Bioacoustic event detection is a demanding endeavor involving recognizing and classifying the sounds animals make in their natural habitats. Traditional supervised learning requires a large amount of labeled data, which are hard to come by in bioacoustics. This paper presents a few-shot learning (FSL) method incorporating transductive inference and data augmentation to address the issues of too few labeled events and small volumes of recordings. Here, transductive inference iteratively alters class prototypes and feature extractors to seize essential patterns, whereas data augmentation applies SpecAugment on Mel spectrogram features to augment training data. The proposed approach is evaluated by using the Detecting and Classifying Acoustic Scenes and Events (DCASE) 2022 and 2021 datasets. Extensive experimental results demonstrate that all components of the proposed method achieve significant F-score improvements of 27% and 10%, for the DCASE-2022 and DCASE-2021 datasets, respectively, compared to recent advanced approaches. Moreover, our method is helpful in FSL tasks because it effectively adapts to sounds from various animal species, recordings, and durations.

Wireless Sensor Networks (WSNs) are widely used for real-time monitoring in industrial automation, environmental sensing, and healthcare. However, sensor faults pose significant challenges, potentially leading to compromised data integrity and reduced system reliability. Traditional fault detection methods often rely on large labeled datasets, which are scarce in real-world scenarios. This paper proposes a novel data-efficient framework that integrates unsupervised learning for fault detection with few-shot learning for fault classification, addressing the challenge of a limited number of fault data in WSNs. For fault detection, we introduce the LSTM Autoencoder with a Temporal Attention Mechanism (LATAM) model, trained on normal data, that identifies anomalies based on reconstruction errors and a dynamic thresholding technique. For fault classification, we employ a few-shot learning approach using Convolutional Kernel Transform (MiniROCKET) for efficient feature extraction, and Prototypical Networks to classify faults with a minimal number of labeled samples. Due to the lack of comprehensive public fault datasets, the proposed framework was evaluated on a dataset combining real-world healthy temperature sensor readings with synthetically generated faults. Experimental results demonstrate that the LATAM model achieves high accuracy in fault detection, with an F-score of 98.90%. The few-shot classification approach performs effectively in low-data scenarios, achieving F-scores of 94.18% in a 15-shot setting and 84.75% in a one-shot setting. Compared to the traditional 1D CNN, LSTM, and Matching Networks methods, the proposed framework exhibits superior performance, particularly in data-scarce environments. By eliminating reliance on large labeled datasets and pre-trained models, this approach effectively addresses the challenges posed by data scarcity in WSNs.

This paper introduces a few-shot transfer learning approach to fault classification in wireless sensor networks (WSNs) with a minimal number of fault samples. WSNs are susceptible to various faults, such as drift, stuck, bias, and spike faults, erratic behavior, and data loss, which can compromise system reliability. Conventional deep learning fault diagnosis methods have achieved promising results, however, the majority of these approaches require a substantial amount of labeled training data, which is not available in real-world scenarios. To address this, we propose a novel method that combines convolutional neural networks (ResNet-18, VGG-16, and MobileNetV2 backbone architectures) with prototypical networks for fault diagnosis in WSNs with only a few fault samples. By transforming time-series data into Gramian Angular Field images, our approach leverages pre-trained deep learning models to extract feature-rich embeddings. These embeddings are then classified using a prototypical network, which enhances the system’s ability to diagnose faults even with a limited amount of labeled data. The proposed model is lightweight and deployable on Internet of Things (IoT) devices, ensuring efficient fault classification with minimal computational resources. Experimental results demonstrate the model’s high accuracy and robustness across various fault types, highlighting its potential for scalable and adaptive IoT applications.

Few-shot (FS) sound event detection (SED) is the process of identifying and recognizing specific sounds or events within an audio recording, specifically in the field of bioacoustics. This task is especially challenging due to the substantial unpredictability, complexity of the sounds, and the limited amount of labeled training data available. We recommend a transudative learning technique with data augmentation in response to the recent success of transudative inference (TI) in the field of computer vision. It aims to make predictions about new, unseen data by maximizing the mutual information between features of the labeled training data and the features of the unlabeled test data. We use the detection and classification of acoustic scenes and events (DCASE)-2021 and DCASE-2022 official datasets from the DCASE to do FS bioacoustics event detection experiments. Research findings illustrate that the recommended approach using TI along with spectrogram augmentation improves the performance of the FS bioacoustics event detection system. The proposed method improves the F-score by 0.10 and 0.16 on DCASE-2021 and DCASE-2022, respectively, compared with the average baseline and state-of-the-art methods.

Autoencoder-based unsupervised intrusion detection systems (IDS) offer a promising approach for identifying zero-day attacks without requiring labeled malicious data. However, residual autoencoder (RAE) frameworks remain limited by static decision thresholds, susceptibility to concept drift (CD), limited adversarial robustness, and restricted adaptability to emerging attack patterns. To address these limitations, we propose the Attention-Boosted Streaming Residual Autoencoder (ASRAE), a unified framework that integrates attention-enhanced representation learning, online CD detection with adaptive thresholding, benign-only adversarial training, post-identification few-shot meta-learning, and explainability through integrated gradients (IG). Experiments on UNSW-NB15, CIC-IDS2017, and CSE-CIC-IDS2018 show that ASRAE outperforms the evaluated unsupervised baselines, achieving ROC-AUC values of 0.952, 0.940, and 0.931, respectively. Under the controlled simulated concept-drift setting, ASRAE maintains a ROC-AUC above 0.91, while adversarial evaluation using Fast Gradient Sign Method (FGSM) and Projected Gradient Descent (PGD) demonstrates improved robustness to feature-space perturbations. The few-shot adaptation stage further enables adaptation to held-out attack families using only a small number of labeled samples while largely preserving benign reconstruction performance. ASRAE requires an average inference time of 9.2 ms per sample in the evaluated experimental environment. Overall, the results demonstrate the effectiveness of ASRAE across the benchmark, simulated-drift, adversarial, and few-shot adaptation settings.