Dataset condensation aims at reducing the network training effort through condensing a cumbersome training set into a compact synthetic one. State-of-the-art approaches largely rely on learning the synthetic data by matching the gradients between the real and synthetic data batches. Despite the intuitive motivation and promising results, such gradient-based methods, by nature, easily overfit to a biased set of samples that produce dominant gradients, and thus lack a global supervision of data distribution. In this paper, we propose a novel scheme to Condense dataset by Aligning FEatures (CAFE), which explicitly attempts to preserve the real-feature distribution as well as the discriminant power of the resulting synthetic set, lending itself to strong generalization capability to various architectures. At the heart of our approach is an effective strategy to align features from the real and synthetic data across various scales, while accounting for the classification of real samples. Our scheme is further backed up by a novel dynamic bi-level optimization, which adaptively adjusts parameter updates to prevent over-/under-fitting. We validate the proposed CAFE across various datasets, and demonstrate that it generally outperforms the state of the art: on the SVHN dataset, for example, the performance gain is up to 11%. Extensive experiments and analysis verify the effectiveness and necessity of proposed designs.

No abstract

No abstract

Hydraulic systems faults have the characteristics of being highly concealed and unclear. Due to the characteristics of the complex vibration transmission mechanism and strong nonlinear time-varying signals in hydraulic systems, it is extremely difficult to achieve fault diagnosis for hydraulic systems. Different components of the system can fail individually or simultaneously. Signal processing faces the problem of coupling between multi-component faults, which makes it more difficult to realise multi-component fault diagnosis. On the one hand, existing techniques rely on hand-designed features and only use a traditional single shallow machine model as the base classifier, and these do not have the ability to self-learn meaningful features. On the other hand, the diagnostic performance of a single base classifier sometimes does not meet engineering requirements. To handle the above problems, a bagging strategy based heterogeneous ensemble deep neural networks (DNNs) approach is proposed for the multiple components fault diagnosis of hydraulic systems. First, Pearson correlation coefficient and neighbourhood component analysis are developed for data channel selection and feature dimensionality reduction. Second, two distinct DNNs are constructed as base learners: a stacked sparse autoencoder and a deep hierarchical extreme-learning machine. Finally, a bagging strategy is adopted to integrate different DNNs to obtain robust diagnostic results. The results from this experiment demonstrate that the proposed method can precisely diagnose hydraulic system faults compared with comparative methods.

No abstract