Self-supervised learning (SSL) methods such as WavLM have shown promising speech separation (SS) results in small-scale simulation-based experiments. In this work, we extend the exploration of the SSL-based SS by massively scaling up both the pre-training data (more than 300K hours) and fine-tuning data (10K hours). We also investigate various techniques to efficiently integrate the pre-trained model with the SS network under a limited computation budget, including a low frame rate SSL model training setup and a fine-tuning scheme using only the part of the pre-trained model. Compared with a supervised baseline and the WavLM-based SS model using feature embeddings obtained with the previously released 94K hours trained WavLM, our proposed model obtains 15.9% and 11.2% of relative word error rate (WER) reductions, respectively, for a simulated far-field speech mixture test set. For conversation transcription on real meeting recordings using continuous speech separation, the proposed model achieves 6.8% and 10.6% of relative WER reductions over the purely supervised baseline on AMI and ICSI evaluation sets, respectively, while reducing the computational cost by 38%.

Nature inspired algorithms has brought solutions to complex problems in optimization where the optimization and solution of complex problems is highly complex and nonlinear. There is a need to use proper design of the cost function or the fitness function in terms of the parameters to be optimized, this can be used in solving any type of such problems. In this paper the nature inspired algorithms has played important role in the optimal design of antenna array with improved radiation characteristics. In this paper, 20 elements linearly spaced array is used as an example of nature inspired optimization in antenna array system. This bridge inspired army ant algorithm(NOABS) is used to reduce the side lobes and to improve the other radiation characteristics to show the effect of the optimization on design characteristics by implementation of NOABS nature inspired algorithm. The entire simulation is carried out on 20 elements linear antenna array.

The aim of this paper is to introduce AHCOA to the electromagnetic and antenna community. AHCOA is a new nature inspired meta heuristic algorithm inspired by how there is a hierarchy and departments in the ant hill colonization. It has high probabilistic potential in solving not only unconstrained but also constrained optimization problems. In this paper the AHCOA is applied to linear antenna array for better pattern synthesis in the following ways : By uniform excitation considering equal spacing of the antenna elements with respect to the uniform array. AHCOA is used in obtaining an array pattern to achieve minimum side lobe levels. The results are compared to other state of the art nature based algorithms such as ant lion optimizer, which show a considerable improvement in AHCOA.

This paper aims to introduce the Ant hill colonization optimization algorithm(AHCOA) to the electromagnetics and antenna community. The ant hill is built by special species of ants known as formicas ants(also meadow ants, fire ants and harvester ants). AHCOA is a novel new nature inspired algorithm mimicking how the ants built and sustain the ant hill for their survival and sustenance for many years. This problem solves constrained and unconstrained optimization problems with wide capability in diverse fields. AHCOA is used by writing equations of volumetric analysis of the ant hill mould the manner in which the structure is architected. In this paper, we have shown how AHCOA is better than the previous paper on ant lion optimizer for controlling side lobe in antenna pattern synthesis in paper [1]. The potential of AHCOA in synthesizing and analyzing for d/ varying from 1.1,0.6,0.5,0.3 and 0.1 linear array is also illustrated. Antenna side lobe level minimization is compared with ant lion optimizer showing why AHCOA is better than the previously simulated ant lion optimizer for side lobe control. The results show why linear arrays are better synthesized for AHCOA then other algorithms used in planar arrays. This paper shows why AHCOA is a strong candidate for antenna optimization used in linear arrays.

为满足商业和科学应用的深度学习的极端计算需求，Dataflow加速器越来越受欢迎。虽然这些“特定于域的”加速器不是完全可编程的CPU和GPU，但它们对数据编程，即数据流和平铺优化来保持不同的灵活性，以提高效率。在设计新的算法和映射方法时，在新硬件上执行目标问题的算法存在若干挑战。以前的作品单独解决了这些挑战。为了解决整体挑战，在这项工作中，我们在流行的MLIR编译基础架构中，我们为一个名为Union的空间加速器提供了HW-SW Co-Design生态系统。我们的框架允许在几种加速器成本模型上探索不同的算法及其映射。联盟还包括一个加速器成本模型和映射器的即插即用库，可以轻松扩展。算法和加速器成本模型通过新颖的映射抽象来连接，该抽象捕获空间加速器的地图空间，该空间加速器可以基于来自硬件，工作负载和映射器的约束来系统地修剪。我们展示了与多个案例研究的社区联盟的价值，该研究将使用不同的映射方案在不同的加速器架构上卸载不同的张量操作（Conv / Gemm / Tensor收缩）。