arXiv subjects

Kai Li

Publications and source records attributed to Kai Li.

At least 199 records · Page 11Linked to original sources

arXiv2025

MultiBooth: Towards Generating All Your Concepts in an Image from Text

This paper introduces MultiBooth, a novel and efficient technique for multi-concept customization in image generation from text. Despite the significant advancements in customized generation methods, particularly with the success of diffusion models, existing methods often struggle with multi-concept scenarios due to low concept fidelity and high inference cost. MultiBooth addresses these issues by dividing the multi-concept generation process into two phases: a single-concept learning phase and a multi-concept integration phase. During the single-concept learning phase, we employ a multi-modal image encoder and an efficient concept encoding technique to learn a concise and discriminative representation for each concept. In the multi-concept integration phase, we use bounding boxes to define the generation area for each concept within the cross-attention map. This method enables the creation of individual concepts within their specified regions, thereby facilitating the formation of multi-concept images. This strategy not only improves concept fidelity but also reduces additional inference cost. MultiBooth surpasses various baselines in both qualitative and quantitative evaluations, showcasing its superior performance and computational efficiency. Project Page: https://multibooth.github.io/

cs.CV↗

arXiv2025

Facial Attractiveness Prediction in Live Streaming: A New Benchmark and Multi-modal Method

Facial attractiveness prediction (FAP) has long been an important computer vision task, which could be widely applied in live streaming for facial retouching, content recommendation, etc. However, previous FAP datasets are either small, closed-source, or lack diversity. Moreover, the corresponding FAP models exhibit limited generalization and adaptation ability. To overcome these limitations, in this paper we present LiveBeauty, the first large-scale live-specific FAP dataset, in a more challenging application scenario, i.e., live streaming. 10,000 face images are collected from a live streaming platform directly, with 200,000 corresponding attractiveness annotations obtained from a well-devised subjective experiment, making LiveBeauty the largest open-access FAP dataset in the challenging live scenario. Furthermore, a multi-modal FAP method is proposed to measure the facial attractiveness in live streaming. Specifically, we first extract holistic facial prior knowledge and multi-modal aesthetic semantic features via a Personalized Attractiveness Prior Module (PAPM) and a Multi-modal Attractiveness Encoder Module (MAEM), respectively, then integrate the extracted features through a Cross-Modal Fusion Module (CMFM). Extensive experiments conducted on both LiveBeauty and other open-source FAP datasets demonstrate that our proposed method achieves state-of-the-art performance. Dataset will be available soon.

cs.CV↗

arXiv2025

Automatic Operator-level Parallelism Planning for Distributed Deep Learning -- A Mixed-Integer Programming Approach

As the artificial intelligence community advances into the era of large models with billions of parameters, distributed training and inference have become essential. While various parallelism strategies-data, model, sequence, and pipeline-have been successfully implemented for popular neural networks on main-stream hardware, optimizing the distributed deployment schedule requires extensive expertise and manual effort. Further more, while existing frameworks with most simple chain-like structures, they struggle with complex non-linear architectures. Mixture-of-experts and multi-modal models feature intricate MIMO and branch-rich topologies that require fine-grained operator-level parallelization beyond the capabilities of existing frameworks. We propose formulating parallelism planning as a scheduling optimization problem using mixed-integer programming. We propose a bi-level solution framework balancing optimality with computational efficiency, automatically generating effective distributed plans that capture both the heterogeneous structure of modern neural networks and the underlying hardware constraints. In experiments comparing against expert-designed strategies like DeepSeek's DualPipe, our framework achieves comparable or superior performance, reducing computational bubbles by half under the same memory constraints. The framework's versatility extends beyond throughput optimization to incorporate hardware utilization maximization, memory capacity constraints, and other considerations or potential strategies. Such capabilities position our solution as both a valuable research tool for exploring optimal parallelization strategies and a practical industrial solution for large-scale AI deployment.

cs.LG↗

arXiv2025

Collective Behavior Clone with Visual Attention via Neural Interaction Graph Prediction

In this paper, we propose a framework, collective behavioral cloning (CBC), to learn the underlying interaction mechanism and control policy of a swarm system. Given the trajectory data of a swarm system, we propose a graph variational autoencoder (GVAE) to learn the local interaction graph. Based on the interaction graph and swarm trajectory, we use behavioral cloning to learn the control policy of the swarm system. To demonstrate the practicality of CBC, we deploy it on a real-world decentralized vision-based robot swarm system. A visual attention network is trained based on the learned interaction graph for online neighbor selection. Experimental results show that our method outperforms previous approaches in predicting both the interaction graph and swarm actions with higher accuracy. This work offers a promising approach for understanding interaction mechanisms and swarm dynamics in future swarm robotics research. Code and data are available.

cs.RO↗

arXiv2025

Dynamic Dictionary Learning for Remote Sensing Image Segmentation

Remote sensing image segmentation faces persistent challenges in distinguishing morphologically similar categories and adapting to diverse scene variations. While existing methods rely on implicit representation learning paradigms, they often fail to dynamically adjust semantic embeddings according to contextual cues, leading to suboptimal performance in fine-grained scenarios such as cloud thickness differentiation. This work introduces a dynamic dictionary learning framework that explicitly models class ID embeddings through iterative refinement. The core contribution lies in a novel dictionary construction mechanism, where class-aware semantic embeddings are progressively updated via multi-stage alternating cross-attention querying between image features and dictionary embeddings. This process enables adaptive representation learning tailored to input-specific characteristics, effectively resolving ambiguities in intra-class heterogeneity and inter-class homogeneity. To further enhance discriminability, a contrastive constraint is applied to the dictionary space, ensuring compact intra-class distributions while maximizing inter-class separability. Extensive experiments across both coarse- and fine-grained datasets demonstrate consistent improvements over state-of-the-art methods, particularly in two online test benchmarks (LoveDA and UAVid). Code is available at https://anonymous.4open.science/r/D2LS-8267/.

cs.CV↗

arXiv2025

Optimizing LLM Inference Throughput via Memory-aware and SLA-constrained Dynamic Batching

The increasing adoption of large language models (LLMs) necessitates inference serving systems that can deliver both high throughput and low latency. Deploying LLMs with hundreds of billions of parameters on memory-constrained GPUs exposes significant limitations in static batching methods. Current inference serving systems often treat batch sizes as fixed hyper-parameters, hindering real-time adaptation to varying system conditions. In this paper, we propose a dynamic batching method that continuously monitors memory utilization and adheres to service-level agreements (SLAs) to enable real-time batch size configuration adjustment. The method comprises two core components: a memory-aware batch scheduler that dynamically allocates GPU resources and a latency feedback mechanism that optimizes decoding processes under SLA constraints. The numerical experiments demonstrate throughput gains of 8% to 28% and capacity improvements of 22% compared to traditional static batching methods, while maintaining full compatibility with existing inference infrastructure. These results highlight the effectiveness of dynamic batching in balancing computational efficiency and quality-of-service requirements for contemporary LLM deployment scenarios. The source code of this work is publicly available at https://github.com/KevinLee1110/dynamic-batching.

cs.DC↗

arXiv2025

SonicSim: A customizable simulation platform for speech processing in moving sound source scenarios

Systematic evaluation of speech separation and enhancement models under moving sound source conditions requires extensive and diverse data. However, real-world datasets often lack sufficient data for training and evaluation, and synthetic datasets, while larger, lack acoustic realism. Consequently, neither effectively meets practical needs. To address this issue, we introduce SonicSim, a synthetic toolkit based on the embodied AI simulation platform Habitat-sim, designed to generate highly customizable data for moving sound sources. SonicSim supports multi-level adjustments, including scene-level, microphone-level, and source-level adjustments, enabling the creation of more diverse synthetic data. Leveraging SonicSim, we constructed a benchmark dataset called SonicSet, utilizing LibriSpeech, Freesound Dataset 50k (FSD50K), Free Music Archive (FMA), and 90 scenes from Matterport3D to evaluate speech separation and enhancement models. Additionally, to investigate the differences between synthetic and real-world data, we selected 5 hours of raw, non-reverberant data from the SonicSet validation set and recorded a real-world speech separation dataset, providing a reference for comparing SonicSet with other synthetic datasets. For speech enhancement, we utilized the real-world dataset RealMAN to validate the acoustic gap between SonicSet and existing synthetic datasets. The results indicate that models trained on SonicSet generalize better to real-world scenarios compared to other synthetic datasets. The code is publicly available at https://cslikai.cn/SonicSim/.

cs.SD↗

arXiv2025

SORRY-Bench: Systematically Evaluating Large Language Model Safety Refusal

Evaluating aligned large language models' (LLMs) ability to recognize and reject unsafe user requests is crucial for safe, policy-compliant deployments. Existing evaluation efforts, however, face three limitations that we address with SORRY-Bench, our proposed benchmark. First, existing methods often use coarse-grained taxonomies of unsafe topics, and are over-representing some fine-grained topics. For example, among the ten existing datasets that we evaluated, tests for refusals of self-harm instructions are over 3x less represented than tests for fraudulent activities. SORRY-Bench improves on this by using a fine-grained taxonomy of 44 potentially unsafe topics, and 440 class-balanced unsafe instructions, compiled through human-in-the-loop methods. Second, linguistic characteristics and formatting of prompts are often overlooked, like different languages, dialects, and more -- which are only implicitly considered in many evaluations. We supplement SORRY-Bench with 20 diverse linguistic augmentations to systematically examine these effects. Third, existing evaluations rely on large LLMs (e.g., GPT-4) for evaluation, which can be computationally expensive. We investigate design choices for creating a fast, accurate automated safety evaluator. By collecting 7K+ human annotations and conducting a meta-evaluation of diverse LLM-as-a-judge designs, we show that fine-tuned 7B LLMs can achieve accuracy comparable to GPT-4 scale LLMs, with lower computational cost. Putting these together, we evaluate over 50 proprietary and open-weight LLMs on SORRY-Bench, analyzing their distinctive safety refusal behaviors. We hope our effort provides a building block for systematic evaluations of LLMs' safety refusal capabilities, in a balanced, granular, and efficient manner. Benchmark demo, data, code, and models are available through https://sorry-bench.github.io.

cs.AI↗

arXiv2025

EdgeLLM: A Highly Efficient CPU-FPGA Heterogeneous Edge Accelerator for Large Language Models

The rapid advancements in artificial intelligence (AI), particularly the Large Language Models (LLMs), have profoundly affected our daily work and communication forms. However, it is still a challenge to deploy LLMs on resource-constrained edge devices (such as robots), due to the intensive computation requirements, heavy memory access, diverse operator types and difficulties in compilation. In this work, we proposed EdgeLLM to address the above issues. Firstly, focusing on the computation, we designed mix-precision processing element array together with group systolic architecture, that can efficiently support both FP16*FP16 for the MHA block (Multi-Head Attention) and FP16*INT4 for the FFN layer (Feed-Forward Network). Meanwhile specific optimization on log-scale structured weight sparsity, has been used to further increase the efficiency. Secondly, to address the compilation and deployment issue, we analyzed the whole operators within LLM models and developed a universal data parallelism scheme, by which all of the input and output features maintain the same data shape, enabling to process different operators without any data rearrangement. Then we proposed an end-to-end compiler to map the whole LLM model on CPU-FPGA heterogeneous system (AMD Xilinx VCU128 FPGA). The accelerator achieves 1.91x higher throughput and 7.55x higher energy efficiency than the commercial GPU (NVIDIA A100-SXM4-80G). When compared with state-of-the-art FPGA accelerator of FlightLLM, it shows 10-24% better performance in terms of HBM bandwidth utilization, energy efficiency and LLM throughput.

cs.AR↗

arXiv2025

Physical parameters of 12201 ASAS-SN contact binaries determined by the Neural Network

In the era of astronomical big data, more than one million contact binaries have been discovered. Traditional approaches of light curve analysis are inadequate for investigating such an extensive number of systems. This paper builds on prior research to present an advanced Neural Network model combined with the Markov Chain Monte Carlo algorithm and including spot parameters. This model was applied to 12785 contact binaries selected from All-Sky Automated Survey for Supernovae. By removing those with goodness of fit less than 0.8, we obtained the physical parameters of 12201 contact binaries. Among these binaries, 4332 are A-subtype systems, while 7869 are W-type systems, and 1594 systems have mass ratios larger than 0.72 (H-subtype system). A statistical study of the physical parameters was carried out, and we found that there are two peaks in the mass ratio distribution and that the probability of the presence of spot is about 50%. In addition, the differences in flux between the two light maxima are from $-$0.1 to 0.1. As the orbital period and the temperature of the primary component decrease, the difference between the two light maxima becomes more pronounced. Based on the relationships between transfer parameter and luminosity ratio, as well as between luminosity ratio and mass ratio, we found that A-, W-, and H-type contact binaries are distributed in distinct regions.

astro-ph.SR↗

arXiv2025

Prompt-Driven Building Footprint Extraction in Aerial Images with Offset-Building Model

More accurate extraction of invisible building footprints from very-high-resolution (VHR) aerial images relies on roof segmentation and roof-to-footprint offset extraction. Existing methods based on instance segmentation suffer from poor generalization when extended to large-scale data production and fail to achieve low-cost human interaction. This prompt paradigm inspires us to design a promptable framework for roof and offset extraction, and transforms end-to-end algorithms into promptable methods. Within this framework, we propose a novel Offset-Building Model (OBM). Based on prompt prediction, we first discover a common pattern of predicting offsets and tailored Distance-NMS (DNMS) algorithms for offset optimization. To rigorously evaluate the algorithm's capabilities, we introduce a prompt-based evaluation method, where our model reduces offset errors by 16.6\% and improves roof Intersection over Union (IoU) by 10.8\% compared to other models. Leveraging the common patterns in predicting offsets, DNMS algorithms enable models to further reduce offset vector loss by 6.5\%. To further validate the generalization of models, we tested them using a newly proposed test set, Huizhou test set, with over 7,000 manually annotated instance samples. Our algorithms and dataset will be available at https://github.com/likaiucas/OBM.

cs.CV↗

arXiv2025

RUN: Reversible Unfolding Network for Concealed Object Segmentation

Existing concealed object segmentation (COS) methods frequently utilize reversible strategies to address uncertain regions. However, these approaches are typically restricted to the mask domain, leaving the potential of the RGB domain underexplored. To address this, we propose the Reversible Unfolding Network (RUN), which applies reversible strategies across both mask and RGB domains through a theoretically grounded framework, enabling accurate segmentation. RUN first formulates a novel COS model by incorporating an extra residual sparsity constraint to minimize segmentation uncertainties. The iterative optimization steps of the proposed model are then unfolded into a multistage network, with each step corresponding to a stage. Each stage of RUN consists of two reversible modules: the Segmentation-Oriented Foreground Separation (SOFS) module and the Reconstruction-Oriented Background Extraction (ROBE) module. SOFS applies the reversible strategy at the mask level and introduces Reversible State Space to capture non-local information. ROBE extends this to the RGB domain, employing a reconstruction network to address conflicting foreground and background regions identified as distortion-prone areas, which arise from their separate estimation by independent modules. As the stages progress, RUN gradually facilitates reversible modeling of foreground and background in both the mask and RGB domains, directing the network's attention to uncertain regions and mitigating false-positive and false-negative results. Extensive experiments demonstrate the superior performance of RUN and highlight the potential of unfolding-based frameworks for COS and other high-level vision tasks. We will release the code and models.

cs.CV↗

arXiv2025

Light communicative materials

The natural interactive materials under far-from-equilibrium conditions have significantly inspired advances in synthetic biomimetic materials. In artificial systems, gradient diffusion serves as the primary means of interaction between individuals, lacking directionality, sufficient interaction ranges and transmission rates. Here, we present a method for constructing highly directed, communicative structures via optical feedback in light responsive materials. We showcase a photomechanical operator system comprising a baffle and a soft actuator. Positive and negative operators are configured to induce light-triggered deformations, alternately interrupting the passage of two light beams in a closed feedback loop. The fundamental functionalities of this optically interconnected material loop include homeostasis-like self-oscillation and signal transmission from one material to another via light. Refinements in alignment facilitate remote sensing, fiber-optic/long-distance communication, and adaptation. These proof-of-concept demonstrations outline a versatile design framework for light-mediated communication among responsive materials, with broad applicability across diverse materials.

physics.optics↗

arXiv2025

SayAnything: Audio-Driven Lip Synchronization with Conditional Video Diffusion

Recent advances in diffusion models have led to significant progress in audio-driven lip synchronization. However, existing methods typically rely on constrained audio-visual alignment priors or multi-stage learning of intermediate representations to force lip motion synthesis. This leads to complex training pipelines and limited motion naturalness. In this paper, we present SayAnything, a conditional video diffusion framework that directly synthesizes lip movements from audio input while preserving speaker identity. Specifically, we propose three specialized modules including identity preservation module, audio guidance module, and editing control module. Our novel design effectively balances different condition signals in the latent space, enabling precise control over appearance, motion, and region-specific generation without requiring additional supervision signals or intermediate representations. Extensive experiments demonstrate that SayAnything generates highly realistic videos with improved lip-teeth coherence, enabling unseen characters to say anything, while effectively generalizing to animated characters.

cs.CV↗

arXiv2025

Hybrid Offline-online Scheduling Method for Large Language Model Inference Optimization

With the development of large language models (LLMs), it has become increasingly important to optimize hardware usage and improve throughput. In this paper, we study the inference optimization of the serving system that deploys LLMs. To optimize system throughput and maximize hardware utilization, we formulate the inference optimization problem as a mixed-integer programming (MIP) model and propose a hybrid offline-online method as solution. The offline method improves large-scale inference systems by introducing a Minimizing Makespan Bin Packing Problem. We further provide a theoretical lower bound computation method. Then, we propose an online sorting and preemptive scheduling method to better utilize hardware. In the online iteration scheduling process, a Lagrangian method is applied to evaluate the cost efficiency of inserting prefill stages versus decode stages at each iteration and dynamically determine when to preempt decoding tasks and insert prefill tasks. Experiments using real-world data from the LLaMA-65B model and the GSM8K dataset demonstrate that system utilization improves from 80.2% to 89.1%, and the total inference time decreases from 201.00 to 190.58 seconds. A 100-cases study shows that our method consistently outperforms the baseline method and improves the utilization rate by 8.0% on average. Finally, we discuss potential future extensions, including stochastic modeling, reinforcement learning-based schedulers, and dynamic decision-making strategies for system throughput and hardware utilization.

cs.DC↗

arXiv2025

Evaluating Zero-Shot Long-Context LLM Compression

This study evaluates the effectiveness of zero-shot compression techniques on large language models (LLMs) under long-context. We identify the tendency for computational errors to increase under long-context when employing certain compression methods. We propose a hypothesis to explain the varied behavior of different LLM compression techniques and explore remedies to mitigate the performance decline observed in some techniques under long-context. This is a course report for COS 598D Machine Learning and Systems by Prof. Kai Li at Princeton University. Due to limited computational resources, our experiments were conducted only on LLaMA-2-7B-32K.

cs.CL↗

arXiv2025

A Tensor-Train Decomposition based Compression of LLMs on Group Vector Systolic Accelerator

Large language models (LLMs) are both storage-intensive and computation-intensive, posing significant challenges when deployed on resource-constrained hardware. As linear layers in LLMs are mainly resource consuming parts, this paper develops a tensor-train decomposition (TTD) for LLMs with a further hardware implementation on FPGA. TTD compression is applied to the linear layers in ChatGLM3-6B and LLaMA2-7B models with compression ratios (CRs) for the whole network 1.94$\times$ and 1.60$\times$, respectively. The compressed LLMs are further implemented on FPGA hardware within a highly efficient group vector systolic array (GVSA) architecture, which has DSP-shared parallel vector PEs for TTD inference, as well as optimized data communication in the accelerator. Experimental results show that the corresponding TTD based LLM accelerator implemented on FPGA achieves 1.45$\times$ and 1.57$\times$ reduction in first token delay for ChatGLM3-6B and LLaMA2-7B models, respectively.

cs.AR↗

arXiv2025

Complementary Subspace Low-Rank Adaptation of Vision-Language Models for Few-Shot Classification

Vision language model (VLM) has been designed for large scale image-text alignment as a pretrained foundation model. For downstream few shot classification tasks, parameter efficient fine-tuning (PEFT) VLM has gained much popularity in the computer vision community. PEFT methods like prompt tuning and linear adapter have been studied for fine-tuning VLM while low rank adaptation (LoRA) algorithm has rarely been considered for few shot fine-tuning VLM. The main obstacle to use LoRA for few shot fine-tuning is the catastrophic forgetting problem. Because the visual language alignment knowledge is important for the generality in few shot learning, whereas low rank adaptation interferes with the most informative direction of the pretrained weight matrix. We propose the complementary subspace low rank adaptation (Comp-LoRA) method to regularize the catastrophic forgetting problem in few shot VLM finetuning. In detail, we optimize the low rank matrix in the complementary subspace, thus preserving the general vision language alignment ability of VLM when learning the novel few shot information. We conduct comparison experiments of the proposed Comp-LoRA method and other PEFT methods on fine-tuning VLM for few shot classification. And we also present the suppression on the catastrophic forgetting problem of our proposed method against directly applying LoRA to VLM. The results show that the proposed method surpasses the baseline method by about +1.0\% Top-1 accuracy and preserves the VLM zero-shot performance over the baseline method by about +1.3\% Top-1 accuracy.

cs.CV↗