Machine anomalous sound detection is a useful technique for various applications, but it often suffers from poor generalization due to the challenges of data collection and complex acoustic environment. To address this issue, we propose a robust machine anomalous sound detection model that leverages self-supervised pre-trained models on large-scale speech data. Specifically, we assign different weights to the features from different layers of the pre-trained model and then use the working condition as the label for self-supervised classification fine-tuning. Moreover, we introduce a data augmentation method that simulates different operating states of the machine to enrich the dataset. Furthermore, we devise a transformer pooling method that fuses the features of different segments. Experiments on the DCASE2023 dataset show that our proposed method outperforms the commonly used reconstruction-based autoencoder and classification-based convolutional network by a large margin, demonstrating the effectiveness of large-scale pre-training for enhancing the generalization and robustness of machine anomalous sound detection. In Task2 of DCASE2023, we achieve 2nd place with these methods.

Automatic detection of machine anomaly remains challenging for machine learning. We believe the capability of generative adversarial network (GAN) suits the need of machine audio anomaly detection, yet rarely has this been investigated by previous work. In this paper, we propose AEGAN-AD, a totally unsupervised approach in which the generator (also an autoencoder) is trained to reconstruct input spectrograms. It is pointed out that the denoising nature of reconstruction deprecates its capacity. Thus, the discriminator is redesigned to aid the generator during both training stage and detection stage. The performance of AEGAN-AD on the dataset of DCASE 2022 Challenge TASK 2 demonstrates the state-of-the-art result on five machine types. A novel anomaly localization method is also investigated. Source code available at: www.github.com/jianganbai/AEGAN-AD

Content-based publish/subscribe systems are an effective paradigm for implementing on-demand event distribution.Each event needs to be matched against subscriptions to identify the target subscribers.To improve the matching performance, many novel data structures have been proposed.However, the predicates included in subscriptions are handled the same way in most existing data structures, which is not efficient given the matching probability of predicates.In this paper, we propose a parallel ensemble matching framework called PEM, which uses multiple algorithms with complementary behavior on predicate matching probabilities.To achieve the performance balance of parallel matching, we design an elastic subscription classification method.We implement a prototype of PEM based on two existing algorithms.The experiment results show that PEM improves the matching performance by 43%.

The content-based publish/subscribe system is an effective paradigm for implementing on-demand event distribution. Each event needs to be matched against subscriptions to identify the target subscribers. To improve the matching performance, many novel data structures have been proposed. However, the predicates contained in subscriptions are handled the same way in most existing data structures, without considering their differences in matching probability. In this paper, we propose the concept of parallel ensemble matching (PEM) based on subscription partitioning. The basic idea is that we have the right algorithm handling the right subscriptions at the right time. First of all, we design a PEM framework by classifying subscriptions according to their matching probabilities and use the proper algorithms to process each subscription category. Furthermore, to deal with high-dimensional subscriptions, we propose a fine-grained PEM (fgPEM) that exploits matching algorithms with complementary behaviors by partitioning subscriptions into sub-subscriptions. We implement the prototype of PEM and fgPEM based on two existing algorithms. The experiment results show that PEM improves the matching performance by 43%. On the basis of PEM, fgPEM further improves the performance by 31%.

Traditional Chinese medicine has been used to treat and prevent infectious diseases for thousands of years, and has accumulated a large number of effective prescriptions. Deep learning methods provide powerful applications in calculating interactions between drugs and targets. In this study, we try to use the method of deep learning to reposition molecules of Chinese medicines (CMs) and the targets of syndrome coronavirus 2 (SARS-CoV-2). A deep convolution neural network with residual module (DCNN-Res) is constructed and trained on KIBA dataset. The accuracy of predicting the binding affinity of drugtarget pairs is 85.33%. By ranking binding affinity scores of 433 molecules in 35 CMs to 6 targets of SARS-Cov-2, DCNN-Res recommends 30 possible repositioning molecules. The consistency between our result and the latest research is 0.827. The molecules in Gancao and Huangqin have a strong binding affinity to targets of SARS-CoV-2, which is also consistent with the latest research.

Smart retail stores open new possibilities for enabling a variety of physical analytics, such as users' shopping trajectories and preferences for certain items. This paper aims to excavate three kinds of relations in physical stores, i.e. user-item, user-user and item-item, which provide abundant information for enhancing users' shopping experiences and boosting retailers' sales. We present ShopEye, a hybrid RFID and smartwatch system to delve into these relations in an implicit and non-intrusive manner. The intuition is that inertial sensors embedded in smartwatches and RFID tags attached to items can capture the user behaviors and the item motions, respectively. ShopEye first pairs users with corresponding items according to correlations between inertial signals and RFID signals, and then incorporates these pairs with the motion behaviors of users to further profile user-user and item-item relations. We have tested the system extensively in our lab environment which mimics the real retail store. Experimental results demonstrate the effectiveness and robustness of ShopEye in excavating these relations.

Smart wearable devices have become pervasive as they are portable and intelligent. The popular method to interact with it is touch-screen, which is error-prone and cumbersome due to its limited size. There are a few innovative works designing a virtual dial plate on the hand back, which need special-purpose sensors or microphones which may suffer from privacy leak. We propose MType, a system only employs sensors already built in the commercial-off-the-shelf (COTS) device with a magnetic ring to expand the interaction space between users and wearable devices. The core idea is to leverage the gravity sensor, linear accelerometer and magnetometer embedded in the standard smartwatch to detect gestures, capture input events and locate keystrokes on the opisthenar and palm. Besides, MType designs a runtime adaptation mechanism to handle the cold start problem and adapt to the variations over the time of usage. We implement MType on the COTS smartwatch and our extensive experiments in different scenarios show that the average accuracy of keystroke localization can reach 93% with a small size initial training set (3 samples for each key) at a low sampling rate (51Hz). Furthermore, when turning on the runtime adaptation mechanism and enlarging the training set, the accuracy can achieve 98%.

Online bookstores are capable of capturing readers preferences by analyzing click logs and transaction records, while physical bookstores or libraries still lack effective methods to gather reader behavioral data. Fortunately, the widespread use of mobile wearable devices and RFID technology opens up new possibilities for uncovering in-store experience. In this paper, we propose ReaderTrack, a system that integrates smartwatch and RFID to excavate interactions between readers and books. We first leverage inertial sensors of smartwatch and backscatter signals of RFID tags to infer reader behaviors and book motions, respectively. Then we associate readers with their corresponding books according to previously inferred behaviors and motions. We implement ReaderTrack with COTS devices and evaluate it extensively in our lab environment which mimics a typical reading room. Experimental results show the effectiveness and robustness of ReaderTrack in reader-book interaction reasoning.

2018 IEEE International Conference on Pervasive Computing and Communications (PerCom), Athens, Greece, March 19-23, 2018