Existing self-knowledge distillation (Self-KD) solutions usually focus on transferring historical predictions of individual instances to the current network. However, this approach tends to create overconfidence for e...
Existing self-knowledge distillation (Self-KD) solutions usually focus on transferring historical predictions of individual instances to the current network. However, this approach tends to create overconfidence for easy instances and underconfidence for hard instances. The widely used temperature-based strategies to smooth or sharpen the predicted distributions can lead to inconsistencies across instances, causing sensitivity issues. To address this, our approach views a queue of instances as an ensemble rather than treating each instance independently. We propose a novel method that distills historical knowledge from a dimensional perspective, utilizing intra class characteristics and interclass relationships within each ensemble. First, we align each dimension distribution from the current network to the historical output. Second, we ensure each dimension is closer to similar dimensions than dissimilar ones, maintaining consistent attitudes from present and historical perspectives. Our insights reveal that distilling historical knowledge from a dimensional perspective is more effective than the traditional instance-based approach, with potential applications in related tasks. Empirical results on three famous datasets and various network architectures demonstrate the superiority of our proposed method. Our code is available at https://***/WenkeHuang/DimSelfKD.
With the rising popularity of online social interactions, emojis play a pivotal role in communication, effectively conveying people's emotions. Hence, accurately converting facial micro-expressions into correspond...
详细信息
Image deraining is a highly ill-posed *** significant progress has been made due to the use of deep convolutional neural networks,this problem still remains challenging,especially for the details restoration and gener...
详细信息
Image deraining is a highly ill-posed *** significant progress has been made due to the use of deep convolutional neural networks,this problem still remains challenging,especially for the details restoration and generalization to real rain *** this paper,we propose a deep residual channel attention network(DeRCAN)for *** channel attention mechanism is able to capture the inherent properties of the feature space and thus facilitates more accurate estimations of structures and details for image *** addition,we further propose an unsupervised learning approach to better solve real rain images based on the proposed *** qualitative and quantitative evaluation results on both synthetic and real-world images demonstrate that the proposed DeRCAN performs favorably against state-of-the-art methods.
Solar flares are one of the strongest outbursts of solar activity,posing a serious threat to Earth’s critical infrastructure,such as communications,navigation,power,and ***,it is essential to accurately predict solar...
详细信息
Solar flares are one of the strongest outbursts of solar activity,posing a serious threat to Earth’s critical infrastructure,such as communications,navigation,power,and ***,it is essential to accurately predict solar flares in order to ensure the safety of human ***,the research focuses on two directions:first,identifying predictors with more physical information and higher prediction accuracy,and second,building flare prediction models that can effectively handle complex observational *** terms of flare observability and predictability,this paper analyses multiple dimensions of solar flare observability and evaluates the potential of observational parameters in *** flare prediction models,the paper focuses on data-driven models and physical models,with an emphasis on the advantages of deep learning techniques in dealing with complex and high-dimensional *** reviewing existing traditional machine learning,deep learning,and fusion methods,the key roles of these techniques in improving prediction accuracy and efficiency are *** prevailing challenges,this study discusses the main challenges currently faced in solar flare prediction,such as the complexity of flare samples,the multimodality of observational data,and the interpretability of *** conclusion summarizes these findings and proposes future research directions and potential technology advancement.
Recommender systems aim to filter information effectively and recommend useful sources to match users' requirements. However, the exponential growth of information in recent social networks may cause low predictio...
详细信息
In the recent years, the development of generative models has stimulated the health care progress, specially medical image generation. The synthetic medical images can be applied to several fields and have many utiliz...
详细信息
Due to the aging of the world's population, the incidence of retinal diseases is on the rise. Machine learning is expected to have a crucial role in identifying retinal disease. Multiple medical institutions coope...
详细信息
The rapid development of the Internet has led to the widespread dissemination of manipulated facial images, significantly impacting people's daily lives. With the continuous advancement of Deepfake technology, the...
详细信息
The rapid development of the Internet has led to the widespread dissemination of manipulated facial images, significantly impacting people's daily lives. With the continuous advancement of Deepfake technology, the generated counterfeit facial images have become increasingly challenging to distinguish. There is an urgent need for a more robust and convincing detection method. Current detection methods mainly operate in the spatial domain and transform the spatial domain into other domains for analysis. With the emergence of transformers, some researchers have also combined traditional convolutional networks with transformers for detection. This paper explores the artifacts left by Deepfakes in various domains and, based on this exploration, proposes a detection method that utilizes the steganalysis rich model to extract high-frequency noise to complement spatial features. We have designed two main modules to fully leverage the interaction between these two aspects based on traditional convolutional neural networks. The first is the multi-scale mixed feature attention module, which introduces artifacts from high-frequency noise into spatial textures, thereby enhancing the model's learning of spatial texture features. The second is the multi-scale channel attention module, which reduces the impact of background noise by weighting the features. Our proposed method was experimentally evaluated on mainstream datasets, and a significant amount of experimental results demonstrate the effectiveness of our approach in detecting Deepfake forged faces, outperforming the majority of existing methods.
The integration of the contrastive learning paradigm into deep clustering has led to enhanced performance in image clustering. However, in existing researches, the samples in the class of the target may be still treat...
详细信息
In the era of network communication,digital image encryption(DIE)technology is critical to ensure the security of image ***,there has been limited research on combining deep learning neural networks with chaotic mappi...
详细信息
In the era of network communication,digital image encryption(DIE)technology is critical to ensure the security of image ***,there has been limited research on combining deep learning neural networks with chaotic mapping for the encryption of digital ***,this paper addresses this gap by studying the generation of pseudo-random sequences(PRS)chaotic signals using dual logistic chaotic *** signals are then predicted using long and short-term memory(LSTM)networks,resulting in the reconstruction of a new chaotic *** the research process,it was discovered that there are numerous training parameters associated with the LSTM network,which can hinder training *** overcome this challenge and improve training efficiency,the paper proposes an improved particle swarm optimization(IPSO)algorithm to optimize the LSTM ***,the obtained chaotic signal from the optimized model training is further scrambled,obfuscated,and diffused to achieve the final encrypted *** research presents a digital image encryption(DIE)algorithm based on a double chaotic map(DCM)and *** algorithm demonstrates a high average NPCR(Number of Pixel Change Rate)of 99.56%and a UACI(Unified Average Changing Intensity)value of 33.46%,indicating a strong ability to resist differential ***,the proposed algorithm realizes secure and sensitive digital image encryption,ensuring the protection of personal information in the Internet environment.
暂无评论