A main problem with reproducing machine learning publications is the variance of metric implementations across papers. A lack of standardization leads to different behavior in mechanisms such as checkpointing, learning rate schedulers or early stopping, that will influence the reported results. For example, a complex metric such as Fréchet inception distance (FID) for synthetic image quality evaluation (Heusel et al., 2017) will differ based on the specific interpolation method used.

No abstract

Since the membrane algorithm was proposed, it has been used for many optimization problems such as, traveling salesman problem, the knapsack problem, and so on. In membrane algorithms, the membranes have two functions: container and comparator. As a container, each membrane contains one evolutionary algorithm like genetic algorithm and ant colony algorithm. These algorithms are called sub-algorithms and used to evolve individuals. As a comparator, the membrane will compare the results of sub-algorithms, and select the best as the base of the next evolvement. This paper proposes a novel evolutionary algorithm called membrane evolutionary algorithm framework (MEAF). Unlike the presented membrane algorithms, the membranes in MEAF will be evolved to solve problems by using four operators that are abstracted from the life cycle of living cells. Based on MEAF, a membrane evolutionary algorithm called MEAMVC is proposed to solve the minimum vertex cover (MVC) problem. The experimental results show the advantages of MEAMVC when MEAMVC is compared with two state-of-the-art MVC algorithms proposed in recent years.

No abstract

This work proposes a neural network to extensively exploit spatial information for multichannel joint speech separation, denoising and dereverberation, named SpatialNet. In the short-time Fourier transform (STFT) domain, the proposed network performs end-to-end speech enhancement. It is mainly composed of interleaved narrow-band and cross-band blocks to respectively exploit narrow-band and cross-band spatial information. The narrow-band blocks process frequencies independently, and use self-attention mechanism and temporal convolutional layers to respectively perform spatial-feature-based speaker clustering and temporal smoothing/filtering. The crossband blocks processes frames independently, and use full-band linear layer and frequency convolutional layers to respectively learn the correlation between all frequencies and adjacent frequencies. Experiments are conducted on various simulated and real datasets, and the results show that 1) the proposed network achieves the state-of-the-art performance on almost all tasks; 2) the proposed network suffers little from the spectral generalization problem; and 3) the proposed network is indeed performing speaker clustering (demonstrated by attention maps).