arXiv subjects

Yebo Wu

Publications and source records attributed to Yebo Wu.

12 recordsLinked to original sources

arXiv2026

Spotlight and Shadow: Attention-Guided Dual-Anchor Introspective Decoding for MLLM Hallucination Mitigation

Multimodal Large Language Models (MLLMs) have demonstrated remarkable reasoning capabilities yet continue to suffer from hallucination, where generated text contradicts visual content. In this paper, we introduce Dual-Anchor Introspective Decoding (DaID), a novel contrastive decoding framework that dynamically calibrates each token generation by mining the model's internal perceptual discrepancies. Specifically, DaID identifies a Spotlight layer to amplify visual factual signals and a Shadow layer to suppress textual inertia. By leveraging visual attention distributions to guide this dual-anchor selection process, our method ensures precise, token-specific adaptation. Experimental results across multiple benchmarks and MLLMs demonstrate that DaID significantly mitigates hallucination while enhancing general reasoning capabilities.

cs.CV↗

arXiv2026

Beyond End-to-End: Dynamic Chain Optimization for Private LLM Adaptation on the Edge

Federated fine-tuning enables privacy-preserving LLM adaptation but faces a critical bottleneck: the disparity between LLMs' high memory demands and edge devices' limited capacity. To break the memory barrier, we propose Chain Federated Fine-Tuning (ChainFed), an innovative paradigm that forgoes end-to-end updates in favor of a sequential, layer-by-layer manner. It first trains the initial adapter to convergence, freezes its weights, and then proceeds to the next. This iterative train-and-freeze process forms an optimization chain, gradually enhancing the model's task-specific proficiency. ChainFed further integrates three core techniques: 1) Dynamic Layer Co-Tuning to bridge semantic gaps between sequentially tuned layers and facilitate information flow; 2) Globally Perceptive Optimization to endow each adapter with foresight beyond its local objective; 3) Function-Oriented Adaptive Tuning to automatically identify the optimal fine-tuning starting point. Extensive experiments on multiple benchmarks demonstrate the superiority of ChainFed over existing methods, boosting average accuracy by up to 46.46\%.

cs.DC↗

arXiv2026

TSEmbed: Unlocking Task Scaling in Universal Multimodal Embeddings

Despite the exceptional reasoning capabilities of Multimodal Large Language Models (MLLMs), their adaptation into universal embedding models is significantly impeded by task conflict. To address this, we propose TSEmbed, a universal multimodal embedding framework that synergizes Mixture-of-Experts (MoE) with Low-Rank Adaptation (LoRA) to explicitly disentangle conflicting task objectives. Moreover, we introduce Expert-Aware Negative Sampling (EANS), a novel strategy that leverages expert routing distributions as an intrinsic proxy for semantic similarity. By dynamically prioritizing informative hard negatives that share expert activation patterns with the query, EANS effectively sharpens the model's discriminative power and refines embedding boundaries. To ensure training stability, we further devise a two-stage learning paradigm that solidifies expert specialization before optimizing representations via EANS. TSEmbed achieves state-of-the-art performance on both the Massive Multimodal Embedding Benchmark (MMEB) and real-world industrial production datasets, laying a foundation for task-level scaling in universal multimodal embeddings.

cs.CL↗

arXiv2026

Floe: Federated Specialization for Real-Time LLM-SLM Inference

Deploying large language models (LLMs) in real-time systems remains challenging due to their substantial computational demands and privacy concerns. We propose Floe, a hybrid federated learning framework designed for latency-sensitive, resource-constrained environments. Floe combines a cloud-based black-box LLM with lightweight small language models (SLMs) on edge devices to enable low-latency, privacy-preserving inference. Personal data and fine-tuning remain on-device, while the cloud LLM contributes general knowledge without exposing proprietary weights. A heterogeneity-aware LoRA adaptation strategy enables efficient edge deployment across diverse hardware, and a logit-level fusion mechanism enables real-time coordination between edge and cloud models. Extensive experiments demonstrate that Floe enhances user privacy and personalization. Moreover, it significantly improves model performance and reduces inference latency on edge devices under real-time constraints compared with baseline approaches.

cs.DC↗

arXiv2025

Elastic Mixture of Rank-Wise Experts for Knowledge Reuse in Federated Fine-Tuning

Federated fine-tuning offers a promising solution for adapting Large Language Models (LLMs) to downstream tasks while safeguarding data privacy. However, its high computational and communication demands hinder its deployment on resource-constrained devices. In this paper, we propose SmartFed, a resource-efficient federated fine-tuning framework. SmartFed intelligently reuses knowledge embedded in existing LoRA modules, eliminating the need for expensive training from scratch when adapting LLMs to new tasks. To effectively exploit this knowledge and ensure scalability, we introduce the Mixture of Rank-Wise Experts (MoRE). MoRE decomposes LoRA modules into fine-grained rank-level experts. These experts are selectively activated and combined based on input semantics and resource budgets. Moreover, to optimize resource utilization, we present the Elastic Expert Quota Allocation (EEQA). EEQA adaptively allocates expert capacity across parameter matrices based on their contribution to model performance, focusing computing resources on the critical experts. Extensive evaluations across multiple benchmarks demonstrate that SmartFed significantly outperforms existing methods in model performance and training efficiency.

cs.DC↗

arXiv2025

Memory-Efficient Federated Fine-Tuning of Large Language Models via Layer Pruning

Federated fine-tuning enables privacy-preserving Large Language Model (LLM) adaptation, but its high memory cost limits participation from resource-constrained devices. We propose FedPruner, an innovative federated fine-tuning paradigm that tackles this via intelligent layer pruning. FedPruner flexibly prunes the global model, creating personalized submodels based on device memory constraints. It employs a macro-micro synergistic pruning framework: a macro-level functionality-driven layer orchestration mechanism groups layers, while a micro-level importance-aware layer selection strategy prunes within groups to build device-specific submodels. We further introduce a fine-grained variant that independently prunes Multi-Head Attention and Feed-Forward Network components to precisely preserve critical architectural elements. Extensive experimental results demonstrate that FedPruner significantly outperforms state-of-the-art approaches, achieving up to a 1.98\% improvement in average model accuracy while reducing peak memory usage by 75\%.

cs.DC↗

arXiv2025

Learning Like Humans: Resource-Efficient Federated Fine-Tuning through Cognitive Developmental Stages

Federated fine-tuning enables Large Language Models (LLMs) to adapt to downstream tasks while preserving data privacy, but its resource-intensive nature limits deployment on edge devices. In this paper, we introduce Developmental Federated Tuning (DevFT), a resource-efficient approach inspired by cognitive development that progressively builds a powerful LLM from a compact foundation. DevFT decomposes the fine-tuning process into developmental stages, each optimizing submodels with increasing parameter capacity. Knowledge from earlier stages transfers to subsequent submodels, providing optimized initialization parameters that prevent convergence to local minima and accelerate training. This paradigm mirrors human learning, gradually constructing comprehensive knowledge structure while refining existing skills. To efficiently build stage-specific submodels, DevFT introduces deconfliction-guided layer grouping and differential-based layer fusion to distill essential information and construct representative layers. Evaluations across multiple benchmarks demonstrate that DevFT significantly outperforms state-of-the-art methods, achieving up to 4.59$\times$ faster convergence, 10.67$\times$ reduction in communication overhead, and 9.07% average performance improvement, while maintaining compatibility with existing approaches.

cs.LG↗

arXiv2025

A Survey on Federated Fine-tuning of Large Language Models

Large Language Models (LLMs) have demonstrated impressive success across various tasks. Integrating LLMs with Federated Learning (FL), a paradigm known as FedLLM, offers a promising avenue for collaborative model adaptation while preserving data privacy. This survey provides a systematic and comprehensive review of FedLLM. We begin by tracing the historical development of both LLMs and FL, summarizing relevant prior research to set the context. Subsequently, we delve into an in-depth analysis of the fundamental challenges inherent in deploying FedLLM. Addressing these challenges often requires efficient adaptation strategies; therefore, we conduct an extensive examination of existing Parameter-Efficient Fine-tuning (PEFT) methods and explore their applicability within the FL framework. To rigorously evaluate the performance of FedLLM, we undertake a thorough review of existing fine-tuning datasets and evaluation benchmarks. Furthermore, we discuss FedLLM's diverse real-world applications across multiple domains. Finally, we identify critical open challenges and outline promising research directions to foster future advancements in FedLLM. This survey aims to serve as a foundational resource for researchers and practitioners, offering valuable insights into the rapidly evolving landscape of federated fine-tuning for LLMs. It also establishes a roadmap for future innovations in privacy-preserving AI. We actively maintain a \href{https://github.com/Clin0212/Awesome-Federated-LLM-Learning}{GitHub repo} to track cutting-edge advancements in this field.

cs.LG↗

arXiv2024

Breaking the Memory Wall for Heterogeneous Federated Learning via Model Splitting

Federated Learning (FL) enables multiple devices to collaboratively train a shared model while preserving data privacy. Ever-increasing model complexity coupled with limited memory resources on the participating devices severely bottlenecks the deployment of FL in real-world scenarios. Thus, a framework that can effectively break the memory wall while jointly taking into account the hardware and statistical heterogeneity in FL is urgently required. In this paper, we propose SmartSplit, a framework that effectively reduces the memory footprint on the device side while guaranteeing the training progress and model accuracy for heterogeneous FL through model splitting.Towards this end, SmartSplit employs a hierarchical structure to adaptively guide the overall training process. In each training round, the central manager, hosted on the server, dynamically selects the participating devices and sets the cutting layer by jointly considering the memory budget, training capacity, and data distribution of each device. The MEC manager, deployed within the edge server, proceeds to split the local model and perform training of the server-side portion. Meanwhile, it fine-tunes the splitting points based on the time-evolving statistical importance. The on-device manager, embedded inside each mobile device, continuously monitors the local training status while employing cost-aware checkpointing to match the runtime dynamic memory budget. Extensive experiments on representative datasets are conducted on both commercial off-the-shelf mobile device testbeds. The experimental results show that SmartSplit excels in FL training on highly memory-constrained mobile SoCs, offering up to a 94% peak latency reduction and 100-fold memory savings. It enhances accuracy performance by 1.49%-57.18% and adaptively adjusts to dynamic memory budgets through cost-aware recomputation.

cs.DC↗

arXiv2024

Heterogeneity-Aware Coordination for Federated Learning via Stitching Pre-trained blocks

Federated learning (FL) coordinates multiple devices to collaboratively train a shared model while preserving data privacy. However, large memory footprint and high energy consumption during the training process excludes the low-end devices from contributing to the global model with their own data, which severely deteriorates the model performance in real-world scenarios. In this paper, we propose FedStitch, a hierarchical coordination framework for heterogeneous federated learning with pre-trained blocks. Unlike the traditional approaches that train the global model from scratch, for a new task, FedStitch composes the global model via stitching pre-trained blocks. Specifically, each participating client selects the most suitable block based on their local data from the candidate pool composed of blocks from pre-trained models. The server then aggregates the optimal block for stitching. This process iterates until a new stitched network is generated. Except for the new training paradigm, FedStitch consists of the following three core components: 1) an RL-weighted aggregator, 2) a search space optimizer deployed on the server side, and 3) a local energy optimizer deployed on each participating client. The RL-weighted aggregator helps to select the right block in the non-IID scenario, while the search space optimizer continuously reduces the size of the candidate block pool during stitching. Meanwhile, the local energy optimizer is designed to minimize energy consumption of each client while guaranteeing the overall training progress. The results demonstrate that compared to existing approaches, FedStitch improves the model accuracy up to 20.93%. At the same time, it achieves up to 8.12% speedup, reduces the memory footprint up to 79.5%, and achieves 89.41% energy saving at most during the learning procedure.

cs.LG↗

arXiv2024

Bridging Memory Gaps: Scaling Federated Learning for Heterogeneous Clients

Federated Learning (FL) enables multiple clients to collaboratively train a shared model while preserving data privacy. However, the high memory demand during model training severely limits the deployment of FL on resource-constrained clients. To this end, we propose \our, a scalable and inclusive FL framework designed to overcome memory limitations through sequential block-wise training. The core idea of \our is to partition the global model into blocks and train them sequentially, thereby reducing training memory requirements. To mitigate information loss during block-wise training, \our introduces a Curriculum Mentor that crafts curriculum-aware training objectives for each block to steer their learning process. Moreover, \our incorporates a Training Harmonizer that designs a parameter co-adaptation training scheme to coordinate block updates, effectively breaking inter-block information isolation. Extensive experiments on both simulation and hardware testbeds demonstrate that \our significantly improves model performance by up to 84.2\%, reduces peak memory usage by up to 50.4\%, and accelerates training by up to 1.9$\times$.

cs.DC↗

arXiv2024

Breaking the Memory Wall for Heterogeneous Federated Learning via Progressive Training

This paper presents ProFL, a new framework that effectively addresses the memory constraints in FL. Rather than updating the full model during local training, ProFL partitions the model into blocks based on its original architecture and trains each block in a progressive fashion. It first trains the front blocks and safely freezes them after convergence. Training of the next block is then triggered. This process progressively grows the model to be trained until the training of the full model is completed. In this way, the peak memory footprint is effectively reduced for feasible deployment on heterogeneous devices. In order to preserve the feature representation of each block, the training process is divided into two stages: model shrinking and model growing. During the model shrinking stage, we meticulously design corresponding output modules to assist each block in learning the expected feature representation and obtain the initialization model parameters. Subsequently, the obtained output modules and initialization model parameters are utilized in the corresponding model growing stage, which progressively trains the full model. Additionally, a novel metric from the scalar perspective is proposed to assess the learning status of each block, enabling us to securely freeze it after convergence and initiate the training of the next one. Finally, we theoretically prove the convergence of ProFL and conduct extensive experiments on representative models and datasets to evaluate its effectiveness. The results demonstrate that ProFL effectively reduces the peak memory footprint by up to 57.4% and improves model accuracy by up to 82.4%.

cs.DC↗