Decentralized Learning

This work presents SeedFlood, a new approach to decentralized LLM fine-tuning designed to scale across large models, large collaborations, and complex network topologies while achieving global consensus with negligible communication overhead. Traditional methods suffer from high communication costs that grow with model size, while information decay over network hops renders global consensus inefficient. SeedFlood takes a significant departure from these practices by exploiting the seed-reconstructible structure of zeroth-order gradients and effectively making the messages to transmit near-zero in size, allowing them to be flooded to every client in the network, and thereby enhancing scalability of decentralized training. Consequently, SeedFlood enables training in regimes previously considered impractical, such as billion-parameter scale models or distributed across hundred of clients. Our experiments on decentralized LLM fine-tuning demonstrate that SeedFlood consistently outperforms the standard zeroth-order baselines in both communication efficiency and generalization performance, and even achieves results comparable to first-order gossip-based methods in large-scale settings, while requiring orders-of-magnitude less communication cost. We also provide theoretical analysis to formalize that SeedFlood avoids topology-dependent consensus terms in the convergence bound while retaining the acceleration enabled by increased client participation.

Decentralized training is often regarded as inferior to centralized training because the consensus errors between workers are thought to undermine convergence and generalization. This work challenges this view by introducing decentralized SGD with Adaptive Consensus (DSGD-AC), which uses a time-dependent scaling mechanism to maintain consensus errors throughout the training. We show that adaptive consensus changes the stationary variance of disagreement modes by balancing two effects: it preserves consensus-error magnitude through weaker graph damping while still allowing curvature-dependent damping to shape the disagreement directions. This balance can produce a stronger Hessian-weighted loss-envelope penalty around the deployed model, even when normalized Hessian alignment is weaker than in standard DSGD. Empirical results on image classification show that DSGD-AC reaches flatter solutions and higher test accuracy than standard DSGD and even centralized SGD. Together, these results support consensus errors as a useful implicit regularizer and open a new perspective on the design of decentralized learning algorithms.

Asynchronous primal-dual methods for decentralized non-smooth convex optimization often require each node to maintain $\mathcal{O}(d)$ auxiliary variables, where $d$ is its degree. This dependence on degree increases memory requirements and can amplify the effects of stale information, especially in dense networks. Motivated by the challenge of frugal memory management in decentralized learning, we introduce Goal-PD, an asynchronous gossip-based primal-dual algorithm that maintains only two variables per node, regardless of the node's degree. We establish almost-sure convergence of Goal-PD to a minimizer of the underlying optimization problem, and prove linear convergence when the objective functions are piecewise linear-quadratic. For decentralized mean estimation, we show that pairwise averaging is a special case of Goal-PD, which establishes a direct link between the proposed primal-dual framework and classical gossip. Experiments on synthetic and real datasets over various network topologies, with non-smooth objectives including median estimation, show that Goal-PD converges faster than existing asynchronous baselines while requiring significantly less memory by design.

Random walk (RW)-based algorithms have long been popular in distributed systems due to low overheads and scalability, with recent growing applications in decentralized learning. However, their reliance on local interactions makes them inherently vulnerable to malicious behavior. In this work, we investigate an adversarial threat that we term the ``Pac-Man'' attack, in which a malicious node probabilistically terminates any RW that visits it. This stealthy behavior gradually eliminates active RWs from the network, effectively halting the learning process without triggering failure alarms. To counter this threat, we propose the CREATE-IF-LATE (CIL) algorithm, which is a fully decentralized, resilient mechanism that enables self-creating RWs and prevents RW extinction in the presence of Pac-Man. Our theoretical analysis shows that the CIL algorithm guarantees several desirable properties, such as (i) non-extinction of the RW population, (ii) almost sure boundedness of the RW population, and (iii) convergence of RW-based stochastic gradient descent even in the presence of Pac-Man with a quantifiable deviation from the true optimum. Moreover, the learning process experiences at most a linear time delay due to Pac-Man interruptions and RW regeneration. Our extensive empirical results on both synthetic and public benchmark datasets validate our theoretical findings.

Adversarial training has been widely studied in recent years due to its role in improving model robustness against adversarial attacks. This paper focuses on comparing different distributed adversarial training algorithms--including centralized and decentralized strategies--within multi-agent learning environments. Previous studies have highlighted the importance of model flatness in determining robustness. To this end, we develop a general theoretical framework to study the escaping efficiency of these algorithms from local minima, which is closely related to the flatness of the resulting models. We show that when the perturbation bound is sufficiently small (i.e., when the attack strength is relatively mild) and a large batch size is used, decentralized adversarial training algorithms--including consensus and diffusion--are guaranteed to escape faster from local minima than the centralized strategy, thereby favoring flatter minima. However, as the perturbation bound increases, this trend may no longer hold. In the simulation results, we illustrate our theoretical findings and systematically compare the performance of models obtained through decentralized and centralized adversarial training algorithms. The results highlight the potential of decentralized strategies to enhance the robustness of models in distributed settings.

Random walk (RW)-based algorithms have long been popular in distributed systems due to low overheads and scalability, with recent growing applications in decentralized learning. However, their reliance on local interactions makes them inherently vulnerable to malicious behavior. In this work, we investigate an adversarial threat that we term the ``Pac-Man'' attack, in which a malicious node probabilistically terminates any RW that visits it. This stealthy behavior gradually eliminates active RWs from the network, effectively halting the learning process without triggering failure alarms. To counter this threat, we propose the Average Crossing (AC) algorithm--a fully decentralized mechanism for duplicating RWs to prevent RW extinction in the presence of Pac-Man. Our theoretical analysis establishes that (i) the RW population remains almost surely bounded under AC and (ii) RW-based stochastic gradient descent remains convergent under AC, even in the presence of Pac-Man, with a quantifiable deviation from the true optimum. Our extensive empirical results on both synthetic and real-world datasets corroborate our theoretical findings. Furthermore, they uncover a phase transition in the extinction probability as a function of the duplication threshold. We offer theoretical insights by analyzing a simplified variant of the AC, which sheds light on the observed phase transition.

Training large language models is generally done on clusters containing thousands of accelerators, communicating over a high-bandwidth interconnect. Scaling up these clusters is expensive and can become impractical, imposing limits on the size of models that can be trained. Several recent studies have proposed training methods that are less communication intensive, avoiding the need for compute clusters with extremely high interconnect speeds. These low communication training methods still employ a global synchronization step for model parameters, which can be too costly with a high number of participants, as the communication cost scales quadratically with group size. In this work, we propose a novel optimization method, NoLoCo, that does not explicitly synchronize all model parameters during training and does not require any collective communication. NoLoCo implicitly synchronizes model weights via a novel variant of the Nesterov momentum optimizer by partially averaging model weights within randomly selected subgroups. We provide both a theoretical convergence analysis of our optimizer and empirical results from language model training. Our method requires significantly less communication than fully sharded data parallel training and DiLoCo, a widely used low-communication baseline. Moreover, our method avoids global blocking communication, thereby reducing accelerator idle time. Our experiments show that NoLoCo is more communication-efficient than DiLoCo, improving final perplexity by up to $4\%$ and converging up to $4\times$ faster in wall-clock time across a range of worker counts, model sizes, and communication bandwidths.

Scaling models has led to significant advancements in deep learning, but training these models in decentralized settings remains challenging due to communication bottlenecks. While existing compression techniques are effective in data-parallel, they do not extend to model parallelism. Unlike data-parallel training, where weight gradients are exchanged, model-parallel requires compressing activations and activation gradients as they propagate through layers, accumulating compression errors. We propose a novel compression algorithm that compresses both forward and backward passes, enabling up to 99% compression with no convergence degradation with negligible memory/compute overhead. By leveraging a recursive structure in transformer networks, we predefine a low-dimensional subspace to confine the activations and gradients, allowing full reconstruction in subsequent layers. Our method achieves up to 100x improvement in communication efficiency and enables training billion-parameter-scale models over low-end GPUs connected via consumer-grade internet speeds as low as 80Mbps, matching the convergence of centralized datacenter systems with 100Gbps connections with model parallel.

As end-user device capability increases and demand for intelligent services at the Internet's edge rises, distributed learning has emerged as a key enabling technology for the intelligent edge. Existing approaches like federated learning (FL) and decentralized FL (DFL) enable privacy-preserving distributed learning among clients, while gossip learning (GL) approaches have emerged to address the potential challenges in resource-constrained, connectivity-challenged infrastructure-less environments. However, most distributed learning approaches assume largely homogeneous data distributions and may not consider or exploit the heterogeneity of clients and their underlying data distributions. This paper introduces Chisme, a novel fully decentralized distributed learning algorithm designed to address the challenges of implementing robust intelligence in network edge contexts characterized by heterogeneous data distributions, episodic connectivity, and sparse network infrastructure or lack thereof. Chisme leverages the affinity between clients' underlying data distributions calculated from received model exchanges to inform how much influence received models have when merging into the local model. By doing so, it enables clients to strategically balance between broader collaboration to build more general knowledge and more selective collaboration to build specific knowledge. We evaluate Chisme against contemporary approaches using image recognition and time-series prediction scenarios while considering different network connectivity conditions, representative of real-world distributed intelligent systems running at the network's edge. Our experiments demonstrate that Chisme outperforms state-of-the-art edge intelligence approaches in almost every case -- clients using Chisme exhibit faster training convergence, lower final loss after training, and lower performance disparity between clients.

Personalized collaborative learning in federated settings faces a critical trade-off between customization and participant trust. Existing approaches typically rely on centralized coordinators or trusted peer groups, limiting their applicability in open, trust-averse environments. While recent decentralized methods explore anonymous knowledge sharing, they often lack global scalability and robust mechanisms against malicious peers. To bridge this gap, we propose TPFed, a \textit{Trust-free Personalized Decentralized Federated Learning} framework. TPFed replaces central aggregators with a blockchain-based bulletin board, enabling participants to dynamically select global communication partners based on Locality-Sensitive Hashing (LSH) and peer ranking. Crucially, we introduce an ``all-in-one'' knowledge distillation protocol that simultaneously handles knowledge transfer, model quality evaluation, and similarity verification via a public reference dataset. This design ensures secure, globally personalized collaboration without exposing local models or data. Extensive experiments demonstrate that TPFed significantly outperforms traditional federated baselines in both learning accuracy and system robustness against adversarial attacks.

A novel Learning-by-Education Node Community framework (LENC) for Collaborative Knowledge Distillation (CKD) is presented, which facilitates continual collective learning through effective knowledge exchanges among diverse deployed Deep Neural Network (DNN) peer nodes. These DNNs dynamically and autonomously adopt either the role of a student, seeking knowledge, or that of a teacher, imparting knowledge, fostering a collaborative learning environment. The proposed framework triggers knowledge transfer via autonomous teacher discovery and stream-driven DNN distillation as needed, while enhancing their learning capabilities and promoting their collaboration. LENC addresses the challenges of handling diverse training data distributions and the limitations of individual DNN node learning abilities. \hl{It enables the exploitation of selected peer-teacher knowledge upon learning a new task and mitigates catastrophic forgetting in DNN nodes.} \hl{Additionally, it supports task-boundary-free continual adaptation in distributed settings via autonomous role assignment and modular forgetting mitigation, as DNN nodes receive no explicit task-boundary metadata during deployment.} Experimental evaluation on a proof-of-concept implementation demonstrates the LENC framework's functionalities and benefits across multiple DNN learning and inference scenarios. The conducted experiments showcase its ability to gradually improve the average test accuracy of the community of interacting DNN nodes in image classification problems, by appropriately leveraging the collective knowledge of all node peers. The LENC framework achieves strong performance in on-line unlabelled CKD.

Federated Learning (FL) emerges as a distributed machine learning approach that addresses privacy concerns by training AI models locally on devices. Decentralized Federated Learning (DFL) extends the FL paradigm by eliminating the central server, thereby enhancing scalability and robustness through the avoidance of a single point of failure. However, DFL faces significant challenges in optimizing security, as most Byzantine-robust algorithms proposed in the literature are designed for centralized scenarios. In this paper, we present a novel Byzantine-robust aggregation algorithm to enhance the security of Decentralized Federated Learning environments, coined WFAgg. This proposal handles adverse conditions and strengthens the robustness of dynamic decentralized topologies at the same time by employing multiple filters to identify and mitigate Byzantine attacks. Experimental results demonstrate the effectiveness of the proposed algorithm in maintaining model accuracy and convergence in the presence of various Byzantine attack scenarios, outperforming state-of-the-art centralized Byzantine-robust aggregation schemes (such as Multi-Krum or Clustering). These algorithms are evaluated on an IID image classification problem in both centralized and decentralized scenarios.

We study the application of federated learning (FL) to indoor fire detection. Such fire-detection systems use edge cameras that record sensitive footage which cannot easily be collected at a central server. Existing federated solutions leave three practical obstacles unaddressed: limited uplink bandwidth, Byzantine (malicious or faulty) clients, and unconditional trust in a single, permanently fixed aggregation server. Our main contributions address all three. In particular, we provide (i) a curated indoor fire-detection dataset assembled from eight public sources; (ii) an edge-deployable detector whose model updates are compressed up to 10 times with only a small loss in balanced accuracy; and (iii) a semi-decentralized Byzantine-robust FL method that combines history-aware aggregation with a rotating coordinator, evicting stealthy attacks that per-round filters miss while removing the fixed-server single point of failure. On the held-out test set the rotating-coordinator method matches its fixed-server counterpart in accuracy and detection speed, and a physically distributed six-node cloud deployment confirms feasibility.