With the Transformer architecture achieving impressive results in the vision domain. It has become a current popular research to explore more potentials of Transformer mixed architectures and explore more suitable mixed combinations. In this paper, Transformer classification network is designed and explored by multi-task architecture search algorithm. A new paradigm for multi-task architecture is designed by combining convolution and Transformer. The designed search architecture can combine the respective advantages of convolution and Transformer and can obtain better performance. At the same time, corresponding knowledge-aware multi-task genetic operators are designed to generate offspring individuals. Inter-task and inter-experience knowledge-aware is utilised to facilitate evolutionary convergence. During the search process, the reference evaluation method is utilised to reduce the redundant computation and time during the search process. In the experimental section, the search results are compared with state-of-the-art architectures and search algorithms. The experimental results confirm the effectiveness and high generalisation of the searched architectures. The ablation experimental part proves the effectiveness of the proposed architectural paradigm, genetic operators and reference evaluation.

Multimodal multiobjective optimization problems (MMOPs) feature multiple Pareto sets (PSs) corresponding to a single Pareto front (PF). While most multimodal multiobjective evolutionary algorithms (MMEAs) focus on locating multiple global PSs, they tend to overlook local PSs that provide valuable decision diversity despite exhibiting slightly inferior objective values. In this paper, a bi-directional coevolutionary algorithm, termed MMEA-BDC, is proposed for MMOPs with local PSs, which integrates tractive and diffusive search mechanisms to simultaneously discover both global and local PSs. Specifically, the tractive mechanism preserves the individuals with the best fitness, which tend to be located around global PSs, thereby guiding the search toward the global PF. Meanwhile, the diffusive mechanism evaluates individual fitness based on an improved local convergence indicator, allowing individuals near local PSs to be identified and preserved, thereby promoting convergence toward local PF. These individuals are further refined through a niche-driven filter to retain those contributing to decision space diversity. Additionally, a dual-space neighbor replacement strategy is designed to comprehensively consider the crowding degree of individuals in both decision and objective spaces, effectively balancing population diversity across the two spaces. Experimental results on several benchmark suites of MMOPs demonstrate that MMEA-BDC effectively discovers both global and local PSs and achieves superior competitiveness compared to seven state-of-the-art MMEAs.

In the evolutionary computing community, the remarkable language-handling capabilities and reasoning power of large language models (LLMs) have significantly enhanced the functionality of evolutionary algorithms (EAs), enabling them to tackle optimization problems involving structured language or program code. Although this field is still in its early stages, its impressive potential has led to the development of various LLM-based EAs. To effectively evaluate the performance and practical applicability of these LLM-based EAs, benchmarks with real-world relevance are essential. In this paper, we focus on LLM-based recommender systems (RSs) and introduce a benchmark problem set, named RSBench, specifically designed to assess the performance of LLM-based EAs in recommendation prompt optimization. RSBench emphasizes session-based recommendations, aiming to discover a set of Pareto optimal prompts that guide the recommendation process, providing accurate, diverse, and fair recommendations. We develop three LLM-based EAs based on established EA frameworks and experimentally evaluate their performance using RSBench. Our study offers valuable insights into the application of EAs in LLM-based RSs. Additionally, we explore key components that may influence the overall performance of the RS, providing meaningful guidance for future research on the development of LLM-based EAs in RSs.

Evolutionary algorithms serve as a pivotal tool in addressing black-box problems, finding widespread applications across diverse academic disciplines and engineering domains. Despite their utility, these algorithms often confront challenges when navigating complex search spaces, impeding a comprehensive exploration of potential solutions. Solely depending on the algorithm’s exploration abilities falls short of fully harnessing the rich information contained within search spaces. To unlock the full potential of solution spaces, we introduce a deep learning method for the reconstruction of the search path. Specifically, discrete data sampled by evolutionary operators during the exploration process are collected, and a uniquely designed fully connected neural network is employed to reconstruct the exploration paths. The neural network’s robust fitting capability facilitates the transformation of initially discrete sampled information into a continuous form. By capitalizing on the reconstructed solution space information, the algorithm excels in identifying superior solutions. We refer to this method of deep learning-based search path reconstruction evolution strategy algorithm (DLES). The effectiveness of DLES is validated across multiple datasets, including CEC 2014, CEC 2018, CEC 2022 and BBOB. Experimental results, compared to several state-of-the-art algorithms, affirm the superiority of the DLES algorithm.

As medical images from multiple modalities provide complementary diagnostic information, multimodal medical image classification (MMIC) leverages their integration to enhance disease diagnosis. Genetic programming (GP), especially multitree GP, offers a flexible framework for evolving multimodal features and holds the potential for interpretability—an essential factor in medical image analysis, where it is crucial to understand the underlying decision-making process. However, current multitree GP methods remain underexplored in MMIC tasks. This article attempts to fill that gap by proposing a novel multitree representation method that enables automatic multimodal feature learning and fusion from different modalities. Based on this new representation, we extend the function and terminal sets to support various types of medical imaging modalities. Additionally, a slice selection factor is proposed to process 3D medical image modalities effectively. Moreover, to fully utilize both modality-specific features and fused multimodal features, fusion is performed at both the feature and decision levels. The accuracy of the proposed method is assessed on eight MMIC tasks, representing different medical scenarios. The new method is compared with 15 competitive algorithms. Experimental results demonstrate that the new method achieves better classification performance than most benchmark methods. Specifically, it achieves the highest accuracy on four out of the eight tasks, GAMMA-Binary, GAMMA-Multi, BraTS2018 and Derm7pt, with improvements of 3.38%, 2.31%, 1.36% and 4.11% respectively over the second-best method. Further analysis reveals that the evolved models show potential for interpretability, owing to their ability to effectively capture the distinctive features of each modality, which may aid in biomarker identification.

Neural Architecture Search (NAS) has emerged as a crucial method for automating the design of deep learning models. Despite its potential, NAS frequently requires substantial computational and hardware resources. To mitigate these challenges, transferable NAS (TNAS) has been introduced, leveraging prior NAS results to enhance performance on new tasks. However, existing methods largely focus on knowledge transfer within identical neural search spaces, overlooking the potential for cross-domain transferability. Motivated by this gap, we explore evolutionary TNAS across heterogeneous search spaces by learning common neural representations. In particular, we introduce a novel approach that encodes both operational and topological information of neural architectures into a unified sequence using a simple tokenizer. This sequence is then processed by a variational auto-encoder, with a Transformer-based encoder to capture rich neural representations and a decoder that reconstructs the original sequence. By utilizing these latent representations, we further establish an inter-domain mapping that acts as a bridge, enabling effective explicit solution transfer among diverse search spaces to enhance the evolutionary NAS process. To harness this capability, we develop an evolutionary sequential transfer optimization approach that transfers knowledge during population initialization, providing both flexibility and adaptability. To the best of our knowledge, this work serves as the first attempt in the literature exploring evolutionary TNAS across diverse spaces. Moreover, we demonstrate the utility of our method through comprehensive empirical studies using different architecture spaces, including NAS-Bench-101, NAS-Bench-201, and the DARTS search space. Our results show that the proposed method significantly enhances the adaptability and performance of NAS across varied domains.

Optimization of the enzyme turnover rate kcat in enzyme-constrained metabolic models, such as S. cerevisiae’s ecYeastGEM, is essential to improve the accuracy of cellular metabolism prediction. This paper constructs a constrained multiobjective optimization problem centered on optimizing kcat, aiming to minimize model calculation error and energy consumption while considering the maximization of specific growth rate. The problem has up to 4261 dimensions and is subject to several stringent constraints, and the existence of multiple feasible solutions for kcat significantly increases the difficulty of the optimization. To address these challenges, we propose the indicator guided bi-surrogate adaptive multimodel multiobjective evolutionary algorithm (IGSMMOEA). The algorithm approximates the constrained optimization landscape via the surrogate model, combines the indicator guided mechanism to achieve adaptive model updating, and ensures that the optimized solution satisfies the enzyme constraints through secondary validation. Experimental results demonstrate that IGSMMOEA exhibits strong generality and efficiently optimizes kcat. The optimized solutions not only surpass existing methods in diversity and convergence but also strictly adhere to biochemical constraints. The parameters derived from the algorithm are more robust, offering valuable support for metabolic engineering design.

Genetic programming-based feature construction has achieved significant success in recent years as an automated machine learning technique to enhance learning performance. However, overfitting remains a challenge that limits its broader applicability. To improve generalization, we prove that vicinal risk, estimated through noise perturbation or mixup-based data augmentation, is bounded by the sum of empirical risk and a regularization termb–either finite difference or the vicinal Jensen gap. Leveraging this decomposition, we propose an evolutionary feature construction framework that jointly optimizes empirical risk and the vicinal Jensen gap to control overfitting. Since datasets may vary in noise levels, we develop a noise estimation strategy to dynamically adjust regularization strength. Furthermore, to mitigate manifold intrusionb–where data augmentation may generate unrealistic samples that fall outside the data manifoldb–we propose a manifold intrusion detection mechanism. Experimental results on 58 datasets demonstrate the effectiveness of Jensen gap minimization compared to other complexity measures. Comparisons with 15 machine learning algorithms further indicate that genetic programming with the proposed overfitting control strategy achieves superior performance.

Nature-inspired methods, such as evolutionary computing and lifetime learning, have shown great promise in advancing autonomous robot design. However, the integration of evolution and learning for the joint optimization of robot morphologies and controllers remains underexplored, particularly in dynamic environments. This paper addresses this gap by investigating the effectiveness of Lamarckian inheritance (a mechanism that allows learned traits to be encoded into the genotype and passed to offspring) in improving robot evolution in nonstationary environments. We compare a Lamarckian system with a traditional Darwinian system, where learned traits are not inherited. Using simulated modular robots in six distinct environmental setups, we analyze the fitness progression, learning ability, and parent-offspring similarity within both systems. Our results demonstrate that the Lamarckian system consistently outperforms the Darwinian system, achieving up to 33% higher fitness in the most challenging conditions. The Lamarckian system also recovers more quickly from environmental changes, showing immediate fitness gains when shifting to complex terrains, whereas the Darwinian system adapts more slowly. Real-world tests validate the robustness of the Lamarckian approach, as the top-performing robots evolved in the most challenging environment exhibit the smallest reality gap. These findings highlight the potential of Lamarckian inheritance as a powerful tool for engineering adaptive robotic systems capable of maintaining high performance in dynamic environments.

When solving expensive constrained multiobjective optimization problems by surrogate-assisted evolutionary algorithms, the evolutionary sampling method is utilized to select valuable candidate solutions for expensive evaluations, therefore playing a vital role. Most existing sampling methods always prioritize feasible solutions over infeasible solutions during the whole evolutionary process. However, sampling infeasible solutions in some cases can not only provide search directions but also enhance diversity. To tackle this issue, a novel competitive multitasking sampling-based framework is proposed. Specifically, three complementary sampling subtasks are constructed: constraint-prioritized subtask, objective-constraint equivalence subtask, and objective-prioritized subtask. The main task is to find the constrained Pareto front. Different sampling subtasks exhibit different performances on various expensive problems. During each iteration, one of the sampling subtasks is selected by hierarchical deep reinforcement learning technique. This technique can effectively learn the relationship between the environmental states and the performances of different subtasks. Based on the selected subtask, the candidate points are generated via surrogate-assisted evolution. Subsequently, after these points undergo expensive evaluations, they can be regarded as knowledge and transferred to both the subtasks and the main task. Experimental results on three benchmark test suites and a real-world application have demonstrated the framework’s superiority over other state-of-the-art competitors.