A transfer learning-based end-to-end speech recognition approach is presented in two levels in our framework. Firstly, a feature extraction approach combining multilingual deep neural network (DNN) training with matrix factorization algorithm is introduced to extract high-level features. Secondly, the advantage of connectionist temporal classification (CTC) is transferred to the target attention-based model through a joint CTC-attention model composed of shallow recurrent neural networks (RNNs) on top of the proposed features. The experimental results show that the proposed transfer learning approach achieved the best performance among all end-to-end methods and could be comparable to the state-of-the-art speech recognition system for TIMIT when further jointly decoded with a RNN language model.

As a promising model in Quantum Machine Learning (QML), Quantum Generative Adversarial Networks (QGANs) are rapidly advancing, offering applications in image processing and generation. However, another emerging paradigm represents an image as a Quantum Implicit Neural Representation (QINR). In this work, we propose a novel architectural technique for building QINR-based QGAN to enhance the quality of images generated by QGANs. Additionally, we integrate classical techniques, such as Gradient Penalty and Wasserstein distance, to train QINR-QGAN. In image generation tasks, we demonstrated that QINR-QGAN can achieve performance comparable to state-of-the-art (SOTA) models while significantly reducing the number of trainable quantum parameters. Specifically, QINR-QGAN reduced the trainable quantum parameters by nearly 10 times compared to PQWGAN (Tsang et al. in IEEE Trans. Quantum Eng. 4:1–19, 2023) and Quantum AnoGAN (Herr et al. Quantum Sci. Technol. 6(4): 045004, 2021), demonstrating its superior efficiency in parameter optimization without sacrificing performance. Furthermore, we conducted experiments on the CelebA dataset to tackle a more complex task and generate larger images ( $78\times 64$ ). The results indicate that our model is capable of successfully completing the face generation task.

The end-to-end speech translation (E2E-ST) model has gradually become a mainstream paradigm due to its low latency and less error propagation. However, it is non-trivial to train such a model well due to the task complexity and data scarcity. The speech-and-text modality differences result in the E2E-ST model performance usually inferior to the corresponding machine translation (MT) model. Based on the above observation, existing methods often use sharing mechanisms to carry out implicit knowledge transfer by imposing various constraints. However, the final model often performs worse on the MT task than the MT model trained alone, which means that the knowledge transfer ability of this method is also limited. To deal with these problems, we propose the FCCL ( F ine- and C oarse- Granularity C ontrastive L earning) approach for E2E-ST, which makes explicit knowledge transfer through cross-modal multi-grained contrastive learning. A key ingredient of our approach is applying contrastive learning at both sentence- and frame-level to give the comprehensive guide for extracting speech representations containing rich semantic information. In addition, we adopt a simple whitening method to alleviate the representation degeneration in the MT model, which adversely affects contrast learning. Experiments on the MuST-C benchmark show that our proposed approach significantly outperforms the state-of-the-art E2E-ST baselines on all eight language pairs. Further analysis indicates that FCCL can free up its capacity from learning grammatical structure information and force more layers to learn semantic information.

We introduce DropDim, a structured dropout method designed for regularizing the self-attention mechanism, which is a key component of the transformer. In contrast to the general dropout method, which randomly drops neurons, DropDim drops part of the embedding dimensions. In this way, the semantic information can be completely discarded. Thus, the excessive co-adapting between different embedding dimensions can be broken, and the self-attention is forced to encode meaningful features with a certain number of embedding dimensions erased. Experiments on a wide range of tasks executed on the MUST-C English-Germany dataset show that DropDim can effectively improve model performance, reduce over-fitting, and show complementary effects with other regularization methods. When combined with label smoothing, the WER can be reduced from 19.1% to 15.1% on the ASR task, and the BLEU value can be increased from 26.90 to 28.38 on the MT task. On the ST task, the model can reach a BLEU score of 22.99, an increase by 1.86 BLEU points compared to the strong baseline.

Variational quantum algorithms (VQAs) have been successfully applied to quantum approximate optimization algorithms, variational quantum compiling, and quantum machine learning models. The performance of VQAs is significantly influenced by the architecture of parameterized quantum circuits (PQCs). Quantum architecture search aims to automatically discover high-performance quantum circuits for specific VQA tasks. Quantum architecture search algorithms that utilize both SuperCircuit training and a parameter-sharing approach can save computational resources. If we directly follow the parameter-sharing approach, the SuperCircuit has to be trained to compensate for the worse search space. To address the challenges posed by the worse search space, we introduce an optimization strategy known as the efficient continuous evolutionary approach using Non-dominated Sorting Genetic Algorithm-II (NSGA-II). Then, we leverage prior information (symmetric property) designing Structure Symmetric Pruning for removing redundant gates of the searched ansatz. Experiments show that the efficient continuous evolutionary approach can search for more quantum architectures with better performance; the number of high-performance ansatzes obtained by our method is 10% higher than that in the literature (Du et al. in npj Quantum Inf. 8:62, 2022). The application of Structure Symmetric Pruning effectively reduces the number of parameters in quantum circuits without compromising their performance significantly. In binary classification tasks, the pruned quantum circuits exhibit an average accuracy reduction of 0.044 compared to their unpruned counterparts.