Mingjia Shi

Yuhao Zhou, Mingjia Shi, Yuanxi Li et al.

Reducing communication overhead in federated learning (FL) is challenging but crucial for large-scale distributed privacy-preserving machine learning. While methods utilizing sparsification or others can largely lower the communication overhead, the convergence rate is also greatly compromised. In this paper, we propose a novel method, named single-step synthetic features compressor (3SFC), to achieve communication-efficient FL by directly constructing a tiny synthetic dataset based on raw gradients. Thus, 3SFC can achieve an extremely low compression rate when the constructed dataset contains only one data sample. Moreover, 3SFC's compressing phase utilizes a similarity-based objective function so that it can be optimized with just one step, thereby considerably improving its performance and robustness. In addition, to minimize the compressing error, error feedback (EF) is also incorporated into 3SFC. Experiments on multiple datasets and models suggest that 3SFC owns significantly better convergence rates compared to competing methods with lower compression rates (up to 0.02%). Furthermore, ablation studies and visualizations show that 3SFC can carry more information than competing methods for every communication round, further validating its effectiveness.

Yinhan He, Yaochen Zhu, Mingjia Shi et al.

Large language models increasingly rely on long chains of thought to improve accuracy, yet such gains come with substantial inference-time costs. We revisit token-efficient post-training and argue that existing sequence-level reward-shaping methods offer limited control over how reasoning effort is allocated across tokens. To bridge the gap, we propose IAPO, an information-theoretic post-training framework that assigns token-wise advantages based on each token's conditional mutual information (MI) with the final answer. This yields an explicit, principled mechanism for identifying informative reasoning steps and suppressing low-utility exploration. We provide a theoretical analysis showing that our IAPO can induce monotonic reductions in reasoning verbosity without harming correctness. Empirically, IAPO consistently improves reasoning accuracy while reducing reasoning length by up to 36%, outperforming existing token-efficient RL methods across various reasoning datasets. Extensive empirical evaluations demonstrate that information-aware advantage shaping is a powerful and general direction for token-efficient post-training. The code is available at https://github.com/YinhanHe123/IAPO.

Mingjia Shi, Yuhao Zhou, Qing Ye et al.

Federated learning (FL for simplification) is a distributed machine learning technique that utilizes global servers and collaborative clients to achieve privacy-preserving global model training without direct data sharing. However, heterogeneous data problem, as one of FL's main problems, makes it difficult for the global model to perform effectively on each client's local data. Thus, personalized federated learning (PFL for simplification) aims to improve the performance of the model on local data as much as possible. Bayesian learning, where the parameters of the model are seen as random variables with a prior assumption, is a feasible solution to the heterogeneous data problem due to the tendency that the more local data the model use, the more it focuses on the local data, otherwise focuses on the prior. When Bayesian learning is applied to PFL, the global model provides global knowledge as a prior to the local training process. In this paper, we employ Bayesian learning to model PFL by assuming a prior in the scaled exponential family, and therefore propose pFedBreD, a framework to solve the problem we model using Bregman divergence regularization. Empirically, our experiments show that, under the prior assumption of the spherical Gaussian and the first order strategy of mean selection, our proposal significantly outcompetes other PFL algorithms on multiple public benchmarks.

Mingjia Shi, Yuhao Zhou, Kai Wang et al.

Classical federated learning (FL) enables training machine learning models without sharing data for privacy preservation, but heterogeneous data characteristic degrades the performance of the localized model. Personalized FL (PFL) addresses this by synthesizing personalized models from a global model via training on local data. Such a global model may overlook the specific information that the clients have been sampled. In this paper, we propose a novel scheme to inject personalized prior knowledge into the global model in each client, which attempts to mitigate the introduced incomplete information problem in PFL. At the heart of our proposed approach is a framework, the PFL with Bregman Divergence (pFedBreD), decoupling the personalized prior from the local objective function regularized by Bregman divergence for greater adaptability in personalized scenarios. We also relax the mirror descent (RMD) to extract the prior explicitly to provide optional strategies. Additionally, our pFedBreD is backed up by a convergence analysis. Sufficient experiments demonstrate that our method reaches the state-of-the-art performances on 5 datasets and outperforms other methods by up to 3.5% across 8 benchmarks. Extensive analyses verify the robustness and necessity of proposed designs.

Xinghao Wu, Jianwei Niu, Xuefeng Liu et al.

In traditional Federated Learning approaches like FedAvg, the global model underperforms when faced with data heterogeneity. Personalized Federated Learning (PFL) enables clients to train personalized models to fit their local data distribution better. However, we surprisingly find that the feature extractor in FedAvg is superior to those in most PFL methods. More interestingly, by applying a linear transformation on local features extracted by the feature extractor to align with the classifier, FedAvg can surpass the majority of PFL methods. This suggests that the primary cause of FedAvg's inadequate performance stems from the mismatch between the locally extracted features and the classifier. While current PFL methods mitigate this issue to some extent, their designs compromise the quality of the feature extractor, thus limiting the full potential of PFL. In this paper, we propose a new PFL framework called FedPFT to address the mismatch problem while enhancing the quality of the feature extractor. FedPFT integrates a feature transformation module, driven by personalized prompts, between the global feature extractor and classifier. In each round, clients first train prompts to transform local features to match the global classifier, followed by training model parameters. This approach can also align the training objectives of clients, reducing the impact of data heterogeneity on model collaboration. Moreover, FedPFT's feature transformation module is highly scalable, allowing for the use of different prompts to tailor local features to various tasks. Leveraging this, we introduce a collaborative contrastive learning task to further refine feature extractor quality. Our experiments demonstrate that FedPFT outperforms state-of-the-art methods by up to 7.08%.

Ziqiao Wang, Wangbo Zhao, Yuhao Zhou et al.

Diffusion Transformers (DiTs) deliver state-of-the-art image quality, yet their training remains notoriously slow. A recent remedy -- representation alignment (REPA) that matches DiT hidden features to those of a non-generative teacher (e.g. DINO) -- dramatically accelerates the early epochs but plateaus or even degrades performance later. We trace this failure to a capacity mismatch: once the generative student begins modelling the joint data distribution, the teacher's lower-dimensional embeddings and attention patterns become a straitjacket rather than a guide. We then introduce HASTE (Holistic Alignment with Stage-wise Termination for Efficient training), a two-phase schedule that keeps the help and drops the hindrance. Phase I applies a holistic alignment loss that simultaneously distills attention maps (relational priors) and feature projections (semantic anchors) from the teacher into mid-level layers of the DiT, yielding rapid convergence. Phase II then performs one-shot termination that deactivates the alignment loss, once a simple trigger such as a fixed iteration is hit, freeing the DiT to focus on denoising and exploit its generative capacity. HASTE speeds up training of diverse DiTs without architecture changes. On ImageNet 256X256, it reaches the vanilla SiT-XL/2 baseline FID in 50 epochs and matches REPA's best FID in 500 epochs, amounting to a 28X reduction in optimization steps. HASTE also improves text-to-image DiTs on MS-COCO, demonstrating to be a simple yet principled recipe for efficient diffusion training across various tasks. Our code is available at https://github.com/NUS-HPC-AI-Lab/HASTE .

Mingjia Shi, Yinhan He, Yaochen Zhu et al.

Vision-language models (VLMs) aim to reason by jointly leveraging visual and textual modalities. While allocating additional inference-time computation has proven effective for large language models (LLMs), achieving similar scaling in VLMs remains challenging. A key obstacle is that visual inputs are typically provided only once at the start of generation, while textual reasoning (e.g., early visual summaries) is generated autoregressively, causing reasoning to become increasingly text-dominated and allowing early visual grounding errors to accumulate. Moreover, vanilla guidance for visual grounding during inference is often coarse and noisy, making it difficult to steer reasoning over long texts. To address these challenges, we propose \emph{Saliency-Aware Principle} (SAP) selection. SAP operates on high-level reasoning principles rather than token-level trajectories, which enable stable control over discrete generation under noisy feedback while allowing later reasoning steps to re-consult visual evidence when renewed grounding is required. In addition, SAP supports multi-route inference, enabling parallel exploration of diverse reasoning behaviors. SAP is model-agnostic and data-free, requiring no additional training. Empirical results show that SAP achieves competitive performance, especially in reducing object hallucination, under comparable token-generation budgets while yielding more stable reasoning and lower response latency than CoT-style long sequential reasoning.

Sifan Zhou, Shuo Wang, Zhihang Yuan et al.

Large Language Models (LLMs) fine-tuning technologies have achieved remarkable results. However, traditional LLM fine-tuning approaches face significant challenges: they require large Floating Point (FP) computation, raising privacy concerns when handling sensitive data, and are impractical for resource-constrained edge devices. While Parameter-Efficient Fine-Tuning (PEFT) techniques reduce trainable parameters, their reliance on floating-point arithmetic creates fundamental incompatibilities with edge hardware. In this work, we introduce a novel framework for on-device LLM fine-tuning that eliminates the need for floating-point operations in both inference and training, named GSQ-Tuning. At its core is the Group-Shared Exponents Integer format, which efficiently represents model parameters in integer format using shared exponents among parameter groups. When combined with LoRA-like adapters, this enables fully integer-based fine-tuning that is both memory and compute efficient. We demonstrate that our approach achieves accuracy comparable to BF16-based fine-tuning while significantly reducing 1.85x memory usage. Moreover, compared to FP8, our method can reduce 5x power consumption and 11x chip area with same performance, making large-scale model adaptation feasible on edge devices.

Zekai Li, Xinhao Zhong, Samir Khaki et al.

In recent years, dataset distillation has provided a reliable solution for data compression, where models trained on the resulting smaller synthetic datasets achieve performance comparable to those trained on the original datasets. To further improve the performance of synthetic datasets, various training pipelines and optimization objectives have been proposed, greatly advancing the field of dataset distillation. Recent decoupled dataset distillation methods introduce soft labels and stronger data augmentation during the post-evaluation phase and scale dataset distillation up to larger datasets (e.g., ImageNet-1K). However, this raises a question: Is accuracy still a reliable metric to fairly evaluate dataset distillation methods? Our empirical findings suggest that the performance improvements of these methods often stem from additional techniques rather than the inherent quality of the images themselves, with even randomly sampled images achieving superior results. Such misaligned evaluation settings severely hinder the development of DD. Therefore, we propose DD-Ranking, a unified evaluation framework, along with new general evaluation metrics to uncover the true performance improvements achieved by different methods. By refocusing on the actual information enhancement of distilled datasets, DD-Ranking provides a more comprehensive and fair evaluation standard for future research advancements.

Zhiyuan Liang, Dongwen Tang, Yuhao Zhou et al.

Modern Parameter-Efficient Fine-Tuning (PEFT) methods such as low-rank adaptation (LoRA) reduce the cost of customizing large language models (LLMs), yet still require a separate optimization run for every downstream dataset. We introduce \textbf{Drag-and-Drop LLMs (\textit{DnD})}, a prompt-conditioned parameter generator that eliminates per-task training by mapping a handful of unlabeled task prompts directly to LoRA weight updates. A lightweight text encoder distills each prompt batch into condition embeddings, which are then transformed by a cascaded hyper-convolutional decoder into the full set of LoRA matrices. Once trained in a diverse collection of prompt-checkpoint pairs, DnD produces task-specific parameters in seconds, yielding i) up to \textbf{12,000$\times$} lower overhead than full fine-tuning, ii) average gains up to \textbf{30\%} in performance over the strongest training LoRAs on unseen common-sense reasoning, math, coding, and multimodal benchmarks, and iii) robust cross-domain generalization despite never seeing the target data or labels. Our results demonstrate that prompt-conditioned parameter generation is a viable alternative to gradient-based adaptation for rapidly specializing LLMs. Our project is available at \href{https://jerryliang24.github.io/DnD}{https://jerryliang24.github.io/DnD}.

Mingjia Shi, Yuhao Zhou, Ruiji Yu et al.

Vision Mamba has shown close to state of the art performance on computer vision tasks, drawing much interest in increasing it's efficiency. A promising approach is token reduction (that has been successfully implemented in ViTs). Pruning informative tokens in Mamba leads to a high loss of key knowledge and degraded performance. An alternative, of merging tokens preserves more information than pruning, also suffers for large compression ratios. Our key insight is that a quick round of retraining after token merging yeilds robust results across various compression ratios. Empirically, pruned Vims only drop up to 0.9% accuracy on ImageNet-1K, recovered by our proposed framework R-MeeTo in our main evaluation. We show how simple and effective the fast recovery can be achieved at minute-level, in particular, a 35.9% accuracy spike over 3 epochs of training on Vim-Ti. Moreover, Vim-Ti/S/B are re-trained within 5/7/17 minutes, and Vim-S only drops 1.3% with 1.2x (up to 1.5x) speed up in inference.

Tianyu Fu, Anyang Su, Chenxu Zhao et al.

Graphical user interfaces (GUIs) are the primary medium for human-computer interaction, yet automating GUI interactions remains challenging due to the complexity of visual elements, dynamic environments, and the need for multi-step reasoning. Existing methods based on vision-language models (VLMs) often suffer from limited resolution, domain mismatch, and insufficient sequential decisionmaking capability. To address these issues, we propose Mano, a robust GUI agent built upon a multi-modal foundation model pre-trained on extensive web and computer system data. Our approach integrates a novel simulated environment for high-fidelity data generation, a three-stage training pipeline (supervised fine-tuning, offline reinforcement learning, and online reinforcement learning), and a verification module for error recovery. Mano demonstrates state-of-the-art performance on multiple GUI benchmarks, including Mind2Web and OSWorld, achieving significant improvements in success rate and operational accuracy. Our work provides new insights into the effective integration of reinforcement learning with VLMs for practical GUI agent deployment, highlighting the importance of domain-specific data, iterative training, and holistic reward design.

Yuhao Zhou, Yuxin Tian, Mingjia Shi et al.

The exponential growth in model sizes has significantly increased the communication burden in Federated Learning (FL). Existing methods to alleviate this burden by transmitting compressed gradients often face high compression errors, which slow down the model's convergence. To simultaneously achieve high compression effectiveness and lower compression errors, we study the gradient compression problem from a novel perspective. Specifically, we propose a systematical algorithm termed Extended Single-Step Synthetic Features Compressing (E-3SFC), which consists of three sub-components, i.e., the Single-Step Synthetic Features Compressor (3SFC), a double-way compression algorithm, and a communication budget scheduler. First, we regard the process of gradient computation of a model as decompressing gradients from corresponding inputs, while the inverse process is considered as compressing the gradients. Based on this, we introduce a novel gradient compression method termed 3SFC, which utilizes the model itself as a decompressor, leveraging training priors such as model weights and objective functions. 3SFC compresses raw gradients into tiny synthetic features in a single-step simulation, incorporating error feedback to minimize overall compression errors. To further reduce communication overhead, 3SFC is extended to E-3SFC, allowing double-way compression and dynamic communication budget scheduling. Our theoretical analysis under both strongly convex and non-convex conditions demonstrates that 3SFC achieves linear and sub-linear convergence rates with aggregation noise. Extensive experiments across six datasets and six models reveal that 3SFC outperforms state-of-the-art methods by up to 13.4% while reducing communication costs by 111.6 times. These findings suggest that 3SFC can significantly enhance communication efficiency in FL without compromising model performance.

Mingjia Shi, Ruihan Lin, Xuxi Chen et al.

Learning to Optimize (L2O) enhances optimization efficiency with integrated neural networks. L2O paradigms achieve great outcomes, e.g., refitting optimizer, generating unseen solutions iteratively or directly. However, conventional L2O methods require intricate design and rely on specific optimization processes, limiting scalability and generalization. Our analyses explore general framework for learning optimization, called Diff-L2O, focusing on augmenting sampled solutions from a wider view rather than local updates in real optimization process only. Meanwhile, we give the related generalization bound, showing that the sample diversity of Diff-L2O brings better performance. This bound can be simply applied to other fields, discussing diversity, mean-variance, and different tasks. Diff-L2O's strong compatibility is empirically verified with only minute-level training, comparing with other hour-levels.

Yuhao Zhou, Yuxin Tian, Jindi Lv et al.

In the realm of high-frequency data streams, achieving real-time learning within varying memory constraints is paramount. This paper presents Ferret, a comprehensive framework designed to enhance online accuracy of Online Continual Learning (OCL) algorithms while dynamically adapting to varying memory budgets. Ferret employs a fine-grained pipeline parallelism strategy combined with an iterative gradient compensation algorithm, ensuring seamless handling of high-frequency data with minimal latency, and effectively counteracting the challenge of stale gradients in parallel training. To adapt to varying memory budgets, its automated model partitioning and pipeline planning optimizes performance regardless of memory limitations. Extensive experiments across 20 benchmarks and 5 integrated OCL algorithms show Ferret's remarkable efficiency, achieving up to 3.7$\times$ lower memory overhead to reach the same online accuracy compared to competing methods. Furthermore, Ferret consistently outperforms these methods across diverse memory budgets, underscoring its superior adaptability. These findings position Ferret as a premier solution for efficient and adaptive OCL framework in real-time environments.

Qing Ye, Yuhao Zhou, Mingjia Shi et al.

Synchronous strategies with data parallelism, such as the Synchronous StochasticGradient Descent (S-SGD) and the model averaging methods, are widely utilizedin distributed training of Deep Neural Networks (DNNs), largely owing to itseasy implementation yet promising performance. Particularly, each worker ofthe cluster hosts a copy of the DNN and an evenly divided share of the datasetwith the fixed mini-batch size, to keep the training of DNNs convergence. In thestrategies, the workers with different computational capability, need to wait foreach other because of the synchronization and delays in network transmission,which will inevitably result in the high-performance workers wasting computation.Consequently, the utilization of the cluster is relatively low. To alleviate thisissue, we propose the Dynamic Batch Size (DBS) strategy for the distributedtraining of DNNs. Specifically, the performance of each worker is evaluatedfirst based on the fact in the previous epoch, and then the batch size and datasetpartition are dynamically adjusted in consideration of the current performanceof the worker, thereby improving the utilization of the cluster. To verify theeffectiveness of the proposed strategy, extensive experiments have been conducted,and the experimental results indicate that the proposed strategy can fully utilizethe performance of the cluster, reduce the training time, and have good robustnesswith disturbance by irrelevant tasks. Furthermore, rigorous theoretical analysis hasalso been provided to prove the convergence of the proposed strategy.