Distributed Training

As model sizes continue to scale, distributed training has become inevitable. Automatic parallelization techniques can derive efficient training parallelism strategies at low cost while achieving superior performance. The difficulty of this problem is jointly determined by the complexity of the model and the underlying compute cluster. Meanwhile, mixture-of-experts (MoE) models are increasingly emerging as the dominant architecture and the rapid evolution of accelerator hardware has made cluster heterogeneity commonplace, posing substantial challenges to automatic parallelization. However, existing approaches typically target either MoE architectures or heterogeneous clusters, failing to generalize to scenarios where both challenges coexist. To this end, we present HAPMoE, a heterogeneity-aware automatic parallelism planner for MoE training. HAPMoE builds a lightweight MoE-aware cost model and efficiently searches a six-dimensional parallel space, producing parallel plans directly deployable on Megatron-LM. Experiments show that HAPMoE improves end-to-end training throughput by up to 3.2$\times$ over baselines across heterogeneous clusters. Its non-uniform pipeline partitioning yields an additional up to 78% gains, and its pruning-enhanced dynamic programming algorithm completes the search within 1 minute, demonstrating high efficiency and practical value in complex hardware environments.

Elastic Compute Cloud (EC2) Spot is 60% to 90% cheaper than On-Demand but can be reclaimed on just a 2-minute notice; for expensive multi-node training this loss can be severe, with one reclaim costing hours of synchronous progress. We build Argus, a Kubernetes operator, and ask empirically, on a CIFAR-10 testbed, when predicting interruptions beats simple checkpointing. Argus on real EKS survives a real Spot drain with a graceful SIGTERM checkpoint, resuming from epoch 8 and losing only the in-progress epoch. Alongside, we further find that in an 80-trial benchmark, the reactive-on-notice degrades toward no protection once interruption outpaces the fixed 2-minute notice, and predictive wasted compute is driven to zero, but with an oversized fixed lead it over-migrates so severely that at the fastest rate only one of five runs completes, while periodic is a strong ML-free baseline. A lead-time sweep turns the lead prediction into a guideline where a small lead suffices for zero waste, but excess lead is wasteful. The predictor built is advisory (a proxy label); real interruption labels and large-model-scale validation are future work.

The pre-training of Large Language Models (LLMs) is increasingly conducted across multiple data centers. As training scales to a larger number of accelerators, the fraction of time spent on computation decreases, while the fraction spent on communication increases. Therefore, frequent synchronization becomes a growing bottleneck. Local update methods reduce this cost by allowing workers to perform several optimizer steps between synchronizations. Most local update methods set the number of local optimizer steps between synchronizations before training and keep this interval fixed throughout the run. However, the best interval can change during the entire train process. If the interval and optimizer are adapted to the current training state, the communication frequency is reduced while maintaining the training performance. In this work, we introduce AutoLoCo, an adaptive training framework to reduce communication in LLM training. It adapts the local interval using scalar training statistics and corrects each outer update. Our method is motivated by two observations: 1) the appropriate local interval varies across training stages, and 2) changing the number of inner steps per interval creates a mismatch with an unchanged outer optimizer, requiring a correction to the outer update. We optimize this mismatch by correction of the outer optimizer for the momentum and the learning rate using the accumulated inner learning rate. Our experiments under communication constraints demonstrate that AutoLoCo reduces communication frequency by 27% relative to DiLoCo while maintaining training performance.

Dynamic routing creates severe load imbalance in large-scale expert-parallel Mixture-of-Experts (MoE) training, turning GPUs that host hot experts into stragglers. As each MoE layer waits for its slowest rank, these stragglers prolong the expert-parallel stage and reduce overall training efficiency. Existing expert-parallelism load-balancing (EPLB) systems commonly compute load-balancing plans on the CPU, incurring device--host data transfers and cross-rank synchronization that make scheduling at every layer and microbatch expensive. Their planning formulations also overlook the hierarchical communication costs of modern scale-up and scale-out GPU clusters. We present \textit{TopoEP}, a GPU-native, topology-aware load-balancing system for large-scale MoE training. At each MoE layer and training microbatch, \textit{TopoEP} converts the current routing result into hot-expert replication and token-rerouting decisions and executes the resulting plan without data-dependent host synchronization, reducing critical-path overhead. To generate these decisions, \textit{TopoEP} uses a deterministic GPU solver that performs inter-node placement followed by intra-node refinement, allowing all ranks to independently produce bitwise-identical plans. On a 32-GPU NVIDIA H800 cluster, integrating \textit{TopoEP} with Megatron-LM improves end-to-end training throughput by 6.2%--11.4% across three representative MoE models.

Fairness-aware distributed learning prioritizes clients with large losses to reduce performance disparities, but label poisoning can create large losses, thereby inducing a fairness--robustness conflict. We propose FairMean to manage this conflict. FairMean weights client gradients using a bounded, nondecreasing function of local loss. The increasing weights prioritize high-loss clients to promote fairness, while the upper bound prevents excessive loss-induced amplification of poisoned-client gradients. In the absence of label poisoning, we show that minimizing the FairMean objective is more conducive to solution fairness than minimizing the standard average-loss objective. Under label poisoning, we establish an average-stationarity bound whose attack-dependent term is proportional to the square of the poisoned-client fraction. Experiments show that FairMean promotes fairness by reducing accuracy variance while improving worst-client accuracy.

The symbiotic scaling of artificial intelligence models and high-performance computing systems continually creates algorithmic challenges in their convergence. Foundation models (FMs) are a crucial example, requiring months-long training on thousands of cutting-edge GPUs. Sharded data parallelism (DP) is the dominant strategy to accelerate such computations by splitting data and models across multiple GPUs. However, it incurs prohibitive communication overhead when deployed at scale, particularly on multi-tier interconnects with heterogeneous performance. Inspired by the efficient communication principles of federated learning (FL), this work introduces two hybrid algorithms - FL+FSDP and FL+HSDP - interleaving sharded DP with FedAvg-style aggregations. Such approaches decouple large DP deployments into smaller, loosely-coupled federation groups, requiring minimal inter-group traffic while keeping the global batch size bounded by the groups' size. Formal analysis of communication costs and experimental validation prove their scalability and flexibility. A Llama3.1 8B pre-training on 512 A100 GPUs shows that, under identical hyperparameters, FL+FSDP and FL+HSDP achieve up to 8.04 faster data processing and 4.48 lower evaluation perplexity than their counterparts, demonstrating superior computational efficiency and improved model quality. These properties stem from reduced communication overhead and the bounded growth of the global batch size relative to the federation group size.

Federated Learning (FL) is experiencing a substantial research interest, with many frameworks being developed to allow practitioners to build federations easily and quickly. Most of these efforts do not consider two main aspects that are key to Machine Learning (ML) software: customizability and performance. This research addresses these issues by implementing an open-source FL framework named FastFederatedLearning (FFL). FFL is implemented in C/C++, focusing on code performance, and allows the user to specify any communication graph between clients and servers involved in the federation, ensuring customizability. FFL is tested against Intel OpenFL, achieving consistent speedups over different computational platforms (x86-64, ARM-v8, RISC-V), ranging from 2.5x and 3.69x. We aim to wrap FFL with a Python interface to ease its use and implement a middleware for different communication backends to be used. We aim to build dynamic federations in which relations between clients and servers are not static, giving life to an environment where federations can be seen as long-time evolving structures and exploited as services.

Open-source language models have a reproducibility problem. Despite releasing weights, training data, and recipes, none of them are provably reproducible due to the non-associativity of floating-point arithmetic. Deep learning frameworks often offer a deterministic execution mode, allowing reproducible operations on the same machines. Unfortunately, this determinism does not carry across hardware such that a user can verify that a released checkpoint was actually produced using the declared training recipe. This leaves room for undisclosed data, injected biases, or backdoors that existing techniques such as proof-of-learning or proof-of-training-data cannot rule out. We introduce a new tier of model transparency, fully auditable, in which every operation on every data sample during training is independently reproducible on heterogeneous commodity hardware with bitwise certainty. By imposing a definite order on the sources of training nondeterminism, GPU kernel reductions, data batch ordering across a data-parallel cluster, and inter/intra-node collective communication, we make it possible to replay any individual step of a large, distributed training run on a single piece of commodity hardware and check it against the published trajectory. Because replaying an entire run on one machine is infeasible, we support this with a collective verification scheme in which many independent auditors each certify individual steps, together covering the whole run. We release Open-1B, a model trained under this regime, together with its full pretraining dataset, every intermediate checkpoint, the training codebase, and the audit harness needed to reproduce and verify any step of its training.

Rising societal and lifestyle complexity has been linked to a growing prevalence of mental distress worldwide. Educational institutions, workplaces, clinics, etc. collect large volumes of mental health survey data to understand and reduce this burden. Collaborative analysis of such data could yield effective generalizable predictive models. Privacy constraints and varied survey designs (i.e., different questions, scales, and formats) hinder direct integration. We propose a schema-aware split learning (SL) framework that preserves privacy, using a large language model (LLM) as a shared semantic encoder to harmonize heterogeneous survey schemas across institutions. We serialize each survey record into a natural-language description, unifying disparate survey schemas into a common format. The LLM is fine-tuned for mental distress assessment via Low-Rank Adaptation (LoRA) and partitioned across client and server. Clients retain the raw survey responses locally and run only a lightweight front-end, so original records never leave the institution that collected them. The resource-intensive backbone runs on the server, minimizing client-side computation. Using LLaMA-3.2-3B-Instruct, the framework attains an average ANLS of 0.708 with only 2,000 training samples, surpasses federated learning (FL) in eight of nine settings, and cuts per-client computation by three orders of magnitude, while generalizing to unseen datasets. Overall, it enables accurate, privacy-preserving, and resource-efficient collaborative learning from heterogeneous mental health survey data.

Instruction-tuning datasets for large language models (LLMs) are often large, redundant, and imbalanced, limiting efficient adaptation. Naive large-batch fine-tuning repeatedly includes overrepresented sample groups while weakly covering underrepresented but informative ones, especially under data parallelism (DP) across multiple GPUs. We propose CluSTER, a Cluster-aware balanced Sampling framework for Training Efficient data Reduction in DP instruction tuning. CluSTER curates a representative reduced dataset through gradient-space clustering and DP-aware balanced allocation, ensuring dual-level coverage across clusters and workers, while preserving the original data distribution by weighted update. As a result, CluSTER reduces redundant computation and improves training stability without compromising model quality. Across multiple instruction-tuning datasets, CluSTER reduces training time by up to 69.6% with almost no accuracy loss compared to prior sampling and data reduction methods. Code is available at https://github.com/kaist-dmlab/CluSTER.

We present BEAST, the first-ever Bayesian Swin Transformer for atmospheric forecasting on 0.25$^\circ$ global resolution able to accurately quantify both aleatoric and epistemic uncertainty. To overcome the associated computational bottlenecks, we devise an orthogonal 4D-parallelization scheme that introduces a unique domain-tensor-parallelism strategy and a novel uncertainty parallel method, enabling us to fully leverage GPU capacity and efficiently scale model training. For a 2.4-billion-parameter model, we achieve a peak performance of 3.96 EFLOP/s on 20,480 NVIDIA GH200 GPUs on the JUPITER supercomputer. We train BEAST as a 700-million-parameter model with 96 random weight samples on 384 nodes on 40 years of data for nearly one million gradient updates. This model achieves predictive skill scores competitive with state-of-the-art probabilistic atmospheric AI models and numerical models, and can predict extreme events with exceptional skill, while generating large ensembles 3 to 4 times faster than the current-best AI model. Our contribution unlocks the potential of high-fidelity uncertainty quantification in atmospheric AI models, heralding a new era for AI-based models in climate and Earth system sciences.

In this paper, we investigate the generalization performance of distributed gradient descent algorithms in a reproducing kernel Hilbert space under a robust loss function $l_σ$. By exploiting the spectral characterization of gradient descent together with the intrinsic properties of robust loss functions, we establish optimal learning rates for the distributed kernel-based robust gradient descent (DKRGD) algorithm with an appropriately chosen scale parameter $σ$. The proposed parameter choice of $σ$ simultaneously alleviates the saturation phenomenon and guarantees statistical robustness. A key technical contribution is a novel error analysis that provides substantially sharper bounds for products of operators, thereby significantly relaxing existing restrictions on the maximum number of local machines while retaining optimal learning rates. Finally, we develop a communication-efficient strategy that further improves the convergence performance of DKRGD.

Kolmogorov-Arnold Networks (KANs) replace the fixed activation functions and linear weights of Multi-Layer Perceptrons (MLPs) with learnable univariate functions on network edges, offering improved interpretability and, in some settings, competitive parameter efficiency. While the approximation properties of KANs have received considerable attention, their behavior under distributed, multi-GPU training has not been systematically characterized. This paper presents an empirical scalability study of data-parallel KAN training on multi-node, multi-GPU high-performance computing (HPC) infrastructure, evaluated along four dimensions: strong scaling, weak scaling, communication overhead, and model-size scaling. Experiments were conducted on the FinisTerrae III supercomputer using up to 8 NVIDIA A100 GPUs across 4 nodes with PyTorch Distributed Data Parallel (DDP). KAN training reaches 74.7% parallel efficiency at 8 GPUs with a 5.97x speedup, consistent with conventional deep learning workloads. Weak scaling shows an initial single-to-multi-GPU throughput drop followed by strong stability. Communication overhead follows a non-monotonic pattern (1.3%-6.1%), driven primarily by All-Reduce algorithm selection and inter-node latency rather than KAN's edge-wise gradient structure. The parameter-to-memory ratio improves with model size even as training time scales unfavorably. These results indicate that operator-level and data-parallel optimizations for KAN are complementary. We provide deployment guidelines for GPU topology and model-size selection, and discuss the limitations of a synthetic-regression evaluation.

Multimodal models are increasingly the default option for perception at the network edge, yet they are trained almost entirely in the datacenter, because a client holding several sensor streams cannot host an encoder per modality. Split Learning makes such training feasible by keeping only the first layers on the device, at the cost of an uplink that must carry smashed activations for every modality at every step. Existing compression schemes give each modality the same keep-ratio, so the shared budget is divided in proportion to smashed-activation dimension, a quantity unrelated to how much each modality contributes to the fused prediction. We make that division an explicit decision and call it inter-modality allocation: under a fixed uplink budget, every policy transmits the same expected payload and differs only in how that payload is split across modalities. Our allocator, ModalShare, sets each modality's keep-ratio from a Shapley contribution score that the server computes over coalitions of activations it has already received. Measuring this score adds no uplink traffic and no client-side computation, and needs no prior knowledge of which stream is which. ModalShare improves accuracy over equal keep-ratios by 15.4 and 12.4 percentage points on CREMA-D and MVSA at matched payload in 5x compression, with strong performance across three compressors, three datasets, and four budgets. We show that existing compressors underperform in multimodal settings, with ModalShare recovering what gains are left behind.

Satellite-based distributed learning promises to train machine-learning models directly in orbit using massive, globally dispersed sensor data, thereby avoiding large-scale data downloads to ground servers. However, training convergence is significantly slowed by severe non-IID data, specifically label imbalance, as each satellite observes different geographic regions with distinct labels. This imbalance extends training duration and increases energy consumption for solar-powered satellites. Existing approaches either fully redistribute data to enforce IID conditions - accelerating convergence but incurring substantial communication delays - or avoid redistribution entirely by modifying local learning algorithms to mitigate the impact of label imbalance, which, however, still prolong training and increase energy use. Both extremes result in excessive total end-to-end learning time (data-transfer delay plus training time) and thus elevated onboard energy consumption. We present SatDL, a data-redistribution framework designed to minimize total end-to-end learning time. At its core, SatDL develops a Distributor-Critic framework that jointly models and optimizes data-transfer delay and training time. Evaluations through trace-driven simulations of a 1,584-satellite Starlink constellation and hardware emulations using NVIDIA Jetson and A100 GPUs across five datasets show SatDL reduces total end-to-end learning time by up to 18.6% and onboard energy consumption by 12.23-88.00%, while maintaining inference accuracy within a few percentage points of state-of-the-art baselines.

In LLM pre-training, synchronization propagates rank-local stalls, slowdowns, and numerical errors into job-wide symptoms, obscuring their origin. Existing diagnosis often relies on in-process monitors that cannot report after the trainer blocks or terminates, or on post-mortem logs that preserve only synchronized symptoms; offline health tests lose the workload and operating conditions that triggered the failure. We present SCOUT, a unified runtime failure-localization framework built on one design principle: identify outliers through strict-majority consensus among equivalent replicas. SCOUT aligns replica progress, timing, and numerical evidence, then uses its Consensus Collective Communication (C3) abstraction to identify ranks whose compact signatures disagree with their peers. An out-of-band CPU observer remains responsive when training hangs, whereas in-situ replay exercises recurring stragglers and silent data corruption (SDC) beside the live job with its model state, kernels, allocations, communication path, and thermal and memory pressure present. Collective fingerprints expose rank-local protocol divergence. Clean replay coverage certifies checkpoint numerical integrity, preventing recovery from selecting state corrupted by SDC. SCOUT integrates with PyTorch, TorchTitan, Megatron-Core, and DeepSpeed without training-loop or framework-source modifications. SCOUT is open source at https://github.com/LMResiliency/lm-resiliency.

Query-Key Normalization (QK-Norm) improves the training stability and quality of modern Large Language Models (LLMs). However, under Tensor Parallelism (TP), layerwise QK-Norm introduces additional cross-GPU communication because the normalization factor depends on the full hidden vector. We present SwiftQK, a multi-GPU RMSNorm kernel that exchanges only scalar normalization statistics and overlaps the remaining Peer-to-Peer reduction with independent element-wise computation in a deadlock-safe persistent kernel. Evaluations on recent LLMs show that SwiftQK reduces QK-Norm latency by 81.4--93.9% relative to the standard TP QK-Norm using full-vector All-Gather. In end-to-end serving, SwiftQK reduces TPOT on average by 29.5% over the All-Gather-based baseline and by 14.3% over an optimized scalar-aggregation implementation.

Graph Neural Networks (GNNs) are widely used across domains such as natural sciences, social network analysis, chip design, and recommendation systems. However, as graph sizes grow, storing and processing them entirely on a single-node CPU-GPU system becomes increasingly impractical. A promising approach is to distribute the graph across multiple remote memory nodes, though this introduces a major bottleneck: inter-node network congestion during training. To address this, we propose LGNNIC, a novel inter-node system architecture that leverages SmartNICs co-located with remote memory nodes-a configuration already available in modern systems-to reduce communication overhead in distributed GNN training. LGNNIC offloads key preprocessing tasks to SmartNICs, reducing the volume of data transferred to computational (training) nodes and alleviating network congestion. We introduce two complementary techniques executed on the SmartNICs during the preprocessing phase: Neighbor Sampling, which performs mini-batch sampling, and Quantization of the sampled batches. To evaluate LGNNIC under different communication infrastructures, we designed both an optimized low-overhead DMA-based synchronization mechanism and a high-overhead socket-based alternative used as a benchmark. We evaluate the core SmartNIC offloading mechanisms across standard GNN workloads and sampling hyperparameters using a proof-of-concept (PoC) system comprising one remote-memory node with an NVIDIA BlueField-2 SmartNIC and one compute node with an A100 GPU. Both Neighbor Sampling and Quantization on the remote node demonstrated substantial training speedups in most configurations. Neighbor Sampling achieved up to 62.4x and 17.5x speedups with Sockets and DOCA-DMA, respectively, primarily due to reduced data transaction time. Quantization provided additional speedups of up to 3.6x and 1.3x, respectively, by reducing data transfer.

Compact AI systems make local language-model experimentation increasingly accessible, yet practical evidence for multi-node training on desktop-class accelerators remains limited. This report presents a proof-of-concept deployment of distributed NanoChat pretraining across two NVIDIA DGX Spark systems, each with a GB10 Grace Blackwell system-on-chip and 128 GB of unified memory, administered remotely over a Tailscale mesh VPN and connected for training by a dedicated 200 Gb/s QSFP56 direct fiber link. PyTorch torchrun, DDP, and NCCL were configured with one process per node, a depth-20 NanoChat model, a local batch size of 32 per node, and a 2,048-token context, giving a global batch of 131,072 tokens per step. The run sustained a step time of about 69.4 s (about 1,890 tokens/s), processing about 653 million tokens over four days. We document link configuration, container setup, interface binding, a step-zero evaluation bug that triggered NCCL timeouts, checkpointing, and troubleshooting lessons, as a reproducibility reference for small labs. We also built a cybersecurity fine-tuning dataset from 77 CISA advisories (338 training, 37 validation conversations) and ran a 17-question held-out evaluation comparing a baseline SFT checkpoint against a CTI-augmented checkpoint with an Ollama-hosted LLM judge. CTI-specific categories improved while general-knowledge categories regressed, for a small overall change from 2.06 to 2.29 on a 0-10 scale. The same cluster supports a 400-level AI course (CS 426) and a query engine for CompTIA Security+ POGIL activities in CBS 255, showing modest local infrastructure can serve both research and teaching. The study establishes feasibility rather than a scaling-efficiency claim, since single-node throughput used for comparison was estimated, not measured under matched conditions. Runbook and scripts are available (see Code Availability).

In gossip learning, a network of nodes trains a shared model collaboratively, without a central coordinator, by repeatedly exchanging parts of their local models. The state-of-the-art protocol, Partitioned Token Gossip Learning (PTGL) of Heged{ü}s et al., splits the weight matrix into S fixed partitions and disseminates them using a token-based fairness mechanism coupled with per-neighbor metadata exchange. We revisit partition scheduling by analogy with peer-to-peer live streaming, where model partitions act as video chunks and partition age acts as chunk scarcity. The analogy yields a design space of two-stage selection strategies (partition first, or neighbor first), from which we instantiate ten concrete protocols collectively called Stream Learning. Our main finding is that the simplest of these protocols, which transmits the locally least-trained partition to a uniformly random neighbor (Ri), matches PTGL on fault-free workloads while requiring neither token counters nor metadata exchange. Under an adversarial 30% permanent crash of the best-performing nodes, Ri matches or outperforms PTGL across all complete-graph configurations tested, with the gap reaching 5.53% on HAR and 5.41% on MNIST in the most heterogeneous regime (Dirichlet $β$ = 0.1). In our experiments, partition fairness, captured by a single local rule on partition age, accounts for the gap; token-based rate control and utility maximization do not improve over this rule and, under heterogeneity, sit below it.

AI training workloads are growing rapidly, making their time, energy, and infrastructure costs increasingly important. In shared cloud clusters, training and fine-tuning jobs compete with co-running workloads for network resources, while network mechanisms and ML training choices are typically optimized separately: networking controls how bytes move, whereas ML systems control when and how much communication occurs. We argue that this separation leaves end-to-end performance on the table. We present ML-for-ML, a cross-layer perspective in which network-side and ML-side knobs are selected jointly under a shared time-to-target-loss objective. Our preliminary prototype shows that by co-optimizing the ML and network parameters, we reach the target loss up to 42% faster.

We report operational experience full-fine-tuning a 32.76B-parameter dense model (Qwen3-32B) on 16 x NVIDIA B300 (two nodes, FSDP / ZeRO-3) -- among the first published field accounts on this accelerator. We claim no new algorithm. The individual mechanisms we use are established practice; our contribution is the integrated field experience and a set of calibrated measurements on new hardware. Concretely we offer four practitioner artifacts. (1) A B300-calibrated power-draw triage table that distinguishes compute / communication / data-starvation / checkpoint-or-deadlock / idle by board wattage (utilization% reads 100% during an NCCL hang). (2) A set of honest negative results that dispel common optimization folklore at this scale: a controlled A/B in which per-step NFS reading matches a pretokenized local cache (~53k tok/s) because the corpus fits in page cache and the job is compute-bound; and a reconstruction of an earlier "throughput collapse" as NFS/CPU contention rather than a storage-medium limit. (3) Calibrated 4/8/16-GPU strong-scaling and GPU-hour numbers on B300 (near-linear, as expected in this regime; we report absolute values as reference data). (4) A worked failure case -- an epoch-end NCCL deadlock from per-rank token-packing imbalance -- together with a 2.7-second pre-run invariant gate and an external watcher that turn multi-hour silent failures into instant rejections. This deadlock and its remedy correspond to PyTorch's documented Join / equalize-to-minimum practice; we position our instantiation against that prior art and report the GPU-hours the failure cost and the gate saves. The transferable takeaway is operational, not algorithmic: for data-dependent data-parallel jobs, watch power rather than utilization, and verify invariants before launch -- a passing smoke test is not evidence of a safe full run.

Credit risk detection, particularly mitigating individual fraud, is crucial for maintaining the stability of digital financial ecosystems. Accurately identifying credit fraud among billions of users is critical for minimizing financial losses and safeguarding the sustainability of inclusive financial services. Given that credit fraud risks are often concealed within heterogeneous user-risk graphs, Graph Neural Networks (GNNs) have emerged as an effective tool for risk mining by capturing complex dependencies. To address the scalability bottleneck of industrial GNNs, distributed training based on subgraphs is indispensable. However, existing strategies often compromise topological integrity for load balancing. This can be catastrophic for risk detection, as it indiscriminately severs the long-tail evidence chains essential for risk propagation. Overlapping subgraphs can restore severed risk contexts but inevitably introduce redundancy and noise, while overlooking the representation alignment across different local subgraphs. In this paper, we propose a risk-aware overlapping subgraph learning framework for large-scale credit risk detection. We first construct base partitions to ensure load balance. Then, we perform budget-constrained sampling that selects informative long-tail nodes, thereby preserving critical risk diffusion patterns while filtering out noise. To mitigate representation inconsistency, we design a cross-subgraph consistency alignment mechanism. By enforcing alignment constraints on the overlapping nodes, we harmonize the local representations into a globally consistent latent space. Extensive experiments on Weixin Pay's production dataset demonstrate that our model significantly outperforms existing strategies for risk detection, offering a scalable and effective solution for industrial graph learning.

In recent years, the prevalence of large-scale data-sets and the demand for sophisti-cated learning models have necessitated the development of efficient distributed ma-chine learning (ML) solutions. Convergence speed is a critical factor influencing the practicality and effectiveness of these distributed frameworks. Recently, non-Lipschitz continuous optimization algorithms have been proposed to improve the slow conver-gence rate of the existing linear solutions. The use of signum-based functions is previ-ously considered in consensus and control literature to reach fast convergence in the prescribed time and also to provide robust algorithms to noisy/outlier data. However, as shown in this work, these algorithms lead to an optimality gap and steady-state re-sidual of the objective function in discrete-time setup. This motivates us to investigate the distributed optimization and ML algorithms in terms of trade-off between conver-gence rate and optimality gap. In this direction, we specifically consider the distributed regression problem and check its convergence rate by applying both linear and non-Lipschitz signum-based functions. We check our distributed regression approach by extensive simulations. Our results show that although adopting signum-based func-tions may give faster convergence, it results in large optimality gaps. The findings pre-sented in this paper may contribute to and advance the ongoing discourse of similar distributed algorithms, e.g., for distributed constrained optimization and distributed estimation.

In large-scale machine learning, distributed training commonly involves multiple workers evaluating the gradients of the model on different dataset partitions. A common challenge is the presence of straggling workers, which may significantly slow down training. Traditional gradient coding (GC) addresses this by duplicating dataset partitions across workers, allowing for the replacement of missing gradients from stragglers. However, GC requires workers to evaluate gradients on multiple dataset partitions in each step, potentially increasing overall training time. In this paper, we propose to pipeline GC, such that gradient evaluation is segmented across multiple steps and each worker evaluates gradients on just a single dataset partition per step. We develop the pipelined version for fractional repetition (FR) and cyclic repetition (CR), two representative dataset placement schemes in GC, and prove convergence guarantees for both. Through extensive simulations and experiments on cloud infrastructure, our schemes not only significantly reduce training time but also accelerate convergence compared to GC and other baselines.

Backdoor attacks severely threaten large-scale AI models. When model owners delegate training to external compute providers within a decentralized training paradigm, adversaries can craft stealthy, low-frequency triggers to inject malicious behavior while evading standard audits. Traditionally, detecting these attacks requires a full re-computation of the training steps--a prohibitive overhead that directly contradicts the owner's resource constraints. To address this, we investigate the resilience of continuous optimization dynamics under Byzantine perturbations, where adversaries are forced to compete against a continuous influx of honest updates. Under a threat model where an adversary compromises f out of n total trainers, we quantify the minimum auditing overhead required by the model owner to probabilistically bound the attack success rate. We formalize this injection-absorption dynamic as a Discrete-Time Markov Chain (DTMC). Using this framework, we prove that the success probability of any bounded adversary asymptotically collapses to zero under a defense strategy combining natural absorption, a randomized scheduler, and lazy verification oracle. Empirical results demonstrate significant backdoor suppression with zero utility degradation even when invoking the verification oracle on merely 10% of the total training steps. This approach yields a provably sound and computationally efficient defense for safety-critical AI.

State-of-the-art large language model (LLM) training takes tens of thousands of graphics processing units (GPUs) for months and encounters failures across the software and hardware stack. Existing fault-tolerance mechanisms either impose non-trivial overhead during failure-free execution or suffer from prolonged recovery latency, particularly under scenarios where a small subset of compute nodes experience permanent failures. %The tradeoff between failure-free overhead and recovery latency forms a space forms a Pareto frontier We present PHOENIX to simultaneously address both optimization objectives. PHOENIX incorporates a fault-tolerance mechanism that restores LLM training via hot-swapping, namely by replacing failed nodes with spare nodes without terminating the complete job. The hot-swapping of PHOENIX is enabled by two ideas: First, it exploits an off-critical-path in-memory checkpointing mechanism for spatial redundancy. Second, it introduces a communicator reconstruction protocol that replaces failed nodes with spare nodes at runtime. PHOENIX efficiently overlaps the in-memory checkpointing with computation, thus introducing zero overhead during error-free execution. Upon permanent node failures, PHOENIX can rebuild memory states with minimal recomputation by leveraging in-memory checkpoints. We evaluate PHOENIX across scales (up to 512 NVIDIA A100 GPUs) and LLMs (up to 65B parameters), and observe zero checkpoint overhead with hot-swapping recovery completing in under 40 seconds. These results show that PHOENIX simultaneously achieves both zero-overhead error-free execution and extremely low recovery cost.

Recent work has established a fundamental trilemma between Byzantine robustness, local differential privacy (LDP), and optimization error in distributed learning. We show that this trilemma does not universally extend to generalization error, but instead depends critically on the privacy regime. Specifically, in the high-noise regime (strong privacy), we prove that increasing privacy reduces the generalization error, i.e., there is no tension between robustness and privacy. In the low-noise regime (weaker privacy), however, the tension between robustness and privacy reappears and increasing privacy indeed degrades generalization. Our theory explains this surprising non-monotonic behavior of the generalization error via matching lower and upper bounds on the algorithmic stability of Byzantine-robust distributed learning under LDP constraints. We corroborate and further analyze these theoretical findings with empirical evaluations.

Machine-learned (ML) operator models can be trained to predict density functional theory (DFT) Hamiltonian/density matrices at significantly reduced computational cost, thus extending electronic-structure calculations to previously unfeasible scales. Here, we introduce MALOQ (Massively Accelerated Learning of Operators for Quantum Transport), an application built to train on and predict electronic-structure matrices for systems made of few to 100k atoms, described by large basis sets, and covering a wide range of atomic elements. Based on a state-of-the-art, SO(2)-equivariant backbone architecture, MALOQ provides (i) custom data-processing kernels to handle high-rank Hamiltonian matrix data and (ii) a scalable edge-wise distribution of atomic graph(s). Trained on the largest molecular Hamiltonian datasets available today, it reduces time-per-epoch by over 30% compared to a molecule-wise-distributed framework, and enables inference on material graphs of arbitrary size. We demonstrate scalable training and inference for 3,000-12,000 atoms on the Alps supercomputer, up to 192 GPUs and 256 GPUs, respectively.

Knowledge Distillation (KD) enables training smaller student models under the guidance of larger teacher models, and the widely adopted TRL library implements it. Yet, TRL treats both models symmetrically, missing opportunities to exploit their pronounced asymmetry in memory footprint, and communication requirements. This paper presents an HPC-aware methodology for KD that decouples teacher and student partitioning efficiently. Our approach achieves up to 67% higher samples-per-second than TRL by avoiding unnecessary teacher-model data structures and selecting the best split strategy. We combine vertical and horizontal partitioning of models, deriving an analytical expression that identifies the existence of inflection points between splitting regimes. These results showed that exploiting teacher--student asymmetry through topology-aware parallelism notably accelerated GKD training on production HPC clusters at our company