Stay updated with breaking news from Eature extraction. Get real-time updates on events, politics, business, and more. Visit us for reliable news and exclusive interviews.
Recent studies have demonstrated that backdoor attacks can cause a significant security threat to federated learning. Existing defense methods mainly focus on detecting or eliminating the backdoor patterns after the model is backdoored. However, these methods either cause model performance degradation or heavily rely on impractical assumptions, such as labeled clean data, which exhibit limited effectiveness in federated learning. To this end, we propose FLPurifier, a novel backdoor defense method in federated learning that can effectively purify the possible backdoor attributes before federated aggregation. Specifically, FLPurifier splits a complete model into a feature extractor and classifier, in which the extractor is trained in a decoupled contrastive manner to break the strong correlation between trigger features and the target label. Compared with existing backdoor mitigation methods, FLPurifier doesn’t rely on impractical assumptions since it can effectively purify the backdoor effects in the training process rather than an already trained model. Moreover, to decrease the negative impact of backdoored classifiers and improve global model accuracy, we further design an adaptive classifier aggregation strategy to dynamically adjust the weight coefficients. Extensive experimental evaluations on six benchmark datasets demonstrate that FLPurifier is effective against known backdoor attacks in federated learning with negligible performance degradation and outperforms the state-of-the-art defense methods.
As a newly emerging task, audio-visual question answering (AVQA) has attracted research attention. Compared with traditional single-modality (e.g., audio or visual) QA tasks, it poses new challenges due to the higher complexity of feature extraction and fusion brought by the multimodal inputs. First, AVQA requires more comprehensive understanding of the scene which involves both audio and visual information; Second, in the presence of more information, feature extraction has to be better connected with a given question; Third, features from different modalities need to be sufficiently correlated and fused. To address this situation, this work proposes a novel framework for multimodal question answering task. It characterises an audiovisual scene at both global and local levels, and within each level, the features from different modalities are well fused. Furthermore, the given question is utilised to guide not only the feature extraction at the local level but also the final fusion of global and local features to predict the answer. Our framework provides a new perspective for audio-visual scene understanding through focusing on both general and specific representations as well as aggregating multimodalities by prioritizing question-related information. As experimentally demonstrated, our method significantly improves the existing audio-visual question answering performance, with the averaged absolute gain of 3.3% and 3.1% on MUSIC-AVQA and AVQA datasets, respectively. Moreover, the ablation study verifies the necessity and effectiveness of our design. Our code will be publicly released.
Hyperspectral image segmentation is an emerging area with numerous applications, including agriculture, forestry, environment monitoring, and remote sensing. This paper proposes a new neural architecture search algorithm, named AdaptorNAS, for hyperspectral image segmentation. AdaptorNAS aims to design the optimum decoder for any given encoder. In our approach, the search space of AdaptorNAS is a large deep neural network (DNN), and the optimal decoder is derived by pruning the large DNN via a perturbation-based pruning strategy. Verified on three popular encoders, i.e., ResNet-34, MobileNet-V2, and EfficientNet-B2, AdaptorNAS can design high-speed decoders that are significantly better than six common hand-crafted decoders. Additionally, with the EfficientNet-B2 encoder, AdaptorNAS (mIoU of 92.47% and mDice of 95.15%) outperforms the state-of-the-art NAS algorithms and hand-crafted network architectures on the hyperspectral image segmentation task. We also introduce a new hyperspectral image dataset of 4,625 images for objective evaluation in hyperspectral image segmentation research.
Freezing of Gait (FoG) is a common symptom of Parkinson’s disease (PD), manifesting as a brief, episodic absence, or marked reduction in walking, despite a patient’s intention to move. Clinical assessment of FoG events from manual observations by experts is both time-consuming and highly subjective. Therefore, machine learning-based FoG identification methods would be desirable. In this article, we address this task as a fine-grained human action recognition problem based on vision inputs. A novel deep learning architecture, namely, higher order polynomial transformer (HP-Transformer), is proposed to incorporate pose and appearance feature sequences to formulate fine-grained FoG patterns. In particular, a higher order self-attention mechanism is proposed based on higher order polynomials. To this end, linear, bilinear, and trilinear transformers are formulated in pursuit of discriminative fine-grained representations. These representations are treated as multiple streams and further fused by a cross-order fusion strategy for FoG detection. Comprehensive experiments on a large in-house dataset collected during clinical assessments demonstrate the effectiveness of the proposed method, and an area under the receiver operating characteristic (ROC) curve (AUC) of 0.92 is achieved for detecting FoG.
We address privacy and latency issues in edge-cloud computing environments where the neural network training is centralized. This paper considers the scenario where the edge devices are the only data sources for the deep learning model to be trained on the central server. Improper access to the massive amounts of data generated by edge devices could lead to privacy concerns. As a result, existing solutions for preserving privacy and reducing network latency in the edge environment rely on auxiliary datasets with no privacy risks or pre-trained models to build the client side feature extractor. However, finding auxiliary datasets or pre-trained models is not always guaranteed and may be challenging. To bridge this gap and eliminate the reliance on auxiliary datasets or pre-trained models of existing solutions, this paper presents DeepGuess, a privacy-preserving and latency-aware deep-learning framework. DeepGuess introduces a new learning mechanism enabled by the AutoEncoder architecture: inductive learning. With inductive learning, sensitive data stays on devices and is not explicitly sent to the central server to engage in back-propagations. To further enhance privacy, we propose a new local differential privacy algorithm that allows edge devices to apply random noise to features extracted from their sensitive data before being transferred to the non-trusted central server. The experimental evaluation of DeepGuess with various datasets and in a real-world scenario shows that our solution achieves comparable or even higher accuracy than existing solutions while reducing data transfer over the network by more than 50%.
Cross-domain person re-identification (Re-ID) is a challenging and important task in monitoring safety and procedure compliance of industrial work places. In this paper, a novel method is proposed to reduce background induced domain shift for adaptive person Re-ID. Specifically, a foreground-background joint clustering module is proposed to extract discriminative foreground and background features and an attention-based feature disentanglement module is designed to reduce the interference of background with the extraction of discriminative foreground features. Experimental results on three widely used person Re-ID benchmarking datasets (Market-1501, DukeMTMC-reID, and MSMT17) have demonstrated that the proposed method achieves promising performance compared with the state-of-the-art methods.
Hyperspectral band selection aims to identify an optimal subset of bands for hyperspectral images (HSIs). For most existing clustering-based band selection methods, they directly stretch each band into a single feature vector and employ the pixelwise features to address band redundancy. In this way, they do not take full consideration of the spatial information and deal with the importance of different regions in HSIs, which leads to a nonoptimal selection. To address these issues, a region-aware hierarchical latent feature representation learning-guided clustering (HLFC) method is proposed. Specifically, in order to fully preserve the spatial information of HSIs, the superpixel segmentation algorithm is adopted to segment HSIs into multiple regions first. For each segmented region, the similarity graph is constructed to reflect the bands-wise similarity, and its corresponding Laplacian matrix is generated for learning low-dimensional latent features in a hierarchical way. All latent features are then fused to form a unified feature representation of HSIs. Finally, $k$ -means clustering is utilized on the unified feature representation matrix to generate multiple clusters from which the band with maximum information entropy is selected to form the final subset of bands. Extensive experimental results demonstrate that the proposed clustering method can achieve superior performance than the state-of-the-art representative methods on the band selection. The demo code of this work is publicly available at https://github.com/WangJun2023/HLFC.
A text semantic classification is an essential approach to recognising the verbal intention of online learners, empowering reliable understanding and inquiry for the regulations of knowledge construction amongst students. However, online learning is increasingly switching from static watching patterns to the collaborative discussion. The current deep learning models, such as CNN and RNN, are ineffective in classifying verbal content contextually. Moreover, the contribution of verbal elements to semantics is often considerably varied, requiring the attachment of weights to these elements to increase verbal recognition precision. The Bi-LSTM is considered to be an adaptive model to investigate semantic relations according to the context. Moreover, the attention mechanism in deep learning simulating human vision could assign weights to target texts effectively. This study proposed to construct a deep learning model combining Bi-LSTM and attention mechanism, in which Bi-LSTM obtained the verbal features and keywords, and the generated keywords were weighed in accordance with the attention mechanism. A total of 12,000 sentences generated in online collaborative discussion activities have been classified into six categories, namely statement, negotiation, question, management, emotion and others. Results showed that the classification accuracy of Attention-Bi-LSTM reached 81.50%, which is higher than that of the baseline Bi-LSTM model. This study theoretically uncovers the features of collaborative discussion of onliners and practically provides an effective approach to automatic behaviour analysis in an online context.
This work discuss the task of automatically detecting satire instances in short articles. It is the study of extracting the most optimal features by using a deep learning architecture combined with carefully handcrafted contextual features. It is found that a few sets can perform well when they are used independently, but the others not so much. However, even the latter sets become very useful after the combination process with the former sets. This shows that each of the feature sets are significant. Finally, the combined feature sets undergoes the classification using well-known machine learning classification algorithms. The best algorithm for this task is found to be Logistic Regression. The outcome of all the experiments are good in all the metrics used. The result comparison to existing works in the same domain shows that the proposed method is slightly better with 0.94 in terms of F1-measure, while existing works managed to obtain 0.91 [1], 0.90 [2] and 0.88 [3]. The performance of each feature sets are also given as additional information. The main purpose of this work is to show that the combination of features extracted using supervised learning with the ones extracted manually can yield a good performance. It is also to open doors for other researchers to take into account the contextual meaning behind a figurative language type such as satire.
Artificial intelligence is bringing revolutionary changes to so many industries, by introducing them to a new era, full of technological advancements. The healthcare industry has been one of the most beneficial to this change, by merging digital transformation and healthcare, to form digital healthcare. Thereby introducing digital pathology, which implements image processing algorithms to help pathologists analyze and examine a diagnosis faster and more efficiently. It not only reduces the long hours pathologists used to take in laboratory analysis but also reduces human error. Therefore, healthcare digitalization has allowed the integration of computer vision into the medical field, with the use of Artificial intelligence techniques such as deep learning and machine learning algorithms. However, past research work has been limited to using AI models to diagnosis one specific disease at a time. Whereas this research work aims to develop an AI model that will automatically perform pathological analysis, to determine the diagnosis for multiple diseases from a medical image, then provide the medical report, while securing the patient's data, and assisting them with any questions they might have regarding the diagnosis. This research applies deep learning and machine learning algorithms for image classification via CNN architectures and feature extraction via Morphological properties. The model achieved great outcomes, with high accuracy and good F1-score results of 90.47% and 0.8332 respectively. The resultant model diagnoses 12 medical disorders, with an overall of 29 diagnostic cases, making it the only one of its kind in digitized healthcare applications.