The ability to incrementally learn new classes is crucial the development of real-world artificial intelligence systems. In this paper, we focus on a challenging but practical few-shot class-incremental learning (FSCIL) problem. FSCIL requires CNN models from very few labelled samples, without forgetting previously learned ones. To address problem, represent knowledge using neural gas (NG) network, which can and preserve topology feature manifold formed by different classes. On basis,...

In this paper, we focus on the challenging few-shot class incremental learning (FSCIL) problem, which requires to transfer knowledge from old tasks new ones and solves catastrophic forgetting. We propose exemplar relation distillation framework balance of old-knowledge preserving new-knowledge adaptation. First, construct an graph represent learned by original network update gradually for learning. Then loss function discovering between different classes is introduced learn structural...

Deep models have shown to be vulnerable catastrophic forgetting, a phenomenon that the recognition performance on old data degrades when pre-trained model is fine-tuned new data. Knowledge distillation (KD) popular incremental approach alleviate forgetting. However, it usually fixes absolute values of neural responses for isolated historical instances, without considering intrinsic structure by convolutional network (CNN) model. To overcome this limitation, we recognize importance global...

Rheumatoid arthritis (RA) is a common chronic immune disease. Berberine, as its main active ingredient, was also contained in variety of medicinal plants such Berberaceae, Buttercup, and Rutaceae, which are widely used digestive system diseases traditional Chinese medicine with anti-inflammatory antibacterial effects. The aims this article were to explore the therapeutic effect mechanism berberine on rheumatoid arthritis.Cell Counting Kit-8 evaluate proliferation RA fibroblast-like...

We develop a fine-grained image classifier using general deep convolutional neural network (DCNN). improve the classification accuracy of DCNN model from following two aspects. First, to better h -level hierarchical label structure classes contained in given training data set, we introduce fully connected (fc) layers replace top fc layer and train them with cascaded softmax loss. Second, propose novel loss function, namely, generalized large-margin (GLM) loss, make explicitly explore...

In this paper, we build a multilabel image classifier using general deep convolutional neural network (DCNN). We propose novel objective function that consists of three parts, i.e., max-margin objective, max-correlation and correntropy loss. The explicitly enforces the minimum score positive labels must be larger than maximum negative by predefined margin, which not only improves accuracies classifier, but also eases threshold determination. can make DCNN model learn latent semantic space,...

We propose a novel method for improving performance accuracies of convolutional neural network (CNN) without the need to increase complexity. accomplish goal by applying proposed Min-Max objective layer below output CNN model in course training. The explicitly ensures that feature maps learned have minimum within-manifold distance each object manifold and maximum between-manifold distances among different manifolds. is general able be applied CNNs with insignificant increases computation...

In this paper, we propose a novel single-task continual learning framework named Bi-Objective Continual Learning (BOCL). BOCL aims at both consolidating historical knowledge and from new data. On one hand, to preserve the old using small set of pillars, develop pillar consolidation (PLC) loss alleviate catastrophic forgetting problem. other contrastive (CPL) term improve classification performance, examine several data sampling strategies for efficient onsite ‘new’ with reasonable amount...

Drug-resistant tuberculosis (TB) poses a major threat to global TB control; consequently, there is an urgent need develop novel anti-TB drugs or strategies. Host-directed therapy (HDT) emerging as effective treatment strategy, especially for drug-resistant TB. This study evaluated the effects of berbamine (BBM), bisbenzylisoquinoline alkaloid, on mycobacterial growth in macrophages. BBM inhibited intracellular Mycobacterium (Mtb) by promoting autophagy and silencing ATG5, partially...

Inspired by the global–local information processing mechanism in human visual system, we propose a novel convolutional neural network (CNN) architecture named cognition-inspired (CogNet) that consists of global pathway, local and top-down modulator. We first use common CNN block to form pathway aims extract fine features input image. Then, transformer encoder capture structural contextual among parts Finally, construct learnable modulator where are modulated representations pathway. For ease...

Virtual prototyping of power electronic modules aims to allow rapid evaluation potential designs without building and testing physical prototypes. Among the interests in thermal models virtual modules, process compact needs effective methodology fast generate small describing performance a design. This study chooses Generalized Minimized Residual (GMRES) Algorithm due its efficiency. Based on that, machine learning aided surrogate model is proposed for prediction since existing approaches...

The ability to incrementally learn new classes is crucial the development of real-world artificial intelligence systems. In this paper, we focus on a challenging but practical few-shot class-incremental learning (FSCIL) problem. FSCIL requires CNN models from very few labelled samples, without forgetting previously learned ones. To address problem, represent knowledge using neural gas (NG) network, which can and preserve topology feature manifold formed by different classes. On basis,...

Solution-phase epitaxy is a versatile method to synthesize functional nanomaterials with customized properties, where supports play central role as they not only serve nucleation templates but also greatly affect the local electronic structures. However, developing remains great challenge. Herein, inspired by commonly observed of minerals in natural environment, we report using calcination-modified kaolinite support for epitaxial growth hexagonal CoO nanoparticles (

The human visual system can recognize object categories accurately and efficiently is robust to complex textures noises. To mimic the analogy-detail dual-pathway cognitive mechanism revealed in recent science studies, this article, we propose a novel convolutional neural network (CNN) architecture named <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">analogy-detail networks</i> (ADNets) for accurate recognition. ADNets disentangle information...

Class Incremental Learning (CIL) is a hot topic in machine learning for CNN models to learn new classes incrementally. However, most of the CIL studies are image classification and object recognition tasks few available video action classification. To mitigate this problem, paper, we present Grow When Required network (GWR) based framework GWR learns knowledge incrementally by modeling manifold frames each encountered class feature space. We also introduce Knowledge Consolidation (KC) method...

Advancements in self-supervised pre-training (SSL) have significantly advanced the field of learning transferable time series representations, which can be very useful enhancing downstream task. Despite being effective, most existing works struggle to achieve cross-domain SSL pre-training, missing valuable opportunities integrate patterns and features from different domains. The main challenge lies significant differences characteristics time-series data across domains, such as variations...