Federated Learning (FL) is a technique that allows multiple participants to collaboratively train a Deep Neural Network (DNN) without the need of centralizing their data. Among other advantages, it comes with privacy-preserving properties making it attractive for application in sensitive contexts, such as health care or the military. Although the data are not explicitly exchanged, the training procedure requires sharing information about participants' models. This makes the individual models vulnerable to theft or unauthorized distribution by malicious actors. To address the issue of ownership rights protection in the context of Machine Learning (ML), DNN Watermarking methods have been developed during the last five years. Most existing works have focused on watermarking in a centralized manner, but only a few methods have been designed for FL and its unique constraints. In this paper, we provide an overview of recent advancements in Federated Learning watermarking, shedding light on the new challenges and opportunities that arise in this field.
相關內容
Deeper Vision Transformers (ViTs) are more challenging to train. We expose a degradation problem in deeper layers of ViT when using masked image modeling (MIM) for pre-training. To ease the training of deeper ViTs, we introduce a self-supervised learning framework called \textbf{M}asked \textbf{I}mage \textbf{R}esidual \textbf{L}earning (\textbf{MIRL}), which significantly alleviates the degradation problem, making scaling ViT along depth a promising direction for performance upgrade. We reformulate the pre-training objective for deeper layers of ViT as learning to recover the residual of the masked image. We provide extensive empirical evidence showing that deeper ViTs can be effectively optimized using MIRL and easily gain accuracy from increased depth. With the same level of computational complexity as ViT-Base and ViT-Large, we instantiate 4.5{$\times$} and 2{$\times$} deeper ViTs, dubbed ViT-S-54 and ViT-B-48. The deeper ViT-S-54, costing 3{$\times$} less than ViT-Large, achieves performance on par with ViT-Large. ViT-B-48 achieves 86.2\% top-1 accuracy on ImageNet. On one hand, deeper ViTs pre-trained with MIRL exhibit excellent generalization capabilities on downstream tasks, such as object detection and semantic segmentation. On the other hand, MIRL demonstrates high pre-training efficiency. With less pre-training time, MIRL yields competitive performance compared to other approaches.
Graph Hypernetworks (GHN) can predict the parameters of varying unseen CNN architectures with surprisingly good accuracy at a fraction of the cost of iterative optimization. Following these successes, preliminary research has explored the use of GHNs to predict quantization-robust parameters for 8-bit and 4-bit quantized CNNs. However, this early work leveraged full-precision float32 training and only quantized for testing. We explore the impact of quantization-aware training and/or other quantization-based training strategies on quantized robustness and performance of GHN predicted parameters for low-precision CNNs. We show that quantization-aware training can significantly improve quantized accuracy for GHN predicted parameters of 4-bit quantized CNNs and even lead to greater-than-random accuracy for 2-bit quantized CNNs. These promising results open the door for future explorations such as investigating the use of GHN predicted parameters as initialization for further quantized training of individual CNNs, further exploration of "extreme bitwidth" quantization, and mixed precision quantization schemes.
Machine unlearning (MU) is gaining increasing attention due to the need to remove or modify predictions made by machine learning (ML) models. While training models have become more efficient and accurate, the importance of unlearning previously learned information has become increasingly significant in fields such as privacy, security, and fairness. This paper presents a comprehensive survey of MU, covering current state-of-the-art techniques and approaches, including data deletion, perturbation, and model updates. In addition, commonly used metrics and datasets are also presented. The paper also highlights the challenges that need to be addressed, including attack sophistication, standardization, transferability, interpretability, training data, and resource constraints. The contributions of this paper include discussions about the potential benefits of MU and its future directions. Additionally, the paper emphasizes the need for researchers and practitioners to continue exploring and refining unlearning techniques to ensure that ML models can adapt to changing circumstances while maintaining user trust. The importance of unlearning is further highlighted in making Artificial Intelligence (AI) more trustworthy and transparent, especially with the increasing importance of AI in various domains that involve large amounts of personal user data.
Distributed training is the de facto standard to scale up the training of Deep Neural Networks (DNNs) with multiple GPUs. The performance bottleneck of distributed training lies in communications for gradient synchronization. Recently, practitioners have observed sparsity in gradient tensors, suggesting the potential to reduce the traffic volume in communication and improve end-to-end training efficiency. Yet, the optimal communication scheme to fully leverage sparsity is still missing. This paper aims to address this gap. We first analyze the characteristics of sparse tensors in popular DNN models to understand the fundamentals of sparsity. We then systematically explore the design space of communication schemes for sparse tensors and find the optimal one. % We then find the optimal scheme based on the characteristics by systematically exploring the design space. We also develop a gradient synchronization system called Zen that approximately realizes it for sparse tensors. We demonstrate that Zen can achieve up to 5.09x speedup in communication time and up to 2.48x speedup in training throughput compared to the state-of-the-art methods.
Machine Translation (MT) is one of the most prominent tasks in Natural Language Processing (NLP) which involves the automatic conversion of texts from one natural language to another while preserving its meaning and fluency. Although the research in machine translation has been going on since multiple decades, the newer approach of integrating deep learning techniques in natural language processing has led to significant improvements in the translation quality. In this paper, we have developed a Neural Machine Translation (NMT) system by training the Transformer model to translate texts from Indian Language Hindi to English. Hindi being a low resource language has made it difficult for neural networks to understand the language thereby leading to a slow growth in the development of neural machine translators. Thus, to address this gap, we implemented back-translation to augment the training data and for creating the vocabulary, we experimented with both word and subword level tokenization using Byte Pair Encoding (BPE) thereby ending up training the Transformer in 10 different configurations. This led us to achieve a state-of-the-art BLEU score of 24.53 on the test set of IIT Bombay English-Hindi Corpus in one of the configurations.
Neural machine translation (NMT) has shown impressive performance when trained on large-scale corpora. However, generic NMT systems have demonstrated poor performance on out-of-domain translation. To mitigate this issue, several domain adaptation methods have recently been proposed which often lead to better translation quality than genetic NMT systems. While there has been some continuous progress in NMT for English and other European languages, domain adaption in Arabic has received little attention in the literature. The current study, therefore, aims to explore the effectiveness of domain-specific adaptation for Arabic MT (AMT), in yet unexplored domain, financial news articles. To this end, we developed carefully a parallel corpus for Arabic-English (AR- EN) translation in the financial domain for benchmarking different domain adaptation methods. We then fine-tuned several pre-trained NMT and Large Language models including ChatGPT-3.5 Turbo on our dataset. The results showed that the fine-tuning is successful using just a few well-aligned in-domain AR-EN segments. The quality of ChatGPT translation was superior than other models based on automatic and human evaluations. To the best of our knowledge, this is the first work on fine-tuning ChatGPT towards financial domain transfer learning. To contribute to research in domain translation, we made our datasets and fine-tuned models available at //huggingface.co/asas-ai/.
Large Language Models (LLMs), acting as a powerful reasoner and generator, exhibit extraordinary performance across various natural language tasks, such as question answering (QA). Among these tasks, Multi-Hop Question Answering (MHQA) stands as a widely discussed category, necessitating seamless integration between LLMs and the retrieval of external knowledge. Existing methods employ LLM to generate reasoning paths and plans, and utilize IR to iteratively retrieve related knowledge, but these approaches have inherent flaws. On one hand, Information Retriever (IR) is hindered by the low quality of generated queries by LLM. On the other hand, LLM is easily misguided by the irrelevant knowledge by IR. These inaccuracies, accumulated by the iterative interaction between IR and LLM, lead to a disaster in effectiveness at the end. To overcome above barriers, in this paper, we propose a novel pipeline for MHQA called Furthest-Reasoning-with-Plan-Assessment (FuRePA), including an improved framework (Furthest Reasoning) and an attached module (Plan Assessor). 1) Furthest reasoning operates by masking previous reasoning path and generated queries for LLM, encouraging LLM generating chain of thought from scratch in each iteration. This approach enables LLM to break the shackle built by previous misleading thoughts and queries (if any). 2) The Plan Assessor is a trained evaluator that selects an appropriate plan from a group of candidate plans proposed by LLM. Our methods are evaluated on three highly recognized public multi-hop question answering datasets and outperform state-of-the-art on most metrics (achieving a 10%-12% in answer accuracy).
Sparsifying the Transformer has garnered considerable interest, as training the Transformer is very computationally demanding. Prior efforts to sparsify the Transformer have either used a fixed pattern or data-driven approach to reduce the number of operations involving the computation of multi-head attention, which is the main bottleneck of the Transformer. However, existing methods suffer from inevitable problems, such as the potential loss of essential sequence features due to the uniform fixed pattern applied across all layers, and an increase in the model size resulting from the use of additional parameters to learn sparsity patterns in attention operations. In this paper, we propose a novel sparsification scheme for the Transformer that integrates convolution filters and the flood filling method to efficiently capture the layer-wise sparse pattern in attention operations. Our sparsification approach reduces the computational complexity and memory footprint of the Transformer during training. Efficient implementations of the layer-wise sparsified attention algorithm on GPUs are developed, demonstrating a new SPION that achieves up to 3.08X speedup over existing state-of-the-art sparse Transformer models, with better evaluation quality.
Knowledge Graph Embedding (KGE) aims to learn representations for entities and relations. Most KGE models have gained great success, especially on extrapolation scenarios. Specifically, given an unseen triple (h, r, t), a trained model can still correctly predict t from (h, r, ?), or h from (?, r, t), such extrapolation ability is impressive. However, most existing KGE works focus on the design of delicate triple modeling function, which mainly tells us how to measure the plausibility of observed triples, but offers limited explanation of why the methods can extrapolate to unseen data, and what are the important factors to help KGE extrapolate. Therefore in this work, we attempt to study the KGE extrapolation of two problems: 1. How does KGE extrapolate to unseen data? 2. How to design the KGE model with better extrapolation ability? For the problem 1, we first discuss the impact factors for extrapolation and from relation, entity and triple level respectively, propose three Semantic Evidences (SEs), which can be observed from train set and provide important semantic information for extrapolation. Then we verify the effectiveness of SEs through extensive experiments on several typical KGE methods. For the problem 2, to make better use of the three levels of SE, we propose a novel GNN-based KGE model, called Semantic Evidence aware Graph Neural Network (SE-GNN). In SE-GNN, each level of SE is modeled explicitly by the corresponding neighbor pattern, and merged sufficiently by the multi-layer aggregation, which contributes to obtaining more extrapolative knowledge representation. Finally, through extensive experiments on FB15k-237 and WN18RR datasets, we show that SE-GNN achieves state-of-the-art performance on Knowledge Graph Completion task and performs a better extrapolation ability.
Pre-trained Language Models (PLMs) have achieved great success in various Natural Language Processing (NLP) tasks under the pre-training and fine-tuning paradigm. With large quantities of parameters, PLMs are computation-intensive and resource-hungry. Hence, model pruning has been introduced to compress large-scale PLMs. However, most prior approaches only consider task-specific knowledge towards downstream tasks, but ignore the essential task-agnostic knowledge during pruning, which may cause catastrophic forgetting problem and lead to poor generalization ability. To maintain both task-agnostic and task-specific knowledge in our pruned model, we propose ContrAstive Pruning (CAP) under the paradigm of pre-training and fine-tuning. It is designed as a general framework, compatible with both structured and unstructured pruning. Unified in contrastive learning, CAP enables the pruned model to learn from the pre-trained model for task-agnostic knowledge, and fine-tuned model for task-specific knowledge. Besides, to better retain the performance of the pruned model, the snapshots (i.e., the intermediate models at each pruning iteration) also serve as effective supervisions for pruning. Our extensive experiments show that adopting CAP consistently yields significant improvements, especially in extremely high sparsity scenarios. With only 3% model parameters reserved (i.e., 97% sparsity), CAP successfully achieves 99.2% and 96.3% of the original BERT performance in QQP and MNLI tasks. In addition, our probing experiments demonstrate that the model pruned by CAP tends to achieve better generalization ability.
小貼士
登錄享
相關主題
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191