Evolutionary Neural Architecture Search (ENAS) algorithms attract great attention since they can automatically search for appropriate network architectures for a given task. However, most ENAS algorithms suffer from a prohibitive computational burden. Moreover, some of these approaches directly use performance predictors for evaluations, which may introduce inaccurate assessments and harm the evolution. To overcome these shortcomings, we propose an efficient ENAS algorithm named EPPGA. EPPGA employs a predictor to pre-select potentially high-performing offspring, enhancing the performance and accelerating the evolution. As the offspring will be further accurately evaluated, even potentially inaccurate predictions will not adversely affect the evolution. Furthermore, a weight inheritance method is suggested to accelerate the evaluation, and new genetic operations are developed to produce offspring that share a substantial proportion of beneficial genetic materials with one parent, improving the performance predictor's effectiveness and promoting weight inheritance. Finally, a new efficient backbone block structure is designed to facilitate the search for lightweight networks. The experimental results demonstrate that EPPGA is a highly competitive algorithm on three benchmarks in terms of accuracy, model size, and computational cost, reveal the superiority of the proposed block structure, and confirm the effectiveness of the proposed performance predictor and weight inheritance method.

To enhance deep learning-based automated interictal epileptiform discharge (IED) detection, this study proposes a multimodal method, vEpiNet, that leverages video and electroencephalogram (EEG) data. Datasets comprise 24 931 IED (from 484 patients) and 166 094 non-IED 4-second video-EEG segments. The video data is processed by the proposed patient detection method, with frame difference and Simple Keypoints (SKPS) capturing patients' movements. EEG data is processed with EfficientNetV2. The video and EEG features are fused via a multilayer perceptron. We developed a comparative model, termed nEpiNet, to test the effectiveness of the video feature in vEpiNet. The 10-fold cross-validation was used for testing. The 10-fold cross-validation showed high areas under the receiver operating characteristic curve (AUROC) in both models, with a slightly superior AUROC (0.9902) in vEpiNet compared to nEpiNet (0.9878). Moreover, to test the model performance in real-world scenarios, we set a prospective test dataset, containing 215 h of raw video-EEG data from 50 patients. The result shows that the vEpiNet achieves an area under the precision-recall curve (AUPRC) of 0.8623, surpassing nEpiNet's 0.8316. Incorporating video data raises precision from 70% (95% CI, 69.8%-70.2%) to 76.6% (95% CI, 74.9%-78.2%) at 80% sensitivity and reduces false positives by nearly a third, with vEpiNet processing one-hour video-EEG data in 5.7 min on average. Our findings indicate that video data can significantly improve the performance and precision of IED detection, especially in prospective real clinic testing. It suggests that vEpiNet is a clinically viable and effective tool for IED analysis in real-world applications.

Deep convolutional neural networks (DCNNs) have achieved surpassing success in the field of computer vision, and a number of elaborately designed networks refresh the performance records in benchmark datasets. Recently, evolutionary neural architecture search (ENAS) has become an emerging area, which could employ an evolutionary computation (EC) technique to automatically construct promising network architectures without human intervention. However, existing algorithms still have limitations: most standard EC approaches cannot process flexible-length architecture representations directly, and most fitness evaluation processes suffer from the exhibitive computational cost and the unreliable prediction results. To overcome these shortcomings, we propose an efficient particle swarm optimization (PSO)-based neural architecture search algorithm to search for appropriate dense blocks on image classification tasks, named EAEPSO. EAEPSO addresses the first limitation by designing an autoencoder to encode variable-length network representations as fixed-length latent vectors, which converts the original search space to a latent space that can facilitate the downstream search. Besides, an efficient and effective hierarchical fitness evaluation method is designed to guide the search process to address the second limitation. The experimental results show that EAEPSO is a very competitive ENAS algorithm that achieves an error rate of 2.74% on CIFAR-10 and 16.17% on CIFAR-100, and reduces the computational cost from hundreds or thousands of GPU-days to only 2.2 and 4 GPU-days, respectively. Further analyses investigate the reduced training data’s effect and confirm the effectiveness of both the proposed autoencoder and the proposed hierarchical fitness evaluation method.

Neural architecture search (NAS) is becoming increasingly popular for its ability to automatically search for an appropriate network architecture, avoiding laborious manual designing processes, and potentially introducing novel structures. However, many NAS methods suffer from heavy computational consumption. One-shot NAS alleviates this issue by training a big supernet and allowing all the candidates to inherit weights from the supernet, avoiding training from scratch. However, the performance evaluations in one-shot methods might not always be reliable due to the weight co-adaption issue inside the supernet. This paper proposes a super-net fine-tuning strategy to allow the supernet's weights to adapt to the new focused search region along with the evolutionary process. Furthermore, a new genetic algorithm-based search method is designed to offer an effective path-sampling strategy in the search region and provide a new population generation method to preclude unfair fitness comparisons between different populations. The experimental results demonstrate the proposed method achieves promising results compared with 32 peer competitors in terms of the algorithm's computational cost and the searched architecture's performance. Specifically, the proposed method achieves classification error rates of 2.50% and 17.07% within only 0.50 and 0.92 GPU-days on CIFAR10 and CIFAR100, respectively.