A collaboration between dataset owners and model owners is needed to facilitate effective machine learning (ML) training. During this collaboration, however, dataset owners and model owners want to protect the confidentiality of their respective assets (i.e., datasets, models and training code), with the dataset owners also caring about the privacy of individual users whose data is in their datasets. Existing solutions either provide limited confidentiality for models and training code, or suffer from privacy issues due to collusion. We present Citadel++, a scalable collaborative ML training system designed to simultaneously protect the confidentiality of datasets, models and training code, as well as the privacy of individual users. Citadel++ enhances differential privacy techniques to safeguard the privacy of individual user data while maintaining model utility. By employing Virtual Machine-level Trusted Execution Environments (TEEs) and improved integrity protection techniques through various OS-level mechanisms, Citadel++ effectively preserves the confidentiality of datasets, models and training code, and enforces our privacy mechanisms even when the models and training code have been maliciously designed. Our experiments show that Citadel++ provides privacy, model utility and performance while adhering to confidentiality and privacy requirements of dataset owners and model owners, outperforming the state-of-the-art privacy-preserving training systems by up to 543x on CPU and 113x on GPU TEEs.

The effectiveness of Large Language Models (LLMs) in solving tasks vastly depends on the quality of the instructions, which often require fine-tuning through extensive human effort. This highlights the need for automated instruction optimization; however, this optimization is particularly challenging when dealing with black-box LLMs, where model parameters and gradients remain inaccessible. We propose ACING, a task-specific prompt optimization approach framed as a stateless continuous-action Reinforcement Learning (RL) problem, known as the continuum bandit setting. ACING leverages an actor-critic-based method to optimize prompts, learning from non-differentiable reward signals. We validate ACING by optimizing prompts for ChatGPT on 30 instruction-based tasks. ACING consistently outperforms baseline methods, achieving a median score improvement of 10 percentage points. Furthermore, ACING not only recovers but also surpasses human-crafted expert instructions, achieving up to a 39 percentage point improvement against human benchmarks.

This work tackles the challenges of data heterogeneity and communication limitations in decentralized federated learning. We focus on creating a collaboration graph that guides each client in selecting suitable collaborators for training personalized models that leverage their local data effectively. Our approach addresses these issues through a novel, communication-efficient strategy that enhances resource efficiency. Unlike traditional methods, our formulation identifies collaborators at a granular level by considering combinatorial relations of clients, enhancing personalization while minimizing communication overhead. We achieve this through a bi-level optimization framework that employs a constrained greedy algorithm, resulting in a resource-efficient collaboration graph for personalized learning. Extensive evaluation against various baselines across diverse datasets demonstrates the superiority of our method, named DPFL. DPFL consistently outperforms other approaches, showcasing its effectiveness in handling real-world data heterogeneity, minimizing communication overhead, enhancing resource efficiency, and building personalized models in decentralized federated learning scenarios.

We propose NeuronaBox, a flexible, user-friendly, and high-fidelity approach to emulate DNN training workloads. We argue that to accurately observe performance, it is possible to execute the training workload on a subset of real nodes and emulate the networked execution environment along with the collective communication operations. Initial results from a proof-of-concept implementation show that NeuronaBox replicates the behavior of actual systems with high accuracy, with an error margin of less than 1% between the emulated measurements and the real system.

This research investigates the enhancement of knowledge distillation (KD) processes in pre-trained models, an emerging field in knowledge transfer with significant implications for distributed training and federated learning environments. These environments benefit from reduced communication demands and accommodate various model architectures. Despite the adoption of numerous KD approaches for transferring knowledge among pre-trained models, a comprehensive understanding of KD's application in these scenarios is lacking. Our study conducts an extensive comparison of multiple KD techniques, including standard KD, tuned KD (via optimized temperature and weight parameters), deep mutual learning, and data partitioning KD. We assess these methods across various data distribution strategies to identify the most effective contexts for each. Through detailed examination of hyperparameter tuning, informed by extensive grid search evaluations, we pinpoint when adjustments are crucial to enhance model performance. This paper sheds light on optimal hyperparameter settings for distinct data partitioning scenarios and investigates KD's role in improving federated learning by minimizing communication rounds and expediting the training process. By filling a notable void in current research, our findings serve as a practical framework for leveraging KD in pre-trained models within collaborative and federated learning frameworks.

In Federated Learning (FL), forgetting, or the loss of knowledge across rounds, hampers algorithm convergence, particularly in the presence of severe data heterogeneity among clients. This study explores the nuances of this issue, emphasizing the critical role of forgetting in FL's inefficient learning within heterogeneous data contexts. Knowledge loss occurs in both client-local updates and server-side aggregation steps; addressing one without the other fails to mitigate forgetting. We introduce a metric to measure forgetting granularly, ensuring distinct recognition amid new knowledge acquisition. Leveraging these insights, we propose Flashback, an FL algorithm with a dynamic distillation approach that is used to regularize the local models, and effectively aggregate their knowledge. Across different benchmarks, Flashback outperforms other methods, mitigates forgetting, and achieves faster round-to-target-accuracy, by converging in 6 to 16 rounds.

In distributed training, communication often emerges as a bottleneck. In response, we introduce Kimad, a solution that offers adaptive gradient compression. By consistently monitoring bandwidth, Kimad refines compression ratios to match specific neural network layer requirements. Our exhaustive tests and proofs confirm Kimad's outstanding performance, establishing it as a benchmark in adaptive compression for distributed deep learning.

Federated learning is an emerging machine learning paradigm that enables clients to train collaboratively without exchanging local data. The clients participating in the training process have a crucial impact on the convergence rate, learning efficiency, and model generalization. In this work, we propose FilFL, a new approach to optimizing client participation and training by introducing client filtering. FilFL periodically filters the available clients to identify a subset that maximizes a combinatorial objective function using an efficient greedy filtering algorithm. From this filtered-in subset, clients are then selected for the training process. We provide a thorough analysis of FilFL convergence in a heterogeneous setting and evaluate its performance across diverse vision and language tasks and realistic federated scenarios with time-varying client availability. Our empirical results demonstrate several benefits of our approach, including improved learning efficiency, faster convergence, and up to 10 percentage points higher test accuracy compared to scenarios where client filtering is not utilized.

Efficient distributed training is a principal driver of recent advances in deep learning. However, communication often proves costly and becomes the primary bottleneck in these systems. As a result, there is a demand for the design of efficient communication mechanisms that can empirically boost throughput while providing theoretical guarantees. In this work, we introduce Global-QSGD, a novel family of quantization operators, engineered to accelerate distributed training based on global scaling. We demonstrate that Global-QSGD is the first theoretically rigorous Allreduce-compatible compression mechanism that achieves a provable speed-up by striking a balance between compression error and communication savings. Importantly, Global-QSGD does not rely on costly error feedback due to its inherent unbiasedness and offers up to $O(\sqrt{n})$ additional compression ratio compared to the popular QSGD quantization ($n$ represents the number of workers). To obtain theoretical guarantees, we generalize the notion of standard unbiased compression operators to incorporate Global-QSGD. We show that this wider class permits standard analysis for unbiased compressors and thus ensures convergence for popular optimization algorithms (e.g., distributed SGD) under typical settings. For the empirical component of our work, we carry out a performance modeling analysis to determine if Global-QSGD can enhance training throughput under specific hardware configurations. We also conduct extensive empirical evaluations on various tasks, testing our theory on both NVLink and PCIe connections as well as a large-scale cloud system.

The learning rate (LR) schedule is one of the most important hyper-parameters needing careful tuning in training DNNs. However, it is also one of the least automated parts of machine learning systems and usually costs significant manual effort and computing. Though there are pre-defined LR schedules and optimizers with adaptive LR, they introduce new hyperparameters that need to be tuned separately for different tasks/datasets. In this paper, we consider the question: Can we automatically tune the LR over the course of training without human involvement? We propose an efficient method, AutoLRS, which automatically optimizes the LR for each training stage by modeling training dynamics. AutoLRS aims to find an LR applied to every τ steps that minimizes the resulted validation loss. We solve this black-box optimization on the fly by Bayesian optimization (BO). However, collecting training instances for BO requires a system to evaluate each LR queried by BO's acquisition function for τ steps, which is prohibitively expensive in practice. This mutual-training process between BO and the loss-prediction model allows us to limit the training steps invested in the BO search. We demonstrate the advantages and the generality of AutoLRS through extensive experiments of training DNNs for tasks from diverse domains using different optimizers. The LR schedules auto-generated by AutoLRS lead to a speedup of 1.22, 1.43, and 1.5 when training ResNet-50, Transformer, and BERT, respectively, compared to the LR schedules in their original papers, and an average speedup of 1.31 over state-of-the-art heavily-tuned LR schedules. In the regime of deep learning, the success of training largely depends on the choice of the learning rate (LR) schedule, since most optimizers will have difficulty traversing a non-smooth and non-convex loss landscape with multiple local minimums and possibly saddle points (Kawaguchi, 2016; Jin et al., 2017; Goodfellow et al., 2016; Li et al., 2018a). To achieve stable and fast convergence towards a solution with good generalization performance, one has to tune the LR schedules carefully for different tasks (Nar & Sastry, 2018; Jastrzębski et al., 2017).