Photon: Federated LLM Pre-Training

Scaling large language models (LLMs) demands extensive data and computing resources, which are traditionally constrained to data centers by the high-bandwidth requirements of distributed training. Low-bandwidth methods like federated learning (FL) could enable collaborative training of larger models across weakly-connected GPUs if they can effectively be used for pre-training. To achieve this, we introduce Photon, the first complete system for federated end-to-end LLM training, leveraging cross-silo FL for global-scale training with minimal communication overheads. Using Photon, we train the first federated family of decoder-only LLMs from scratch. We show that: (1) Photon can train model sizes up to 7B in a federated fashion while reaching an even better perplexity than centralized pre-training; (2) Photon model training time decreases with available compute, achieving a similar compute-time trade-off to centralized; and (3) Photon outperforms the wall-time of baseline distributed training methods by 35% via communicating 64x-512xless. Our proposal is robust to data heterogeneity and converges twice as fast as previous methods like DiLoCo. This surprising data efficiency stems from a unique approach combining small client batch sizes with extremely high learning rates, enabled by federated averaging's robustness to hyperparameters. Photon thus represents the first economical system for global internet-wide LLM pre-training.

DEPT: Decoupled Embeddings for Pre-training Language Models

Language Model pre-training benefits from a broader data mixture to enhance performance across domains and languages. However, training on such heterogeneous text corpora is complex, requiring extensive and cost-intensive efforts. Since these data sources vary in lexical, syntactic, and semantic aspects, they cause negative interference or the "curse of multilinguality". We propose a novel pre-training framework to alleviate this curse. Our method, DEPT, decouples the embedding layers from the transformer body while simultaneously training the latter in multiple contexts. DEPT enables the model to train without being bound to a shared global vocabulary. DEPT: (1) can train robustly and effectively under significant data heterogeneity, (2) reduces the parameter count of the token embeddings by up to 80% and the communication costs by 675x for billion-scale models (3) enhances model generalization and plasticity in adapting to new languages and domains, and (4) allows training with custom optimized vocabulary per data source. We prove DEPT's potential by performing the first vocabulary-agnostic federated multilingual pre-training of a 1.3 billion-parameter model across high and low-resource languages, reducing its parameter count by 409 million.

PISTOL: Dataset Compilation Pipeline for Structural Unlearning of LLMs

Recently, machine unlearning, which seeks to erase specific data stored in the pre-trained or fine-tuned models, has emerged as a crucial protective measure for LLMs. However, unlearning approaches for LLMs that have been considered thus far have focused on the removal of independent data points and have not taken into account that the stored facts are logically connected to one another and form an implicit knowledge graph. To facilitate the development of structural unlearning methods, which are essential for the practical application of unlearning, we propose PISTOL, a pipeline for compiling multi-scenario datasets for benchmarking structural LLM unlearning. Additionally, leveraging sample datasets synthesized using PISTOL, we conducted benchmarks with four distinct unlearning methods on both Llama2-7B and Mistral-7B models. This analysis helps to illustrate the prevailing challenges in effectively and robustly removing highly inter-connected data, batched data, or data skewed towards a specific domain. It also highlights the choice of pre-trained model can impact unlearning performance. This work not only advances our understandings on the limitation of current LLMs unlearning methods and proposes future research directions, but also provides a replicable framework for ongoing exploration and validation in the field.

FLea: Addressing Data Scarcity and Label Skew in Federated Learning via Privacy-preserving Feature Augmentation

Federated Learning (FL) enables model development by leveraging data distributed across numerous edge devices without transferring local data to a central server. However, existing FL methods still face challenges when dealing with scarce and label-skewed data across devices, resulting in local model overfitting and drift, consequently hindering the performance of the global model. In response to these challenges, we propose a pioneering framework called FLea, incorporating the following key components: i) A global feature buffer that stores activation-target pairs shared from multiple clients to support local training. This design mitigates local model drift caused by the absence of certain classes; ii) A feature augmentation approach based on local and global activation mix-ups for local training. This strategy enlarges the training samples, thereby reducing the risk of local overfitting; iii) An obfuscation method to minimize the correlation between intermediate activations and the source data, enhancing the privacy of shared features. To verify the superiority of FLea, we conduct extensive experiments using a wide range of data modalities, simulating different levels of local data scarcity and label skew. The results demonstrate that FLea consistently outperforms state-of-the-art FL counterparts (among 13 of the experimented 18 settings, the improvement is over 5% while concurrently mitigating the privacy vulnerabilities associated with shared features. Code is available at https://github.com/XTxiatong/FLea.git.

Sheaf HyperNetworks for Personalized Federated Learning

Graph hypernetworks (GHNs), constructed by combining graph neural networks (GNNs) with hypernetworks (HNs), leverage relational data across various domains such as neural architecture search, molecular property prediction and federated learning. Despite GNNs and HNs being individually successful, we show that GHNs present problems compromising their performance, such as over-smoothing and heterophily. Moreover, we cannot apply GHNs directly to personalized federated learning (PFL) scenarios, where a priori client relation graph may be absent, private, or inaccessible. To mitigate these limitations in the context of PFL, we propose a novel class of HNs, sheaf hypernetworks (SHNs), which combine cellular sheaf theory with HNs to improve parameter sharing for PFL. We thoroughly evaluate SHNs across diverse PFL tasks, including multi-class classification, traffic and weather forecasting. Additionally, we provide a methodology for constructing client relation graphs in scenarios where such graphs are unavailable. We show that SHNs consistently outperform existing PFL solutions in complex non-IID scenarios. While the baselines' performance fluctuates depending on the task, SHNs show improvements of up to 2.7% in accuracy and 5.3% in lower mean squared error over the best-performing baseline.

The Future of Large Language Model Pre-training is Federated

Generative pre-trained large language models (LLMs) have demonstrated impressive performance over a wide range of tasks, thanks to the unprecedented amount of data they have been trained on. As established scaling laws indicate, LLMs' future performance improvement depends on the amount of computing and data sources we can leverage for pre-training. Federated learning (FL) has the potential to unleash the majority of the planet's data and computational resources, which are underutilized by the data-center-focused training methodology of current LLM practice. Our work presents a robust, flexible, reproducible FL approach that enables large-scale collaboration across institutions to train LLMs. This would mobilize more computational and data resources while matching or potentially exceeding centralized performance. We further show the effectiveness of the federated training scales with model size and present our approach for training a billion-scale federated LLM using limited resources. This will help data-rich actors to become the protagonists of LLMs pre-training instead of leaving the stage to compute-rich actors alone.

FedAnchor: Enhancing Federated Semi-Supervised Learning with Label Contrastive Loss for Unlabeled Clients

Federated learning (FL) is a distributed learning paradigm that facilitates collaborative training of a shared global model across devices while keeping data localized. The deployment of FL in numerous real-world applications faces delays, primarily due to the prevalent reliance on supervised tasks. Generating detailed labels at edge devices, if feasible, is demanding, given resource constraints and the imperative for continuous data updates. In addressing these challenges, solutions such as federated semi-supervised learning (FSSL), which relies on unlabeled clients' data and a limited amount of labeled data on the server, become pivotal. In this paper, we propose FedAnchor, an innovative FSSL method that introduces a unique double-head structure, called anchor head, paired with the classification head trained exclusively on labeled anchor data on the server. The anchor head is empowered with a newly designed label contrastive loss based on the cosine similarity metric. Our approach mitigates the confirmation bias and overfitting issues associated with pseudo-labeling techniques based on high-confidence model prediction samples. Extensive experiments on CIFAR10/100 and SVHN datasets demonstrate that our method outperforms the state-of-the-art method by a significant margin in terms of convergence rate and model accuracy.

FedL2P: Federated Learning to Personalize

Federated learning (FL) research has made progress in developing algorithms for distributed learning of global models, as well as algorithms for local personalization of those common models to the specifics of each client's local data distribution. However, different FL problems may require different personalization strategies, and it may not even be possible to define an effective one-size-fits-all personalization strategy for all clients: depending on how similar each client's optimal predictor is to that of the global model, different personalization strategies may be preferred. In this paper, we consider the federated meta-learning problem of learning personalization strategies. Specifically, we consider meta-nets that induce the batch-norm and learning rate parameters for each client given local data statistics. By learning these meta-nets through FL, we allow the whole FL network to collaborate in learning a customized personalization strategy for each client. Empirical results show that this framework improves on a range of standard hand-crafted personalization baselines in both label and feature shift situations.

FedVal: Different good or different bad in federated learning

Federated learning (FL) systems are susceptible to attacks from malicious actors who might attempt to corrupt the training model through various poisoning attacks. FL also poses new challenges in addressing group bias, such as ensuring fair performance for different demographic groups. Traditional methods used to address such biases require centralized access to the data, which FL systems do not have. In this paper, we present a novel approach FedVal for both robustness and fairness that does not require any additional information from clients that could raise privacy concerns and consequently compromise the integrity of the FL system. To this end, we propose an innovative score function based on a server-side validation method that assesses client updates and determines the optimal aggregation balance between locally-trained models. Our research shows that this approach not only provides solid protection against poisoning attacks but can also be used to reduce group bias and subsequently promote fairness while maintaining the system's capability for differential privacy. Extensive experiments on the CIFAR-10, FEMNIST, and PUMS ACSIncome datasets in different configurations demonstrate the effectiveness of our method, resulting in state-of-the-art performances. We have proven robustness in situations where 80% of participating clients are malicious. Additionally, we have shown a significant increase in accuracy for underrepresented labels from 32% to 53%, and increase in recall rate for underrepresented features from 19% to 50%.

vFedSec: Efficient Secure Aggregation for Vertical Federated Learning via Secure Layer

Most work in privacy-preserving federated learning (FL) has been focusing on horizontally partitioned datasets where clients share the same sets of features and can train complete models independently. However, in many interesting problems, individual data points are scattered across different clients/organizations in a vertical setting. Solutions for this type of FL require the exchange of intermediate outputs and gradients between participants, posing a potential risk of privacy leakage when privacy and security concerns are not considered. In this work, we present vFedSec - a novel design with an innovative Secure Layer for training vertical FL securely and efficiently using state-of-the-art security modules in secure aggregation. We theoretically demonstrate that our method does not impact the training performance while protecting private data effectively. Empirically results also show its applicability with extensive experiments that our design can achieve the protection with negligible computation and communication overhead. Also, our method can obtain 9.1e2 ~ 3.8e4 speedup compared to widely-adopted homomorphic encryption (HE) method.