Mitigating Catastrophic Forgetting in Large Language Models with Self-Synthesized Rehearsal

Large language models (LLMs) suffer from catastrophic forgetting during continual learning. Conventional rehearsal-based methods rely on previous training data to retain the model's ability, which may not be feasible in real-world applications. When conducting continual learning based on a publicly-released LLM checkpoint, the availability of the original training data may be non-existent. To address this challenge, we propose a framework called Self-Synthesized Rehearsal (SSR) that uses the LLM to generate synthetic instances for rehearsal. Concretely, we first employ the base LLM for in-context learning to generate synthetic instances. Subsequently, we utilize the latest LLM to refine the instance outputs based on the synthetic inputs, preserving its acquired ability. Finally, we select diverse high-quality synthetic instances for rehearsal in future stages. Experimental results demonstrate that SSR achieves superior or comparable performance compared to conventional rehearsal-based approaches while being more data-efficient. Besides, SSR effectively preserves the generalization capabilities of LLMs in general domains.

Wasserstein Differential Privacy

Differential privacy (DP) has achieved remarkable results in the field of privacy-preserving machine learning. However, existing DP frameworks do not satisfy all the conditions for becoming metrics, which prevents them from deriving better basic private properties and leads to exaggerated values on privacy budgets. We propose Wasserstein differential privacy (WDP), an alternative DP framework to measure the risk of privacy leakage, which satisfies the properties of symmetry and triangle inequality. We show and prove that WDP has 13 excellent properties, which can be theoretical supports for the better performance of WDP than other DP frameworks. In addition, we derive a general privacy accounting method called Wasserstein accountant, which enables WDP to be applied in stochastic gradient descent (SGD) scenarios containing sub-sampling. Experiments on basic mechanisms, compositions and deep learning show that the privacy budgets obtained by Wasserstein accountant are relatively stable and less influenced by order. Moreover, the overestimation on privacy budgets can be effectively alleviated. The code is available at https://github.com/Hifipsysta/WDP.

HiMTM: Hierarchical Multi-Scale Masked Time Series Modeling for Long-Term Forecasting

Time series forecasting is crucial and challenging in the real world. The recent surge in interest regarding time series foundation models, which cater to a diverse array of downstream tasks, is noteworthy. However, existing methods often overlook the multi-scale nature of time series, an aspect crucial for precise forecasting. To bridge this gap, we propose HiMTM, a hierarchical multi-scale masked time series modeling method designed for long-term forecasting. Specifically, it comprises four integral components: (1) hierarchical multi-scale transformer (HMT) to capture temporal information at different scales; (2) decoupled encoder-decoder (DED) forces the encoder to focus on feature extraction, while the decoder to focus on pretext tasks; (3) multi-scale masked reconstruction (MMR) provides multi-stage supervision signals for pre-training; (4) cross-scale attention fine-tuning (CSA-FT) to capture dependencies between different scales for forecasting. Collectively, these components enhance multi-scale feature extraction capabilities in masked time series modeling and contribute to improved prediction accuracy. We conduct extensive experiments on 7 mainstream datasets to prove that HiMTM has obvious advantages over contemporary self-supervised and end-to-end learning methods. The effectiveness of HiMTM is further showcased by its application in the industry of natural gas demand forecasting.

Federated Learning in Big Model Era: Domain-Specific Multimodal Large Models

Multimodal data, which can comprehensively perceive and recognize the physical world, has become an essential path towards general artificial intelligence. However, multimodal large models trained on public datasets often underperform in specific industrial domains. This paper proposes a multimodal federated learning framework that enables multiple enterprises to utilize private domain data to collaboratively train large models for vertical domains, achieving intelligent services across scenarios. The authors discuss in-depth the strategic transformation of federated learning in terms of intelligence foundation and objectives in the era of big model, as well as the new challenges faced in heterogeneous data, model aggregation, performance and cost trade-off, data privacy, and incentive mechanism. The paper elaborates a case study of leading enterprises contributing multimodal data and expert knowledge to city safety operation management , including distributed deployment and efficient coordination of the federated learning platform, technical innovations on data quality improvement based on large model capabilities and efficient joint fine-tuning approaches. Preliminary experiments show that enterprises can enhance and accumulate intelligent capabilities through multimodal model federated learning, thereby jointly creating an smart city model that provides high-quality intelligent services covering energy infrastructure safety, residential community security, and urban operation management. The established federated learning cooperation ecosystem is expected to further aggregate industry, academia, and research resources, realize large models in multiple vertical domains, and promote the large-scale industrial application of artificial intelligence and cutting-edge research on multimodal federated learning.

The Prospect of Enhancing Large-Scale Heterogeneous Federated Learning with Transformers

Federated learning (FL) addresses data privacy concerns by enabling collaborative training of AI models across distributed data owners. Wide adoption of FL faces the fundamental challenges of data heterogeneity and the large scale of data owners involved. In this paper, we investigate the prospect of Transformer-based FL models for achieving generalization and personalization in this setting. We conduct extensive comparative experiments involving FL with Transformers, ResNet, and personalized ResNet-based FL approaches under various scenarios. These experiments consider varying numbers of data owners to demonstrate Transformers' advantages over deep neural networks in large-scale heterogeneous FL tasks. In addition, we analyze the superior performance of Transformers by comparing the Centered Kernel Alignment (CKA) representation similarity across different layers and FL models to gain insight into the reasons behind their promising capabilities.

Hierarchical Federated Learning Incentivization for Gas Usage Estimation

Accurately estimating gas usage is essential for the efficient functioning of gas distribution networks and saving operational costs. Traditional methods rely on centralized data processing, which poses privacy risks. Federated learning (FL) offers a solution to this problem by enabling local data processing on each participant, such as gas companies and heating stations. However, local training and communication overhead may discourage gas companies and heating stations from actively participating in the FL training process. To address this challenge, we propose a Hierarchical FL Incentive Mechanism for Gas Usage Estimation (HI-GAS), which has been testbedded in the ENN Group, one of the leading players in the natural gas and green energy industry. It is designed to support horizontal FL among gas companies, and vertical FL among each gas company and heating station within a hierarchical FL ecosystem, rewarding participants based on their contributions to FL. In addition, a hierarchical FL model aggregation approach is also proposed to improve the gas usage estimation performance by aggregating models at different levels of the hierarchy. The incentive scheme employs a multi-dimensional contribution-aware reward distribution function that combines the evaluation of data quality and model contribution to incentivize both gas companies and heating stations within their jurisdiction while maintaining fairness. Results of extensive experiments validate the effectiveness of the proposed mechanism.

Efficient Training of Large-scale Industrial Fault Diagnostic Models through Federated Opportunistic Block Dropout

Artificial intelligence (AI)-empowered industrial fault diagnostics is important in ensuring the safe operation of industrial applications. Since complex industrial systems often involve multiple industrial plants (possibly belonging to different companies or subsidiaries) with sensitive data collected and stored in a distributed manner, collaborative fault diagnostic model training often needs to leverage federated learning (FL). As the scale of the industrial fault diagnostic models are often large and communication channels in such systems are often not exclusively used for FL model training, existing deployed FL model training frameworks cannot train such models efficiently across multiple institutions. In this paper, we report our experience developing and deploying the Federated Opportunistic Block Dropout (FEDOBD) approach for industrial fault diagnostic model training. By decomposing large-scale models into semantic blocks and enabling FL participants to opportunistically upload selected important blocks in a quantized manner, it significantly reduces the communication overhead while maintaining model performance. Since its deployment in ENN Group in February 2022, FEDOBD has served two coal chemical plants across two cities in China to build industrial fault prediction models. It helped the company reduce the training communication overhead by over 70% compared to its previous AI Engine, while maintaining model performance at over 85% test F1 score. To our knowledge, it is the first successfully deployed dropout-based FL approach.