 16th layer preserves accuracy while reducing response latency. Our results demonstrate that activation-based confidence modeling offers a scalable, architecture-aware path toward trustworthy RAG deployment.

cs.CL↗

arXiv2025

Introduction to the Chinese Space Station Survey Telescope (CSST)

The Chinese Space Station Survey Telescope (CSST) is an upcoming Stage-IV sky survey telescope, distinguished by its large field of view (FoV), high image quality, and multi-band observation capabilities. It can simultaneously conduct precise measurements of the Universe by performing multi-color photometric imaging and slitless spectroscopic surveys. The CSST is equipped with five scientific instruments, i.e. Multi-band Imaging and Slitless Spectroscopy Survey Camera (SC), Multi-Channel Imager (MCI), Integral Field Spectrograph (IFS), Cool Planet Imaging Coronagraph (CPI-C), and THz Spectrometer (TS). Using these instruments, CSST is expected to make significant contributions and discoveries across various astronomical fields, including cosmology, galaxies and active galactic nuclei (AGN), the Milky Way and nearby galaxies, stars, exoplanets, Solar System objects, astrometry, and transients and variable sources. This review aims to provide a comprehensive overview of the CSST instruments, observational capabilities, data products, and scientific potential.

astro-ph.IM↗

arXiv2025

Optical+NIR analysis of a Newly Confirmed Einstein ring at z$\sim$1 from the Kilo-Degree Survey: Dark matter fraction, total and dark matter density slope and IMF

We report the spectroscopic confirmation of a bright blue Einstein ring in the Kilo Degree Survey (KiDS) footprint: the Einstein ``blue eye''. Spectroscopic data from X-Shooter at the Very Large Telescope (VLT) show that the lens is a typical early-type galaxy (ETG) at $z_l=0.9906$, while the background source is a Ly$α$ emitter at $z_s=2.823$. The reference lens modeling was performed on a high-resolution $Y-$band adaptive-optics image from HAWK-I at VLT. Assuming a singular isothermal ellipsoid (SIE) total mass density profile, we inferred an Einstein radius $R_{Ein}=10.47 \pm 0.06$ kpc. The average slope of the total mass density inside the Einstein radius, as determined by a joint analysis of lensing and isotropic Jeans equations is $γ_{tot}=2.14^{+0.06}_{-0.07}$, showing no systematic deviation from the slopes of lower redshift galaxies, This can be the evidence of ETGs developing through dry mergers plus moderate dissipationless accretion. Stellar population analysis with 8-band ($gri$ZYJHK$s$) photometries from KiDS and VIKING shows that the total stellar mass of the lens is $M*=(3.95\pm 0.35)\times 10^{11} M_\odot$ (Salpeter Initial Mass Function, IMF), implying a dark matter fraction inside the effective radius to be $f_{\rm DM}=0.307\pm 0.151$. We finally explored the dark matter halo slope and found a strong degeneracy with the dynamic stellar mass. Dark matter adiabatic contraction is needed to explain the posterior distribution of the slope unless IMF heavier than Salpeter is assumed.

astro-ph.GA↗

arXiv2025

Efficient Inference for Large Reasoning Models: A Survey

Large Reasoning Models (LRMs) significantly improve the reasoning ability of Large Language Models (LLMs) by learning to reason, exhibiting promising performance in solving complex tasks. However, their deliberative reasoning process leads to inefficiencies in token usage, memory consumption, and inference time. Thus, this survey provides a review of efficient inference methods designed specifically for LRMs, focusing on mitigating token inefficiency while preserving the reasoning quality. The overview structure of this paper is shown in Figure~\ref{fig:paper_structure}. First, we introduce a taxonomy to group the recent methods into two main categories: (a) explicit compact Chain-of-Thought (CoT), which reduces tokens while keeping the explicit reasoning structure, and (b) implicit latent CoT, which encodes reasoning steps within hidden representations instead of explicit tokens. Meanwhile, we discuss their strengths and weaknesses. Then, we conduct empirical analyses on existing methods from reasoning scenarios, object functions, and performance \& efficiency aspects. Besides, we present open challenges in this field, including human-centric controllable reasoning, trade-off between interpretability and efficiency of reasoning, ensuring the safety of efficient reasoning, and broader applications of efficient reasoning. In addition, we highlight key insights for enhancing LRMs' inference efficiency via techniques such as model merging, new architectures, and agent routers. We hope this work serves as a valuable guide, helping researchers overcome challenges in this vibrant field. A collection of efficient reasoning methods for LRMs (papers and codes) is provided at this link: https://github.com/yueliu1999/Awesome-Efficient-Inference-for-LRMs.

cs.CL↗

arXiv2025

An Optimal In-Situ Multipole Algorithm for the Isotropic Three-Point Correlation Functions

We present an optimised multipole algorithm for computing the three-point correlation function (3PCF), tailored for application to large-scale cosmological datasets. The algorithm builds on a $in\, situ$ interpretation of correlation functions, wherein spatial displacements are implemented via translation window functions. In Fourier space, these translations correspond to plane waves, whose decomposition into spherical harmonics naturally leads to a multipole expansion framework for the 3PCF. To accelerate computation, we incorporate density field reconstruction within the framework of multiresolution analysis, enabling efficient summation using either grid-based or particle-based schemes. In addition to the shared computational cost of reconstructing the multipole-decomposed density fields - scaling as $\mathcal{O}(L^2_{\text{trun}} N_g \log N_g)$ (where $N_g$ is the number of grids and $L_{\text{trun}}$ is the truncation order of the multipole expansion) - the final summation step achieves a complexity of $\mathcal{O}(D^6_{\text{sup}} N_g)$ for the grid-based approach and $\mathcal{O}(D^3_{\text{sup}} N_p)$ for the particle-based scheme (where $D_{\text{sup}}$ is the support of the basis function and $N_p$ is the number of particles). The proposed $in\, situ$ multipole algorithm is fully GPU-accelerated and implemented in the open-source $Hermes$ toolkit for cosmic statistics. This development enables fast, scalable higher-order clustering analyses for large-volume datasets from current and upcoming cosmological surveys such as Euclid, DESI, LSST, and CSST.

astro-ph.CO↗

arXiv2025

TruthTorchLM: A Comprehensive Library for Predicting Truthfulness in LLM Outputs

Generative Large Language Models (LLMs)inevitably produce untruthful responses. Accurately predicting the truthfulness of these outputs is critical, especially in high-stakes settings. To accelerate research in this domain and make truthfulness prediction methods more accessible, we introduce TruthTorchLM an open-source, comprehensive Python library featuring over 30 truthfulness prediction methods, which we refer to as Truth Methods. Unlike existing toolkits such as Guardrails, which focus solely on document-grounded verification, or LM-Polygraph, which is limited to uncertainty-based methods, TruthTorchLM offers a broad and extensible collection of techniques. These methods span diverse tradeoffs in computational cost, access level (e.g., black-box vs white-box), grounding document requirements, and supervision type (self-supervised or supervised). TruthTorchLM is seamlessly compatible with both HuggingFace and LiteLLM, enabling support for locally hosted and API-based models. It also provides a unified interface for generation, evaluation, calibration, and long-form truthfulness prediction, along with a flexible framework for extending the library with new methods. We conduct an evaluation of representative truth methods on three datasets, TriviaQA, GSM8K, and FactScore-Bio. The code is available at https://github.com/Ybakman/TruthTorchLM

cs.CL↗

arXiv2025

Anti-Interference Diffractive Deep Neural Networks for Multi-Object Recognition

Optical neural networks (ONNs) are emerging as a promising neuromorphic computing paradigm for object recognition, offering unprecedented advantages in light-speed computation, ultra-low power consumption, and inherent parallelism. However, most of ONNs are only capable of performing simple object classification tasks. These tasks are typically constrained to single-object scenarios, which limits their practical applications in multi-object recognition tasks. Here, we propose an anti-interference diffractive deep neural network (AI D2NN) that can accurately and robustly recognize targets in multi-object scenarios, including intra-class, inter-class, and dynamic interference. By employing different deep-learning-based training strategies for targets and interference, two transmissive diffractive layers form a physical network that maps the spatial information of targets all-optically into the power spectrum of the output light, while dispersing all interference as background noise. We demonstrate the effectiveness of this framework in classifying unknown handwritten digits under dynamic scenarios involving 40 categories of interference, achieving a simulated blind testing accuracy of 87.4% using terahertz waves. The presented framework can be physically scaled to operate at any electromagnetic wavelength by simply scaling the diffractive features in proportion to the wavelength range of interest. This work can greatly advance the practical application of ONNs in target recognition and pave the way for the development of real-time, high-throughput, low-power all-optical computing systems, which are expected to be applied to autonomous driving perception, precision medical diagnosis, and intelligent security monitoring.

physics.optics↗

arXiv2025

CountLLM: Towards Generalizable Repetitive Action Counting via Large Language Model

Repetitive action counting, which aims to count periodic movements in a video, is valuable for video analysis applications such as fitness monitoring. However, existing methods largely rely on regression networks with limited representational capacity, which hampers their ability to accurately capture variable periodic patterns. Additionally, their supervised learning on narrow, limited training sets leads to overfitting and restricts their ability to generalize across diverse scenarios. To address these challenges, we propose CountLLM, the first large language model (LLM)-based framework that takes video data and periodic text prompts as inputs and outputs the desired counting value. CountLLM leverages the rich clues from explicit textual instructions and the powerful representational capabilities of pre-trained LLMs for repetitive action counting. To effectively guide CountLLM, we develop a periodicity-based structured template for instructions that describes the properties of periodicity and implements a standardized answer format to ensure consistency. Additionally, we propose a progressive multimodal training paradigm to enhance the periodicity-awareness of the LLM. Empirical evaluations on widely recognized benchmarks demonstrate CountLLM's superior performance and generalization, particularly in handling novel and out-of-domain actions that deviate significantly from the training data, offering a promising avenue for repetitive action counting.

cs.CV↗

arXiv2025

Kimi-VL Technical Report

We present Kimi-VL, an efficient open-source Mixture-of-Experts (MoE) vision-language model (VLM) that offers advanced multimodal reasoning, long-context understanding, and strong agent capabilities - all while activating only 2.8B parameters in its language decoder (Kimi-VL-A3B). Kimi-VL demonstrates strong performance across challenging domains: as a general-purpose VLM, Kimi-VL excels in multi-turn agent tasks (e.g., OSWorld), matching flagship models. Furthermore, it exhibits remarkable capabilities across diverse challenging vision language tasks, including college-level image and video comprehension, OCR, mathematical reasoning, and multi-image understanding. In comparative evaluations, it effectively competes with cutting-edge efficient VLMs such as GPT-4o-mini, Qwen2.5-VL-7B, and Gemma-3-12B-IT, while surpassing GPT-4o in several key domains. Kimi-VL also advances in processing long contexts and perceiving clearly. With a 128K extended context window, Kimi-VL can process diverse long inputs, achieving impressive scores of 64.5 on LongVideoBench and 35.1 on MMLongBench-Doc. Its native-resolution vision encoder, MoonViT, further allows it to see and understand ultra-high-resolution visual inputs, achieving 83.2 on InfoVQA and 34.5 on ScreenSpot-Pro, while maintaining lower computational cost for common tasks. Building upon Kimi-VL, we introduce an advanced long-thinking variant: Kimi-VL-Thinking-2506. Developed through long chain-of-thought (CoT) supervised fine-tuning (SFT) and reinforcement learning (RL), the latest model exhibits strong long-horizon reasoning capabilities (64.0 on MMMU, 46.3 on MMMU-Pro, 56.9 on MathVision, 80.1 on MathVista, 65.2 on VideoMMMU) while obtaining robust general abilities. Code and models are publicly accessible at https://github.com/MoonshotAI/Kimi-VL.

cs.CV↗

arXiv2025

Kimi k1.5: Scaling Reinforcement Learning with LLMs

Language model pretraining with next token prediction has proved effective for scaling compute but is limited to the amount of available training data. Scaling reinforcement learning (RL) unlocks a new axis for the continued improvement of artificial intelligence, with the promise that large language models (LLMs) can scale their training data by learning to explore with rewards. However, prior published work has not produced competitive results. In light of this, we report on the training practice of Kimi k1.5, our latest multi-modal LLM trained with RL, including its RL training techniques, multi-modal data recipes, and infrastructure optimization. Long context scaling and improved policy optimization methods are key ingredients of our approach, which establishes a simplistic, effective RL framework without relying on more complex techniques such as Monte Carlo tree search, value functions, and process reward models. Notably, our system achieves state-of-the-art reasoning performance across multiple benchmarks and modalities -- e.g., 77.5 on AIME, 96.2 on MATH 500, 94-th percentile on Codeforces, 74.9 on MathVista -- matching OpenAI's o1. Moreover, we present effective long2short methods that use long-CoT techniques to improve short-CoT models, yielding state-of-the-art short-CoT reasoning results -- e.g., 60.8 on AIME, 94.6 on MATH500, 47.3 on LiveCodeBench -- outperforming existing short-CoT models such as GPT-4o and Claude Sonnet 3.5 by a large margin (up to +550%).

cs.AI↗