arXiv subjects

Kai Liu

Publications and source records attributed to Kai Liu.

At least 181 records · Page 10Linked to original sources

arXiv2025

Mixture-of-Instructions: Aligning Large Language Models via Mixture Prompting

With the proliferation of large language models (LLMs), the comprehensive alignment of such models across multiple tasks has emerged as a critical area of research. Existing alignment methodologies primarily address single task, such as multi-turn dialogue, coding, mathematical problem-solving, and tool usage. Although there is a large amount of high-quality data available for those tasks, most of them provide only questions and answers without including the system prompt. Though a detailed analysis of the Qwen language model, we found that the system prompt has a significant impact on both training and inference processes of LLM. We attributes this phenomenon to overfitting to the system prompt. In address this issue, we introduce a novel technique termed Mixture-of-Instructions (MoI), which employs a strategy of instruction packing combined with diverse system prompts to boost the alignment efficiency of language models. We have also compiled a diverse set of seven benchmark datasets to rigorously evaluate the alignment efficacy of the MoI-enhanced language model. Our methodology was applied to the open-source Qwen-7B-chat model, culminating in the development of Qwen-SFT-MoI. This enhanced model demonstrates significant advancements in generative capabilities across coding, mathematics, and tool use tasks.

cs.CL↗

arXiv2025

BiMaCoSR: Binary One-Step Diffusion Model Leveraging Flexible Matrix Compression for Real Super-Resolution

While super-resolution (SR) methods based on diffusion models (DM) have demonstrated inspiring performance, their deployment is impeded due to the heavy request of memory and computation. Recent researchers apply two kinds of methods to compress or fasten the DM. One is to compress the DM into 1-bit, aka binarization, alleviating the storage and computation pressure. The other distills the multi-step DM into only one step, significantly speeding up inference process. Nonetheless, it remains impossible to deploy DM to resource-limited edge devices. To address this problem, we propose BiMaCoSR, which combines binarization and one-step distillation to obtain extreme compression and acceleration. To prevent the catastrophic collapse of the model caused by binarization, we proposed sparse matrix branch (SMB) and low rank matrix branch (LRMB). Both auxiliary branches pass the full-precision (FP) information but in different ways. SMB absorbs the extreme values and its output is high rank, carrying abundant FP information. Whereas, the design of LRMB is inspired by LoRA and is initialized with the top r SVD components, outputting low rank representation. The computation and storage overhead of our proposed branches can be safely ignored. Comprehensive comparison experiments are conducted to exhibit BiMaCoSR outperforms current state-of-the-art binarization methods and gains competitive performance compared with FP one-step model. BiMaCoSR achieves a 23.8x compression ratio and a 27.4x speedup ratio compared to FP counterpart. Our code and model are available at https://github.com/Kai-Liu001/BiMaCoSR.

cs.CV↗

arXiv2025

SCDiar: a streaming diarization system based on speaker change detection and speech recognition

In hours-long meeting scenarios, real-time speech stream often struggles with achieving accurate speaker diarization, commonly leading to speaker identification and speaker count errors. To address this challenge, we propose SCDiar, a system that operates on speech segments, split at the token level by a speaker change detection (SCD) module. Building on these segments, we introduce several enhancements to efficiently select the best available segment for each speaker. These improvements lead to significant gains across various benchmarks. Notably, on real-world meeting data involving more than ten participants, SCDiar outperforms previous systems by up to 53.6\% in accuracy, substantially narrowing the performance gap between online and offline systems.

eess.AS↗

arXiv2025

Modeling Extensive Defects in Metals through Classical Potential-Guided Sampling and Automated Configuration Reconstruction

Extended defects such as dislocation networks and general grain boundaries are ubiquitous in metals, and accurately modeling these extensive defects is crucial for understanding their deformation mechanisms. Existing machine learning interatomic potentials (MLIPs) often fall short in adequately describing these defects, as their significant characteristic sizes exceed the computational limits of first-principles calculations. In this study, we address these challenges by establishing a comprehensive defect genome through empirical interatomic potential-guided sampling. To further enable accurate first-principles calculations on this defect genome, we have developed an automated configuration reconstruction technique. This method transforms defect atomic clusters into periodic configurations through precise atom insertion, utilizing Grand Canonical Monte Carlo simulations. These strategies enable the development of highly accurate and transferable MLIPs for modeling extensive defects in metals. Using body-centered cubic tungsten as a model system, we develop an MLIP that reveals unique plastic mechanisms in simulations of nanoindentation. This framework not only improves the modeling accuracy of extensive defects in crystalline materials but also establishes a robust foundation for further advancement of MLIP development through the strategic use of defect genomes.

cond-mat.mtrl-sci↗

arXiv2025

Spontaneous Donor Defects and Voltage-Assisted Hole Doping in Beta-Gallium Oxides under Multiple Epitaxy Conditions

Beta-phase gallium oxide (beta-Ga2O3) is prone to the spontaneous formation of donor defects but poses a formidable challenge in achieving high-quality p-type doping, mainly due to its exceptionally low valence band maximum (VBM). In this study, we utilize first-principles computations to investigate the origin of spontaneous donor defects in beta-Ga2O3 grown by three typical techniques: molecular beam epitaxy (MBE), metal organic chemical vapor deposition (MOCVD), and halide vapor phase epitaxy (HVPE). Our findings elucidate that the primary donor defects vary with the growth techniques, specifically Gai3+ for MBE, Hi+ and CGa+ for MOCVD, and (2VGa+Gai+2VO)+ and ClO+ for HVPE under unintentionally doped conditions. Employing a theoretically proposed voltage-assisted doping method, we computationally demonstrate that the dominant spontaneous donors can be significantly reduced accompanied by a noticeable increase in acceptors, leading to a stepwise reduction of Fermi level to 0.52, 0.88, and 2.10 eV above VBM for the MOCVD, HVPE, and MBE methods, and a hole concentration of 8.5*10^17, 8.7*10^11, and 2.7*10^-9 cm-3, respectively, at room temperature without the use of external dopants. By introducing Mg doping, we further reduce the Fermi level for both the MBE and HVPE experiments.

cond-mat.mtrl-sci↗

arXiv2025

CAMEL: Cross-Attention Enhanced Mixture-of-Experts and Language Bias for Code-Switching Speech Recognition

Code-switching automatic speech recognition (ASR) aims to transcribe speech that contains two or more languages accurately. To better capture language-specific speech representations and address language confusion in code-switching ASR, the mixture-of-experts (MoE) architecture and an additional language diarization (LD) decoder are commonly employed. However, most researches remain stagnant in simple operations like weighted summation or concatenation to fuse languagespecific speech representations, leaving significant opportunities to explore the enhancement of integrating language bias information. In this paper, we introduce CAMEL, a cross-attention-based MoE and language bias approach for code-switching ASR. Specifically, after each MoE layer, we fuse language-specific speech representations with cross-attention, leveraging its strong contextual modeling abilities. Additionally, we design a source attention-based mechanism to incorporate the language information from the LD decoder output into text embeddings. Experimental results demonstrate that our approach achieves state-of-the-art performance on the SEAME, ASRU200, and ASRU700+LibriSpeech460 Mandarin-English code-switching ASR datasets.

cs.SD↗

arXiv2024

ESOD: Efficient Small Object Detection on High-Resolution Images

Enlarging input images is a straightforward and effective approach to promote small object detection. However, simple image enlargement is significantly expensive on both computations and GPU memory. In fact, small objects are usually sparsely distributed and locally clustered. Therefore, massive feature extraction computations are wasted on the non-target background area of images. Recent works have tried to pick out target-containing regions using an extra network and perform conventional object detection, but the newly introduced computation limits their final performance. In this paper, we propose to reuse the detector's backbone to conduct feature-level object-seeking and patch-slicing, which can avoid redundant feature extraction and reduce the computation cost. Incorporating a sparse detection head, we are able to detect small objects on high-resolution inputs (e.g., 1080P or larger) for superior performance. The resulting Efficient Small Object Detection (ESOD) approach is a generic framework, which can be applied to both CNN- and ViT-based detectors to save the computation and GPU memory costs. Extensive experiments demonstrate the efficacy and efficiency of our method. In particular, our method consistently surpasses the SOTA detectors by a large margin (e.g., 8% gains on AP) on the representative VisDrone, UAVDT, and TinyPerson datasets. Code is available at https://github.com/alibaba/esod.

cs.CV↗

arXiv2024

Switchable Chern insulator, isospin competitions and charge density waves in rhombohedral graphene moire superlattices

Graphene-based moire superlattices provide a versatile platform for exploring novel correlated and topological electronic states, driven by enhanced Coulomb interactions within flat bands. The intrinsic tunability of graphene s multiple degrees of freedom enables precise control over these complex quantum phases. In this study, we observe a range of competing phases and their transitions in rhombohedrally stacked hexalayer graphene on hexagonal boron nitride (r-6G/hBN) moire superlattices. When electrons are polarized away from the moire superlattice, we firstly identify a Chern insulator with reversible Chern numbers at v = 1 (one electron per moire cell), attributed to the competition between bulk and edge magnetizations.Then, we detect transitions between three distinct insulating states at v = 2, driven by vertical displacement field D and vertical magnetic field B. These insulating phases are distinguished as spin-antiferromagnetic, spin-polarized, and valley-polarized insulators, based on their responses to parallel and perpendicular magnetic fields. When electrons are polarized toward the moire superlattice, in a device with large twist angle, insulating states appear at v = 1/3 and 2/3 at zero magnetic field, and v = 1/2 in a magnetic field. Our findings reveal a rich interplay of charge, isospin, topology and magnetic field in rhombohedral graphene moire superlattices.

cond-mat.str-el↗

arXiv2024

Magneto-Ionic Physical Reservoir Computing

Recent progresses in magnetoionics offer exciting potentials to leverage its non-linearity, short-term memory, and energy-efficiency to uniquely advance the field of physical reservoir computing. In this work, we experimentally demonstrate the classification of temporal data using a magneto-ionic (MI) heterostructure. The device was specifically engineered to induce non-linear ion migration dynamics, which in turn imparted non-linearity and short-term memory (STM) to the magnetization. These capabilities, key features for enabling reservoir computing, were investigated, and the role of the ion migration mechanism, along with its history-dependent influence on STM, was explained. These attributes were utilized to distinguish between sine and square waveforms within a randomly distributed set of pulses. Additionally, two important performance metrics, short-term memory and parity check capacity (PC), were quantified, yielding promising values of 1.44 and 2, respectively, comparable to those of other state-of-the-art reservoirs. Our work paves the way for exploiting the relaxation dynamics of solid-state magneto-ionic platforms and developing energy-efficient magneto-ionic reservoir computing devices.

cond-mat.mes-hall↗

arXiv2024

Real-space study of zero-field correlation in tetralayer rhombohedral graphene

Rhombohedral graphene (RG) has emerged as a promising platform for exploring exotic quantum phenomena, such as quantum magnetism, unconventional superconductivity, and fractional quantum anomalous Hall effects. Despite its potential, atomic-scale investigations of RG remain limited, hindering a detailed microscopic understanding of the origins of these correlated states. In this study, we employ scanning probe microscopy and spectroscopy to probe the intrinsic electronic states in trilayer and tetralayer RG. We identify a correlated insulating state with a 17 meV gap at the charge neutrality point in tetralayer RG, which is absent in the trilayer configuration. This gap is suppressed by applying a perpendicular magnetic field or doping the charge carrier density and does not exhibit inter-valley coherence patterns. We attribute this phenomenon to a symmetry-broken layer antiferromagnetic state, characterized by ferrimagnetic ordering in the outermost layers and antiferromagnetic coupling between them. To further investigate this magnetic correlated state, we conduct local scattering experiments. Within the correlated regime, a bound state emerges around a non-magnetic impurity but is absent near magnetic impurities, suggesting that non-magnetic doping induces a spin texture in the ferrimagnetic surface layers. Outside the correlated regime, Friedel oscillations are observed, allowing precise determination of the band dispersion in tetralayer RG. These findings provide atomic-scale evidences of zero-field correlations in RG and may be extended to study other exotic phases in RG.

cond-mat.mes-hall↗

arXiv2024

Learning Identifiable Factorized Causal Representations of Cellular Responses

The study of cells and their responses to genetic or chemical perturbations promises to accelerate the discovery of therapeutic targets. However, designing adequate and insightful models for such data is difficult because the response of a cell to perturbations essentially depends on its biological context (e.g., genetic background or cell type). For example, while discovering therapeutic targets, one may want to enrich for drugs that specifically target a certain cell type. This challenge emphasizes the need for methods that explicitly take into account potential interactions between drugs and contexts. Towards this goal, we propose a novel Factorized Causal Representation (FCR) learning method that reveals causal structure in single-cell perturbation data from several cell lines. Based on the framework of identifiable deep generative models, FCR learns multiple cellular representations that are disentangled, comprised of covariate-specific ($\mathbf{z}_x$), treatment-specific ($\mathbf{z}_{t}$), and interaction-specific ($\mathbf{z}_{tx}$) blocks. Based on recent advances in non-linear ICA theory, we prove the component-wise identifiability of $\mathbf{z}_{tx}$ and block-wise identifiability of $\mathbf{z}_t$ and $\mathbf{z}_x$. Then, we present our implementation of FCR, and empirically demonstrate that it outperforms state-of-the-art baselines in various tasks across four single-cell datasets.

cs.LG↗

arXiv2024

Delving into the Reversal Curse: How Far Can Large Language Models Generalize?

While large language models (LLMs) showcase unprecedented capabilities, they also exhibit certain inherent limitations when facing seemingly trivial tasks. A prime example is the recently debated "reversal curse", which surfaces when models, having been trained on the fact "A is B", struggle to generalize this knowledge to infer that "B is A". In this paper, we examine the manifestation of the reversal curse across various tasks and delve into both the generalization abilities and the problem-solving mechanisms of LLMs. This investigation leads to a series of significant insights: (1) LLMs are able to generalize to "B is A" when both A and B are presented in the context as in the case of a multiple-choice question. (2) This generalization ability is highly correlated to the structure of the fact "A is B" in the training documents. For example, this generalization only applies to biographies structured in "[Name] is [Description]" but not to "[Description] is [Name]". (3) We propose and verify the hypothesis that LLMs possess an inherent bias in fact recalling during knowledge application, which explains and underscores the importance of the document structure to successful learning. (4) The negative impact of this bias on the downstream performance of LLMs can hardly be mitigated through training alone. These findings offer a novel perspective on interpreting LLMs' generalization through their intrinsic mechanisms and provide insights for developing more effective learning methods. Our code and data are available at https://github.com/alibaba/thinking_bias.git.

cs.CL↗

arXiv2024

Tunable surface electron gas and effect of phonons in Sr$_2$CuO$_3$: A first-principles study

While the conducting CuO$_2$ planes in cuprate superconductors have been widely recognized as a crucial component in producing high superconducting $T_\text{c}$, recent experimental and theoretical studies on Ba$_{2-x}$Sr$_x$CuO$_{3+}$$_δ$ have also drawn much attention to the importance of Cu-O chains in one-dimensional (1D) cuprates. To better understand the cuprates containing Cu-O chains, here we have studied the electronic, magnetic, and phonon properties of Sr$_2$CuO$_3$ bulk and films based on the spin-polarized density functional theory calculations. We first reproduced the typical Mott insulator feature of the cuprate parent compound for bulk Sr$_2$CuO$_3$, and then built a Sr$_2$CuO$_3$ thin film with Cu-O chains exposed on the surface to directly investigate their characteristics. Different from the insulating bulk phase, the Sr$_2$CuO$_3$ surface shows interesting metallic properties. Further electronic structure calculations reveal the existence of spin-polarized electron gas between surface Sr atoms that strongly depends on the interchain coupling of Cu spins. Moreover, the phonon modes that involve the vibrations of in-chain and out-of-chain O atoms can induce strong charge and spin fluctuations in the surface layer of Sr$_2$CuO$_3$ film, which suggests significant multiple degree-of-freedom couplings that may be important for the superconductivity in 1D cuprates. Our work provides a comprehensive viewpoint of the properties of Cu-O chains in Sr$_2$CuO$_3$, facilitating a complete understanding of 1D cuprate superconductors.

cond-mat.supr-con↗

arXiv2024

Category-Extensible Out-of-Distribution Detection via Hierarchical Context Descriptions

The key to OOD detection has two aspects: generalized feature representation and precise category description. Recently, vision-language models such as CLIP provide significant advances in both two issues, but constructing precise category descriptions is still in its infancy due to the absence of unseen categories. This work introduces two hierarchical contexts, namely perceptual context and spurious context, to carefully describe the precise category boundary through automatic prompt tuning. Specifically, perceptual contexts perceive the inter-category difference (e.g., cats vs apples) for current classification tasks, while spurious contexts further identify spurious (similar but exactly not) OOD samples for every single category (e.g., cats vs panthers, apples vs peaches). The two contexts hierarchically construct the precise description for a certain category, which is, first roughly classifying a sample to the predicted category and then delicately identifying whether it is truly an ID sample or actually OOD. Moreover, the precise descriptions for those categories within the vision-language framework present a novel application: CATegory-EXtensible OOD detection (CATEX). One can efficiently extend the set of recognizable categories by simply merging the hierarchical contexts learned under different sub-task settings. And extensive experiments are conducted to demonstrate CATEX's effectiveness, robustness, and category-extensibility. For instance, CATEX consistently surpasses the rivals by a large margin with several protocols on the challenging ImageNet-1K dataset. In addition, we offer new insights on how to efficiently scale up the prompt engineering in vision-language models to recognize thousands of object categories, as well as how to incorporate large language models (like GPT-3) to boost zero-shot applications. Code is publicly available at https://github.com/alibaba/catex.

cs.CV↗

arXiv2024

Comprehensive Study on the Slat Noise of 30P30N High-Lift Airfoil Basd on High-Order Wall-Resolved Large-Eddy Simulation

This study presents wall-resolved large-eddy simulations (WRLES) of a high-lift airfoil, based on high-order flux reconstruction (FR) commercial software Dimaxer, which runs on consumer level GPUs. A series of independence tests are conducted, including various Ffowcs Williams-Hawkings sampling surfaces, different mesh densities, simulations at 4th and 5th order accuracies, and varying spanwise lengths, to establish best practice for predicting slat noise through high-order WRLES. The results show excellent agreement with experimental data while requiring significantly fewer computational resources than traditional second-order methods. An investigation on the effects of Reynolds number (Re) is performed by scaling the airfoil size, with Reynolds numbers ranging from 8.55e5 to a real aircraft level of 1.71e7. By applying simple scaling through Strouhal number (St), spanwise correction, and distance from the receiver, the far-field noise spectra for different Reynolds numbers can be coincided. Additionally, simulations are performed at four angles of attack: 3°, 5.5°, 9.5°, and 14°. The results indicate that higher angles of attack lead to a less intense feedback loop, resulting in lower tonal noise frequencies and reduced noise amplitude. The maximum noise reduction observed is over 14dB when comparing 14° to 3°. Furthermore, an improved formula is proposed to enhance the prediction of slat noise tonal frequencies and to better elucidate the mechanism behind tonal noise generation.

physics.flu-dyn↗

arXiv2024

Hunyuan-Large: An Open-Source MoE Model with 52 Billion Activated Parameters by Tencent

In this paper, we introduce Hunyuan-Large, which is currently the largest open-source Transformer-based mixture of experts model, with a total of 389 billion parameters and 52 billion activation parameters, capable of handling up to 256K tokens. We conduct a thorough evaluation of Hunyuan-Large's superior performance across various benchmarks including language understanding and generation, logical reasoning, mathematical problem-solving, coding, long-context, and aggregated tasks, where it outperforms LLama3.1-70B and exhibits comparable performance when compared to the significantly larger LLama3.1-405B model. Key practice of Hunyuan-Large include large-scale synthetic data that is orders larger than in previous literature, a mixed expert routing strategy, a key-value cache compression technique, and an expert-specific learning rate strategy. Additionally, we also investigate the scaling laws and learning rate schedule of mixture of experts models, providing valuable insights and guidances for future model development and optimization. The code and checkpoints of Hunyuan-Large are released to facilitate future innovations and applications. Codes: https://github.com/Tencent/Hunyuan-Large Models: https://huggingface.co/tencent/Tencent-Hunyuan-Large

cs.CL↗

arXiv2024

Rethinking Out-of-Distribution Detection on Imbalanced Data Distribution

Detecting and rejecting unknown out-of-distribution (OOD) samples is critical for deployed neural networks to void unreliable predictions. In real-world scenarios, however, the efficacy of existing OOD detection methods is often impeded by the inherent imbalance of in-distribution (ID) data, which causes significant performance decline. Through statistical observations, we have identified two common challenges faced by different OOD detectors: misidentifying tail class ID samples as OOD, while erroneously predicting OOD samples as head class from ID. To explain this phenomenon, we introduce a generalized statistical framework, termed ImOOD, to formulate the OOD detection problem on imbalanced data distribution. Consequently, the theoretical analysis reveals that there exists a class-aware bias item between balanced and imbalanced OOD detection, which contributes to the performance gap. Building upon this finding, we present a unified training-time regularization technique to mitigate the bias and boost imbalanced OOD detectors across architecture designs. Our theoretically grounded method translates into consistent improvements on the representative CIFAR10-LT, CIFAR100-LT, and ImageNet-LT benchmarks against several state-of-the-art OOD detection approaches. Code is available at https://github.com/alibaba/imood.

cs.CV↗

arXiv2024

Enhancing LLM's Cognition via Structurization

When reading long-form text, human cognition is complex and structurized. While large language models (LLMs) process input contexts through a causal and sequential perspective, this approach can potentially limit their ability to handle intricate and complex inputs effectively. To enhance LLM's cognition capability, this paper presents a novel concept of context structurization. Specifically, we transform the plain, unordered contextual sentences into well-ordered and hierarchically structurized elements. By doing so, LLMs can better grasp intricate and extended contexts through precise attention and information-seeking along the organized structures. Extensive evaluations are conducted across various model architectures and sizes (including a series of auto-regressive LLMs as well as BERT-like masking models) on a diverse set of NLP tasks (e.g., context-based question-answering, exhaustive hallucination evaluation, and passage-level dense retrieval). Empirical results show consistent and significant performance gains afforded by a single-round structurization. In particular, we boost the open-sourced LLaMA2-70B model to achieve comparable performance against GPT-3.5-Turbo as the hallucination evaluator. Besides, we show the feasibility of distilling advanced LLMs' language processing abilities to a smaller yet effective StruXGPT-7B to execute structurization, addressing the practicality of our approach. Code is available at https://github.com/alibaba/struxgpt.

cs.CL↗