Non-Matching Reference-based Speech Quality Assessment models typically require numerous references during inference to ensure stable and accurate predictions. However, this dependency introduces significant computational overhead, limiting their suitability for real-time applications. In this paper, we propose a novel training paradigm that directly addresses prediction instability at its source by integrating multiple references during training rather than during inference, as in existing approaches. This method allows the model to capture the inherent variability of reference signals, thereby enhancing prediction reliability. Additionally, we introduce an auxiliary variance loss function to minimize inconsistencies across predictions, ensuring stable assessments regardless of the number of references used. Experiments on the NISQA datasets demonstrate that, with the same training time, our method achieves consistent predictions with a single reference during inference, resulting in a 100-fold reduction in computational time while maintaining high accuracy.

The process of transcribing speech plays a critical role in developing automatic speech recognition (ASR) systems. During this process, human transcribers listen to spoken utterances and convert them into accurately written texts. However, our initial investigations using telephone conversation datasets have shown that approximately 40% of the recorded utterances are challenging to transcribe and often disregarded by transcribers due to various factors, such as unintelligibility. As a result, transcribers waste significant time and effort listening to such utterances, significantly slowing down the overall transcribing speed. In this paper, we explore the potential of pretrained speech recognition models to model transcribers’ behavior in the speech transcribing process. By leveraging the knowledge and insights encoded in these models, we aim to filter out problematic utterances that transcribers typically disregard before they reach human transcribers, thus optimizing their time and effort. Through experiments on practical telephone datasets, our approach utilizing pretrained speech recognition models successfully reduces unwanted utterances by 50% while preserving the high quality of the collected dataset.

Current years have witnessed an increment in the number of research activities on improving the efficacy of multitasking algorithms for tackling challenging optimization problems. However, current approaches often present two potential problems. First, although tasks may have different characteristics, existing literature usually utilizes only one search operator for all of them. Second, while multitasking environments comprise tasks of varying difficulty, previous proposals treat them equally. This article proposes an algorithm named ensemble multifactorial evolution with biased skill-factor inheritance (EME-BI) for optimizing a large number of tasks simultaneously. In EME-BI, an effective parameter adaptation based on the knowledge transfer quality with biased skill-factor inheritance mechanism is designed to minimize negative transfer and allocate generated offspring to tasks that need resources. Besides, instead of using only one fixed search operator, EME-BI can automatically select the most appropriate one for each task at each evolutionary stage. Finally, the proposed algorithm is armed with a dynamically adjusted population size to promote exploitation. Empirical studies on various many-task benchmark problems and a real-world problem are conducted to verify the efficiency of EME-BI. The results portrayed that EME-BI achieves highly competitive performance compared to several state-of-the-art algorithms regarding the solution quality, convergence trend, and computation time. This proposal also won first prize at the CEC2021 Competition on Evolutionary Multitask Optimization, multitask single-objective optimization.

Recent years have witnessed an increment in the number of network components communicating through many network scenarios such as multi-layer and multi-domain, and it may result in a negative impact on resource utilization. An urgent requirement arises for routing the packets most efficiently and economically in large multi-domain networks. In tackling this complicated area, we consider the Inter-Domain Path Computation problem under Node-defined Domain Uniqueness Constraint (IDPC-NDU), which intends to find the minimum routing cost path between two nodes that traverses every domain at most once. Owing to the NP-Hard property of the IDPC-NDU, applying metaheuristic algorithms to solve this problem usually proves more efficient. In like manner, this paper proposes a Two-level Genetic Algorithm (PGA), where the first level determines the order of the visited domains, and the second level finds the shortest path between the two given nodes. Furthermore, to facilitate the finding process, a method to minimize the search space and a new chromosome encoding that would reduce the chromosome length to the number of domains are integrated into this proposed algorithm. To evaluate the efficiency of the proposal, experiments on various instances were conducted. The results demonstrated that PGA outperforms other algorithms and gives results no more than twice the optimal values.

Inspired by the ability of human multitasking, there is a growing body of literature in the computational intelligence community dedicated to solving multiple problems concurrently. One of the research areas that have been receiving much attention in this topic is evolutionary multitasking, which is able to solve multiple complicated optimization problems together, yielding a better result than solving them in isolation. However, researches on evolutionary multitasking mostly focus on solving a small number of problems together. In an attempt to improve evolutionary multitasking, we propose Many-Task Multi-Armed Bandit Evolutionary Algorithm (Ma2BEA), including a new structure for the evolutionary multitasking algorithm. It also adopts a well-proven result of Multi-Armed Bandit (MAB) as the method to control the adaptive knowledge exchange between different tasks. In particular, the action of selecting which task to perform inter-task crossover is learned and decided by the designed MAB agent. We verify that Ma2BEA correctly learned the underlying relationship between tasks using the simple 10-task benchmarks. Besides, Ma2BEA is compared with other evolutionary many-tasking algorithms that have recently been proposed using the Single-Objective Many-task benchmark from the WCCI 2020 Competition on Evolutionary Multi-task Optimization. Empirical results show that Ma2BEA is competitive in terms of high solution quality and reasonable execution time.

Nowadays, connectivity among communication devices in networks has been playing a significant role, especially when the number of devices is increasing dramatically that requires network service providers to have a better architecture of management system. One of the popular approach is to divide those devices inside a network into different domains, in which the problem of minimizing path computation in general or Inter-Domain Path Computation under Domain Uniqueness constraint (IDPC-DU) problem in specific has received much attention from the research community. Since the IDPC-DU is NP-complete, an approximate approach is usually taken to tackle this problem when the dimensionality is high. Although Multifactorial Evolutionary Algorithm (MFEA) has emerged as an effective approximation algorithm to deal with various fields of problems, there are still some difficulties to apply directly MFEA to solve the IDPC-DU problem, i.e. different chromosomes may have different numbers of genes or to construct a feasible solution not violating the problem's constraint. Therefore, to overcome these limitations, MFEA algorithm with a new solution representation based on Priority-based Encoding is introduced. With the new representation of the solution, a chromosome consists of two parts: the first part encodes the priority of the vertex while the second part encodes information of edges in the solution. Besides, the paper also proposed a corresponding decoding method as well as novel crossover and mutation operators. Those evolutionary operators always produce valid solutions. For examining the efficiency of the proposed MFEA, experiments on a wide range of test sets of instances were implemented and the results pointed out the effectiveness of the proposed algorithm. Finally, the characteristics of the proposed algorithm are also indicated and carefully analyzed.