While normalizations aim to fix the exploding and vanishing gradient problem in deep neural networks, they have drawbacks in speed or accuracy because of their dependency on the data set statistics. This work is a comprehensive study of a novel method based on unitary synaptic weights derived from Lie Group to construct intrinsically stable neural systems. Here we show that unitary convolutional neural networks deliver up to 32% faster inference speeds while maintaining competitive prediction accuracy. Unlike prior arts restricted to square synaptic weights, we expand the unitary networks to weights of any size and dimension.

The monitoring of rotating machinery is an essential task in today's production processes. Currently, several machine learning and deep learning-based modules have achieved excellent results in fault detection and diagnosis. Nevertheless, to further increase user adoption and diffusion of such technologies, users and human experts must be provided with explanations and insights by the modules. Another issue is related, in most cases, with the unavailability of labeled historical data that makes the use of supervised models unfeasible. Therefore, a new approach for fault detection and diagnosis in rotating machinery is here proposed. The methodology consists of three parts: feature extraction, fault detection and fault diagnosis. In the first part, the vibration features in the time and frequency domains are extracted. Secondly, in the fault detection, the presence of fault is verified in an unsupervised manner based on anomaly detection algorithms. The modularity of the methodology allows different algorithms to be implemented. Finally, in fault diagnosis, Shapley Additive Explanations (SHAP), a technique to interpret black-box models, is used. Through the feature importance ranking obtained by the model explainability, the fault diagnosis is performed. Two tools for diagnosis are proposed, namely: unsupervised classification and root cause analysis. The effectiveness of the proposed approach is shown on three datasets containing different mechanical faults in rotating machinery. The study also presents a comparison between models used in machine learning explainability: SHAP and Local Depth-based Feature Importance for the Isolation Forest (Local- DIFFI). Lastly, an analysis of several state-of-art anomaly detection algorithms in rotating machinery is included.

We demonstrate the use of an extensive deep neural network to localize instances of objects in images. The EDNN is naturally able to accurately perform multi-class counting using only ground truth count values as labels. Without providing any conceptual information, object annotations, or pixel segmentation information, the neural network is able to formulate its own conceptual representation of the items in the image. Using images labelled with only the counts of the objects present, the structure of the extensive deep neural network can be exploited to perform localization of the objects within the visual field. We demonstrate that a trained EDNN can be used to count objects in images much larger than those on which it was trained. In order to demonstrate our technique, we introduce seven new datasets: five progressively harder MNIST digit-counting data sets, and two data sets of 3d-rendered rubber ducks in various situations. On most of these datasets, the EDNN achieves greater than 99% test set accuracy in counting objects.

Sentiment analysis is known as one of the most crucial tasks in the field of natural language processing and Convolutional Neural Network (CNN) is one of those prominent models that is commonly used for this aim. Although convolutional neural networks have obtained remarkable results in recent years, they are still confronted with some limitations. Firstly, they consider that all words in a sentence have equal contributions in the sentence meaning representation and are not able to extract informative words. Secondly, they require a large number of training data to obtain considerable results while they have many parameters that must be accurately adjusted. To this end, a convolutional neural network integrated with a hierarchical attention layer is proposed which is able to extract informative words and assign them higher weight. Moreover, the effect of transfer learning that transfers knowledge learned in the source domain to the target domain with the aim of improving the performance is also explored. Based on the empirical results, the proposed model not only has higher classification accuracy and can extract informative words but also applying incremental transfer learning can significantly enhance the classification performance.

Realistic use of neural networks often requires adhering to multiple constraints on latency, energy and memory among others. A popular approach to find fitting networks is through constrained Neural Architecture Search (NAS), however, previous methods enforce the constraint only softly. Therefore, the resulting networks do not exactly adhere to the resource constraint and their accuracy is harmed. In this work we resolve this by introducing Hard Constrained diffeRentiable NAS (HardCoRe-NAS), that is based on an accurate formulation of the expected resource requirement and a scalable search method that satisfies the hard constraint throughout the search. Our experiments show that HardCoRe-NAS generates state-of-the-art architectures, surpassing other NAS methods, while strictly satisfying the hard resource constraints without any tuning required.

Suffering from the semantic insufficiency and domain-shift problems, most of existing state-of-the-art methods fail to achieve satisfactory results for Zero-Shot Learning (ZSL). In order to alleviate these problems, we propose a novel generative ZSL method to learn more generalized features from multi-knowledge with continuously generated new semantics in semantic-to-visual embedding. In our approach, the proposed Multi-Knowledge Fusion Network (MKFNet) takes different semantic features from multi-knowledge as input, which enables more relevant semantic features to be trained for semantic-to-visual embedding, and finally generates more generalized visual features by adaptively fusing visual features from different knowledge domain. The proposed New Feature Generator (NFG) with adaptive genetic strategy is used to enrich semantic information on the one hand, and on the other hand it greatly improves the intersection of visual feature generated by MKFNet and unseen visual faetures. Empirically, we show that our approach can achieve significantly better performance compared to existing state-of-the-art methods on a large number of benchmarks for several ZSL tasks, including traditional ZSL, generalized ZSL and zero-shot retrieval.

Aided by recent advances in Deep Learning, Image Caption Generation has seen tremendous progress over the last few years. Most methods use transfer learning to extract visual information, in the form of image features, with the help of pre-trained Convolutional Neural Network models followed by transformation of the visual information using a Caption Generator module to generate the output sentences. Different methods have used different Convolutional Neural Network Architectures and, to the best of our knowledge, there is no systematic study which compares the relative efficacy of different Convolutional Neural Network architectures for extracting the visual information. In this work, we have evaluated 17 different Convolutional Neural Networks on two popular Image Caption Generation frameworks: the first based on Neural Image Caption (NIC) generation model and the second based on Soft-Attention framework. We observe that model complexity of Convolutional Neural Network, as measured by number of parameters, and the accuracy of the model on Object Recognition task does not necessarily co-relate with its efficacy on feature extraction for Image Caption Generation task.

Deep Neural Networks have achieved unprecedented success in the field of face recognition such that any individual can crawl the data of others from the Internet without their explicit permission for the purpose of training high-precision face recognition models, creating a serious violation of privacy. Recently, a well-known system named Fawkes (published in USENIX Security 2020) claimed this privacy threat can be neutralized by uploading cloaked user images instead of their original images. In this paper, we present Oriole, a system that combines the advantages of data poisoning attacks and evasion attacks, to thwart the protection offered by Fawkes, by training the attacker face recognition model with multi-cloaked images generated by Oriole. Consequently, the face recognition accuracy of the attack model is maintained and the weaknesses of Fawkes are revealed. Experimental results show that our proposed Oriole system is able to effectively interfere with the performance of the Fawkes system to achieve promising attacking results. Our ablation study highlights multiple principal factors that affect the performance of the Oriole system, including the DSSIM perturbation budget, the ratio of leaked clean user images, and the numbers of multi-cloaks for each uncloaked image. We also identify and discuss at length the vulnerabilities of Fawkes. We hope that the new methodology presented in this paper will inform the security community of a need to design more robust privacy-preserving deep learning models.

Medical time-series datasets have unique characteristics that make prediction tasks challenging. Most notably, patient trajectories often contain longitudinal variations in their input-output relationships, generally referred to as temporal conditional shift. Designing sequence models capable of adapting to such time-varying distributions remains a prevailing problem. To address this we present Model-Attentive Ensemble learning for Sequence modeling (MAES). MAES is a mixture of time-series experts which leverages an attention-based gating mechanism to specialize the experts on different sequence dynamics and adaptively weight their predictions. We demonstrate that MAES significantly out-performs popular sequence models on datasets subject to temporal shift.

Following identification of a target protein, hit identification, which finds small organic molecules that bind to the target, is an important first step of a structure-based drug design project. In this study, we demonstrate a target-specific drug design method that can autonomously generate a series of target-favorable compounds. This method utilizes the seq2seq model based on a deep learning algorithm and a water pharmacophore. Water pharmacophore models are used to screen compounds that are favorable to a given target in a large compound database, and seq2seq compound generators are used to train the screened compounds and generate entirely new compounds based on the training model. Our method was tested through binding energy calculation studies of six pharmaceutically relevant targets in the directory of useful decoys (DUD) set with docking.