Large Language Model (LLM) training is increasingly concentrated in homogeneous datacenters, while private data and underutilized GPUs across universities, laboratories, and edge sites remain difficult to use. This extended abstract presents preliminary results from a geo-distributed LLM training prototype that treats networking constraints as first-order design concerns. The prototype connects three heterogeneous GPU sites via cloud-hosted parameter servers, outbound-only gRPC streams, two-stage delta compression (INT8 quantization + Huffman coding, achieving up to 4× payload reduction), and fault-tolerant rejoin. In real deployments, GPT-2 Medium pretraining achieves stable loss reduction and reaches the target loss 15.2% faster in wall-clock time than the best tested baseline; Llama3-1B pretraining remains stable under larger communication pressure; and cross-site latency traces reveal site-dependent WAN spikes of up to 200s. These results motivate adaptive networking support for synchronization, compression, placement, telemetry, and recovery in geo-distributed LLM training.

Data privacy is of great concern in cloud machine-learning service platforms, when sensitive data are exposed to service providers. While private computing environments (e.g., secure enclaves), and cryptographic approaches (e.g., homomorphic encryption) provide strong privacy protection, their computing performance still falls short compared to cloud GPUs. To achieve privacy protection with high computing performance, we propose Delta, a new private training and inference framework, with comparable model performance as non-private centralized training. Delta features two asymmetric data flows: the main information-sensitive flow and the residual flow. The main part flows into a small model while the residuals are offloaded to a large model. Specifically, Delta embeds the information-sensitive representations into a low-dimensional space while pushing the information-insensitive part into high-dimension residuals. To ensure privacy protection, the low-dimensional information-sensitive part is secured and fed to a small model in a private environment. On the other hand, the residual part is sent to fast cloud GPUs, and processed by a large model. To further enhance privacy and reduce the communication cost, Delta applies a random binary quantization technique along with a DP-based technique to the residuals before sharing them with the public platform. We theoretically show that Delta guarantees differential privacy in the public environment and greatly reduces the complexity in the private environment. We conduct empirical analyses on CIFAR-10, CIFAR-100 and ImageNet datasets and ResNet-18 and ResNet-34, showing that Delta achieves strong privacy protection, fast training, and inference without significantly compromising the model utility.

In Federated Learning (FL), clients may have weak devices that cannot train the full model or even hold it in their memory space. To implement large-scale FL applications, thus, it is crucial to develop a distributed learning method that enables the participation of such weak clients. We proposeEmbracingFL, a general FL framework that allows all available clients to join the distributed training regardless of their system resource capacity. The framework is built upon a novel form of partial model training method in which each client trains as many consecutive output-side layers as its system resources allow. Our study demonstrates thatEmbracingFLencourages each layer to have similar data representations across clients, improving FL efficiency. The proposed partial model training method guarantees convergence to a neighbor of stationary points for non-convex and smooth problems. We evaluate the efficacy ofEmbracingFLunder a variety of settings with a mixed number of strong, moderate ($\sim 40\%$memory), and weak ($\sim 15\%$memory) clients, datasets (CIFAR-10, FEMNIST, and IMDB), and models (ResNet20, CNN, and LSTM). Our empirical study shows thatEmbracingFLconsistently achieves high accuracy as like all clients are strong, outperforming the state-of-the-art width reduction methods (i.e. HeteroFL and FjORD).

Decentralized (i.e., serverless) training across edge nodes can suffer substantially from potential Byzantine nodes that can degrade the training performance. However, detection and mitigation of Byzantine behaviors in a decentralized learning setting is a daunting task, especially when the data distribution at the users is heterogeneous. As our main contribution, we proposeBasil, a fast and computationally efficient Byzantine-robust algorithm for decentralized training systems, which leverages a novel sequential, memory-assisted and performance-based criteria for training over a logical ring while filtering the Byzantine users. In the IID dataset setting, we provide the theoretical convergence guarantees ofBasil, demonstrating its linear convergence rate. Furthermore, for the IID setting, we experimentally demonstrate thatBasilis robust to various Byzantine attacks, including the strong Hidden attack, while providing up to absolute ~16% higher test accuracy over the state-of-the-art Byzantine-resilient decentralized learning approach. Additionally, we generalizeBasilto the non-IID setting by proposing Anonymous Cyclic Data Sharing (ACDS), a technique that allows each node to anonymously share a random fraction of its local non-sensitive dataset (e.g., landmarks images) with all other nodes. Finally, to reduce the overall latency ofBasilresulting from its sequential implementation over the logical ring, we proposeBasil+that enables Byzantine-robust parallel training across groups of logical rings, and at the same time, it retains the performance gains ofBasildue to sequential training within each group. Furthermore, we experimentally demonstrate the scalability gains ofBasil+through different sets of experiments.

We focus on the commonly used synchronous Gradient Descent paradigm for large-scale distributed learning, for which there has been a growing interest to develop efficient and robust gradient aggregation strategies that overcome two key system bottlenecks: communication bandwidth and stragglers’ delays. In particular, Ring-AllReduce (RAR) design has been proposed to avoid bandwidth bottleneck at any particular node by allowing each worker to only communicate with its neighbors that are arranged in a logical ring. On the other hand, Gradient Coding (GC) has been recently proposed to mitigate stragglers in a master-worker topology by allowing carefully designed redundant allocation of the data set to the workers. We propose a joint communication topology design and data set allocation strategy, named CodedReduce (CR), that combines the best of bothRARandGC. That is, it parallelizes the communications over a tree topology leading to efficient bandwidth utilization, and carefully designs a redundant data set allocation and coding strategy at the nodes to make the proposed gradient aggregation scheme robust to stragglers. In particular, we quantify the communication parallelization gain and resiliency of the proposedCRscheme, and prove its optimality when the communication topology is a regular tree. Moreover, we characterize the expected run-time ofCRand show order-wise speedups compared to the benchmark schemes. Finally, we empirically evaluate the performance of our proposedCRdesign over Amazon EC2 and demonstrate that it achieves speedups of up to$27.2\times $and$7.0\times $, respectively over the benchmarksGCandRAR.

Federated learning enables training a global model from data located at the client nodes, without data sharing and moving client data to a centralized server. Performance of federated learning in a multi-access edge computing (MEC) network suffers from slow convergence due to heterogeneity and stochastic fluctuations in compute power and communication link qualities across clients. We propose a novel coded computing framework, CodedFedL, that injects structured coding redundancy into federated learning for mitigating stragglers and speeding up the training procedure. CodedFedL enables coded computing for non-linear federated learning by efficiently exploiting distributed kernel embedding via random Fourier features that transforms the training task into computationally favourable distributed linear regression. Furthermore, clients generate local parity datasets by coding over their local datasets, while the server combines them to obtain the global parity dataset. Gradient from the global parity dataset compensates for straggling gradients during training, and thereby speeds up convergence. For minimizing the epoch deadline time at the MEC server, we provide a tractable approach for finding the amount of coding redundancy and the number of local data points that a client processes during training, by exploiting the statistical properties of compute as well as communication delays. We also characterize the leakage in data privacy when clients share their local parity datasets with the server. Additionally, we analyze the convergence rate and iteration complexity of CodedFedL under simplifying assumptions, by treating CodedFedL as a stochastic gradient descent algorithm. Finally, for demonstrating gains that CodedFedL can achieve in practice, we conduct numerical experiments using practical network parameters and benchmark datasets, in which CodedFedL speeds up the overall training time by up to 15× in comparison to the benchmark schemes.

Client devices at the edge are generating increasingly large amounts of rich data suitable for learning powerful statistical models. However, privacy concerns and heavy communication load make it infeasible to move the client data to a centralized location for training. In many distributed learning setups, client nodes carry out gradient computations on their local data while the central master server receives the local gradients and aggregates them to take the global model update step. To guarantee robustness against straggling communication links, we consider a hierarchical setup with neclients and nhreliable helper nodes that are available to aid in gradient aggregation at the master. To achieve resiliency against straggling client-to-helpers links, we propose two approaches leveraging coded redundancy. First is the Aligned Repetition Coding (ARC) that repeats gradient components on the helper links, allowing significant partial aggregations at the helpers, resulting in a helpers-to-master communication load (CHM) of O(nh). ARC however results in a client-to-helpers communication load (CEH) of Θ(nh), which is prohibitive for client nodes due to limited and costly bandwidth. We thus propose Aligned Minimum Distance Separable Coding (AMC) that achieves optimal CEHof Θ(1) for a given resiliency threshold by using MDS code over the gradient components, while achieving a CHMof O(ne).

Many distributed computing systems have been developed recently for implementing graph based algorithms such as PageRank over large-scale graph-structured datasets such as social networks. Performance of these systems significantly suffers from communication bottleneck as a large number of messages are exchanged among servers at each step of the computation. Motivated by graph based MapReduce, we propose a coded computing framework that leverages computation redundancy to alleviate the communication bottleneck in distributed graph processing. As a key contribution of this work, we develop a novel coding scheme that systematically injects structured redundancy in the computation phase to enable coded multicasting opportunities during message exchange between servers, reducing the communication load substantially in large-scale graph processing. For theoretical analysis, we consider random graph models, and focus on schemes in which subgraph allocation and Reduce allocation are only dependent on vertex ID while the Shuffle design varies with graph connectivity. Specifically, we prove that our proposed scheme enables an (asymptotically) inverse-linear trade-off between computation load and average communication load for two popular random graph models - Erdös-Rényi model, and power law model. Particularly, for a given computation load r, (i.e. when each graph vertex is carefully stored at r servers), the proposed scheme slashes the average communication load by (nearly) a multiplicative factor of r. Furthermore, for the Erdös-Rényi model, we prove that our proposed scheme is optimal asymptotically as the graph size increases by providing an information-theoretic converse. To illustrate the benefits of our scheme in practice, we implement PageRank over Amazon EC2, using artificial as well as real-world datasets, demonstrating gains of up to 50.8% in comparison to the conventional PageRank implementation. Additionally, we specialize our coded scheme and extend our theoretical results to two other random graph models - random bi-partite model, and stochastic block model. Our specialized schemes asymptotically enable inverse-linear trade-offs between computation and communication loads in distributed graph processing for these popular random graph models as well. We complement the achievability results with converse bounds for both of these models.

Scaling up distributed machine learning systems face two major bottlenecks - delays due to stragglers and limited communication bandwidth. Recently, a number of coding theoretic strategies have been proposed for mitigating these bottlenecks. In particular, the Gradient Coding (GC) scheme was proposed to speed up distributed gradient descent algorithm in a synchronous master-worker setting by providing robustness to stragglers. A major drawback of the master-worker architecture for distributed learning is however, the bandwidth contention at the master, which can significantly deteriorate the performance as the cluster size increases. In this paper, we propose a new framework named Tree Gradient Coding (TGC) for distributed gradient aggregation, which parallelizes communication over a tree topology while providing straggler robustness. As our main contribution, we characterize the minimum computation load for TGC for a given tree topology and straggler resiliency, and design a tree gradient coding algorithm that achieves this optimal computation load. Furthermore, we provide results from experiments over Amazon EC2, where TGC speeds up the training time by up to 18.8× in comparison to GC.

In wireless mobile edge computing platforms, such as those supported by vehicular networks, location specific machine learning models can be trained by distributing computations using resources available at the edge. The computational and communication resources at the wireless edge are heterogeneous and unreliable, which can lead to straggler effects that significantly slow down recursive learning tasks, such as gradient descent. In this paper we propose a coded computation framework, which utilizes statistical knowledge of resource heterogeneity to determine optimal encoding and load balancing of training data using Random Linear codes, while avoiding an explicit step for decoding gradients. Results show that the proposed coding framework speeds up the training time for linear regression models, by up to ten times, when compared with repetition coding schemes.

Many distributed graph computing systems have been developed recently for efficient processing of massive graphs. These systems require many messages to be exchanged among computing machines at each step of the computation, making communication bandwidth a major performance bottleneck. We present a coded computing framework that systematically injects redundancy in the computation phase to enable coding opportunities in the communication phase thus reducing the communication load substantially. Specifically, we propose coded schemes that enable an inverse-linear trade-off (asymptotically) between computation load and average communication load for Erdös-Rényi (ER) random graph. The proposed scheme for ER graph is shown to be optimal asymptotically as the graph size n → ∞. For finite n, we demonstrate via numerical analysis that for a given computation load r, i.e. when each graph vertex is carefully stored at r servers, the proposed scheme slashes the average communication load by (nearly) r.

In large-scale distributed computing clusters, such as Amazon EC2, there are several types of “system noise” that can result in major degradation of performance: system failures, bottlenecks due to limited communication bandwidth, latency due to straggler nodes, and so on. There have been recent results that demonstrate the impact of coding for efficient utilization of computation and storage redundancy to alleviate the effect of stragglers and communication bottlenecks in homogeneous clusters. In this paper, we focus on general heterogeneous distributed computing clusters consist of a variety of computing machines with different capabilities. We propose a coding framework for speeding up distributed computing in heterogeneous clusters by trading redundancy for reducing the latency of computation. In particular, we propose heterogeneous coded matrix multiplication (HCMM) algorithm for performing distributed matrix multiplication over heterogeneous clusters that are provably asymptotically optimal for a broad class of processing time distributions. Moreover, we show that HCMM is unboundedly faster than any uncoded scheme that partitions the total workload among the workers. To demonstrate how the proposed HCMM scheme can be applied in practice, we provide results from numerical studies and Amazon EC2 experiments comparing HCMM with three benchmark load allocation schemes-uniform uncoded, load-balanced uncoded, and uniform coded. In particular, in our numerical studies, HCMM achieves speedups of up to 73%, 56%, and 42%, respectively, over the three benchmark schemes mentioned earlier. Furthermore, we carry out experiments over Amazon EC2 clusters and demonstrate how HCMM can be combined with rateless codes with nearly linear decoding complexity. In particular, we show that HCMM combined with the Luby transform codes can significantly reduce the overall execution time. HCMM is found to be up to 61%, 46%, and 36% faster than the aforementioned three benchmark schemes, respectively. Additionally, we provide a generalization to the problem of optimal load allocation in heterogeneous settings, where we take into account the monetary costs associated with distributed computing clusters. We argue that HCMM is asymptotically optimal for budget-constrained scenarios as well. In particular, we characterize the minimum possible expected cost associated with a computation task over a given cluster of machines. Furthermore, we develop a heuristic algorithm for (HCMM) load allocation for the distributed implementation of budget-limited computation tasks.