Although large foundation models pre-trained by self-supervised learning have achieved state-of-the-art performance in many tasks including automatic speech recognition (ASR), knowledge distillation (KD) is often required in practice to transfer the knowledge learned by large teacher models into much smaller student models with affordable computation and memory costs. This paper proposes a novel two-stage KD framework to distil the knowledge from multiple speech foundation models as teachers into a single student neural transducer model for ASR. In the first stage, the student model encoder is pre-trained using the embeddings extracted from multiple teacher models. In the second stage, the student encoder is fine-tuned with the audio-text pairs based on the ASR task. Experiments on the LibriSpeech 100-hour subset show that the proposed KD framework improves the performance of both streaming and non-streaming student models when using only one teacher. The performance of the student model can be further enhanced when multiple teachers are used jointly, achieving word error rate reductions (WERRs) of 17.5% and 10.6%. Our proposed framework can be combined with other existing KD methods to achieve further improvements. Further WERRs were obtained by incorporating extra unlabelled data during encoder pre-training, leading to a total relative WERR of 55.0% on the non-streaming student model.
Large deep learning models have shown great potential for delivering exceptional results in various applications. However, the training process can be incredibly challenging due to the models' vast parameter sizes, often consisting of hundreds of billions of parameters. Common distributed training methods, such as data parallelism, tensor parallelism, and pipeline parallelism, demand significant data communication throughout the process, leading to prolonged wait times for some machines in physically distant distributed systems. To address this issue, we propose a novel solution called Hulk, which utilizes a modified graph neural network to optimize distributed computing systems. Hulk not only optimizes data communication efficiency between different countries or even different regions within the same city, but also provides optimal distributed deployment of models in parallel. For example, it can place certain layers on a machine in a specific region or pass specific parameters of a model to a machine in a particular location. By using Hulk in experiments, we were able to improve the time efficiency of training large deep learning models on distributed systems by more than 20\%. Our open source collection of unlabeled data:https://github.com/DLYuanGod/Hulk.
Word-piece models (WPMs) are commonly used subword units in state-of-the-art end-to-end automatic speech recognition (ASR) systems. For multilingual ASR, due to the differences in written scripts across languages, multilingual WPMs bring the challenges of having overly large output layers and scaling to more languages. In this work, we propose a universal monolingual output layer (UML) to address such problems. Instead of one output node for only one WPM, UML re-associates each output node with multiple WPMs, one for each language, and results in a smaller monolingual output layer shared across languages. Consequently, the UML enables to switch in the interpretation of each output node depending on the language of the input speech. Experimental results on an 11-language voice search task demonstrated the feasibility of using UML for high-quality and high-efficiency multilingual streaming ASR.
Pre-trained models have been used in many fields in recent years, ranging from natural language understanding to computer vision and natural language generation. Nowadays, the performance of these natural language generation models is overly dependent on the model's scale and the dataset's size. While the larger language model is excellent in some respects, it cannot learn up-to-date knowledge and is relatively difficult to relearn. In this paper, a new adversarial process learning method is called Auto-Learning, which can improve the performance of any natural language generation model without the help of additional datasets. Auto-Learning includes two models: $G$ is a text generation model, and $D$ can test whether the data generated by G is legitimate. Firstly, the fine-tuned $D$ model is used as the brain's knowledge base before the process. Then the text generated by the $G$ model is used as the input of $D$ to determine whether the text is legitimate. Finally, $G$ is fine-tuned according to the output of $D$. This adversarial process is like a self-escalation of the brain through some a priori knowledge. When this adversarial system wants to learn something new, simply fine-tune the $D$ model. Our approach applies to Autoregressive Language Modeling for all Transformer classes. Auto-Learning enables 8 models to achieve stable improvement in 10 natural language processing tasks without any change in structure.
Recently, inspired by DETR variants, query-based end-to-end instance segmentation (QEIS) methods have outperformed CNN-based models on large-scale datasets. Yet they would lose efficacy when only a small amount of training data is available since it's hard for the crucial queries/kernels to learn localization and shape priors. To this end, this work offers a novel unsupervised pre-training solution for low-data regimes. Inspired by the recent success of the Prompting technique, we introduce a new pre-training method that boosts QEIS models by giving Saliency Prompt for queries/kernels. Our method contains three parts: 1) Saliency Masks Proposal is responsible for generating pseudo masks from unlabeled images based on the saliency mechanism. 2) Prompt-Kernel Matching transfers pseudo masks into prompts and injects the corresponding localization and shape priors to the best-matched kernels. 3) Kernel Supervision is applied to supply supervision at the kernel level for robust learning. From a practical perspective, our pre-training method helps QEIS models achieve a similar convergence speed and comparable performance with CNN-based models in low-data regimes. Experimental results show that our method significantly boosts several QEIS models on three datasets. Code will be made available.
The growing demand for optimal and low-power energy consumption paradigms for Internet of Things (IoT) devices has garnered significant attention due to their cost-effectiveness, simplicity, and intelligibility. We propose an Artificial Intelligence (AI) hardware energy-efficient framework to achieve optimal energy savings in heterogeneous computing through appropriate power consumption management. A deep reinforcement learning framework is employed, utilizing the Actor-Critic architecture to provide a simple and precise method for power saving. The results of the study demonstrate the proposed approach's suitability for different hardware configurations, achieving notable energy consumption control while adhering to strict performance requirements. The evaluation of the proposed power-saving framework shows that it is more stable, and has achieved more than 23% efficiency improvement, outperforming other methods by more than 5%.
Training deep neural networks (DNNs) with limited supervision has been a popular research topic as it can significantly alleviate the annotation burden. Self-training has been successfully applied in semi-supervised learning tasks, but one drawback of self-training is that it is vulnerable to the label noise from incorrect pseudo labels. Inspired by the fact that samples with similar labels tend to share similar representations, we develop a neighborhood-based sample selection approach to tackle the issue of noisy pseudo labels. We further stabilize self-training via aggregating the predictions from different rounds during sample selection. Experiments on eight tasks show that our proposed method outperforms the strongest self-training baseline with 1.83% and 2.51% performance gain for text and graph datasets on average. Our further analysis demonstrates that our proposed data selection strategy reduces the noise of pseudo labels by 36.8% and saves 57.3% of the time when compared with the best baseline. Our code and appendices will be uploaded to https://github.com/ritaranx/NeST.
Online media data, in the forms of images and videos, are becoming mainstream communication channels. However, recent advances in deep learning, particularly deep generative models, open the doors for producing perceptually convincing images and videos at a low cost, which not only poses a serious threat to the trustworthiness of digital information but also has severe societal implications. This motivates a growing interest of research in media tampering detection, i.e., using deep learning techniques to examine whether media data have been maliciously manipulated. Depending on the content of the targeted images, media forgery could be divided into image tampering and Deepfake techniques. The former typically moves or erases the visual elements in ordinary images, while the latter manipulates the expressions and even the identity of human faces. Accordingly, the means of defense include image tampering detection and Deepfake detection, which share a wide variety of properties. In this paper, we provide a comprehensive review of the current media tampering detection approaches, and discuss the challenges and trends in this field for future research.
In tunnel boring machine (TBM) underground projects, an accurate description of the rock-soil types distributed in the tunnel can decrease the construction risk ({\it e.g.} surface settlement and landslide) and improve the efficiency of construction. In this paper, we propose an active learning framework, called AL-iGAN, for tunnel geological reconstruction based on TBM operational data. This framework contains two main parts: one is the usage of active learning techniques for recommending new drilling locations to label the TBM operational data and then to form new training samples; and the other is an incremental generative adversarial network for geological reconstruction (iGAN-GR), whose weights can be incrementally updated to improve the reconstruction performance by using the new samples. The numerical experiment validate the effectiveness of the proposed framework as well.