Ming Tang

Chao Huang, Geng Tian, Ming Tang

Federated learning (FL) enables collaborative model training across distributed clients (e.g., edge devices) without sharing raw data. Yet, FL can be computationally expensive as the clients need to train the entire model multiple times. SplitFed learning (SFL) is a recent distributed approach that alleviates computation workload at the client device by splitting the model at a cut layer into two parts, where clients only need to train part of the model. However, SFL still suffers from the \textit{client drift} problem when clients' data are highly non-IID. To address this issue, we propose MiniBatch-SFL. This algorithm incorporates MiniBatch SGD into SFL, where the clients train the client-side model in an FL fashion while the server trains the server-side model similar to MiniBatch SGD. We analyze the convergence of MiniBatch-SFL and show that the bound of the expected loss can be obtained by analyzing the expected server-side and client-side model updates, respectively. The server-side updates do not depend on the non-IID degree of the clients' datasets and can potentially mitigate client drift. However, the client-side model relies on the non-IID degree and can be optimized by properly choosing the cut layer. Perhaps counter-intuitive, our empirical result shows that a latter position of the cut layer leads to a smaller average gradient divergence and a better algorithm performance. Moreover, numerical results show that MiniBatch-SFL achieves higher accuracy than conventional SFL and FL. The accuracy improvement can be up to 24.1\% and 17.1\% with highly non-IID data, respectively.

Xing Liu, Lizhuo Luo, Ming Tang et al.

Distributed inference serves as a promising approach to enabling the inference of large language models (LLMs) at the network edge. It distributes the inference process to multiple devices to ensure that the LLMs can fit into the device memory. Recent pipeline-based approaches have the potential to parallelize communication and computation, which helps reduce inference latency. However, the benefit diminishes when the inference request at the network edge is sparse, where pipeline is typically at low utilization. To enable efficient distributed LLM inference at the edge, we propose \textbf{FlowSpec}, a pipeline-parallel tree-based speculative decoding framework. FlowSpec incorporates three key mechanisms to improve decoding efficiency: 1) score-based step-wise verification prioritizes more important draft tokens to bring earlier accpeted tokens; 2) efficient draft management to prune invalid tokens while maintaining correct causal relationship during verification; 3) dynamic draft expansion strategies to supply high-quality speculative inputs. These techniques work in concert to enhance both pipeline utilization and speculative efficiency. We evaluate FlowSpec on a real-world testbed with other baselines. Experimental results demonstrate that our proposed framework significantly improves inference speed across diverse models and configurations, achieving speedup ratios 1.28$\times$-1.79$\times$ compared to baselines. Our code is publicly available at \href{https://github.com/Leosang-lx/FlowSpec#}{https://github.com/Leosang-lx/FlowSpec\#}

Yizhou Yan, Xinyu Tang, Chao Huang et al.

Federated Learning (FL) is a distributed machine learning scheme that enables clients to train a shared global model without exchanging local data. The presence of label noise can severely degrade the FL performance, and some existing studies have focused on algorithm design for label denoising. However, they ignored the important issue that clients may not apply costly label denoising strategies due to them being self-interested and having heterogeneous valuations on the FL performance. To fill this gap, we model the clients' interactions as a novel label denoising game and characterize its equilibrium. We also analyze the price of stability, which quantifies the difference in the system performance (e.g., global model accuracy, social welfare) between the equilibrium outcome and the socially optimal solution. We prove that the equilibrium outcome always leads to a lower global model accuracy than the socially optimal solution does. We further design an efficient algorithm to compute the socially optimal solution. Numerical experiments on MNIST dataset show that the price of stability increases as the clients' data become noisier, calling for an effective incentive mechanism.

Pengchao Han, Chao Huang, Geng Tian et al.

Split federated learning (SFL) is a recent distributed approach for collaborative model training among multiple clients. In SFL, a global model is typically split into two parts, where clients train one part in a parallel federated manner, and a main server trains the other. Despite the recent research on SFL algorithm development, the convergence analysis of SFL is missing in the literature, and this paper aims to fill this gap. The analysis of SFL can be more challenging than that of federated learning (FL), due to the potential dual-paced updates at the clients and the main server. We provide convergence analysis of SFL for strongly convex and general convex objectives on heterogeneous data. The convergence rates are $O(1/T)$ and $O(1/\sqrt[3]{T})$, respectively, where $T$ denotes the total number of rounds for SFL training. We further extend the analysis to non-convex objectives and the scenario where some clients may be unavailable during training. Experimental experiments validate our theoretical results and show that SFL outperforms FL and split learning (SL) when data is highly heterogeneous across a large number of clients.

Guanghao Li, Zhihui Fu, Min Fang et al.

As large language models (LLMs) scale up, accuracy improves, but the autoregressive (AR) nature of decoding increases latency since each token requires a serial forward pass. Speculative decoding addresses this by employing a fast drafter to propose multi-token drafts, which are then verified in parallel by the target model. However, many deployments still rely on AR drafters, where sequential passes limit wall-clock gains. We revisit the drafting stage and present DiffuSpec, a training-free drop-in framework that uses a pretrained diffusion language model (DLM) to produce multi-token drafts in a single forward pass, while remaining compatible with standard AR verifiers. Because DLM drafts are generated under bidirectional conditioning, parallel per-position candidates form a token lattice in which the locally highest-probability token at each position need not form a causal left-to-right path. Moreover, DLM drafting requires pre-specifying a draft length, inducing a speed-quality trade-off. To address these challenges, we introduce two practical components: (i) a causal-consistency path search (CPS) over this lattice that extracts a left-to-right path aligned with AR verification; and (ii) an adaptive draft-length (ADL) controller that adjusts next proposal size based on recent acceptance feedback and realized generated length. Across benchmarks, DiffuSpec yields up to 3x wall-clock speedup, establishing diffusion-based drafting as a robust alternative to autoregressive drafters for speculative decoding.