Metal–organic frameworks (MOFs) have emerged as an important and unique class of functional crystalline hybrid porous materials in the past two decades. Due to their modular structures and adjustable ...pore system, such distinctive materials have exhibited remarkable prospects in key applications pertaining to adsorption such as gas storage, gas and liquid separations, and trace impurity removal. Evidently, gaining a better understanding of the structure–property relationship offers great potential for the enhancement of a given associated MOF property either by structural adjustments via isoreticular chemistry or by the design and construction of new MOF structures via the practice of reticular chemistry. Correspondingly, the application of isoreticular chemistry paves the way for the microfine design and structure regulation of presented MOFs. Explicitly, the microfine tuning is mainly based on known MOF platforms, focusing on the modification and/or functionalization of a precise part of the MOF structure or pore system, thus providing an effective approach to produce richer pore systems with enhanced performances from a limited number of MOF platforms. Here, the latest progress in this field is highlighted by emphasizing the differences and connections between various methods. Finally, the challenges together with prospects are also discussed.
The application of microfine design in metal–organic framework (MOF) structures and their influence on properties are summarized in five aspects including size effect, counterions, functional groups, metal diversities, and defects. The differences and connections between each of the regulation methods are discussed. Finally, current challenges and perspectives in microfine design are also presented.
H2 gas sensors for different applications require various detection ranges, such as 1–100 ppm for exhale breath test and 0–40000 ppm for H2 energy vehicles. Coarse-tuning of the detection range could ...be realized by the selection of the type of H2 sensors. The fine-tuning of the detection range within one type of H2 sensor, however, is little concerned and reported. Herein, we propose to achieve the fine-tuning of the H2 gas detection range of the AlGaN/GaN HEMT devices by adjusting the Pt gate thickness. Devices with various Pt gate thicknesses of 2, 20, 60, and 100 nm were fabricated and investigated. Results show that the HEMT devices have excellent pinch-off characteristic with an on-to-off ratio of ∼four orders of magnitude. For the 100 nm thick device exposed to 500 ppm H2, ultrafast response time of 1.5 s is observed together with high response. With the decrease of gate thickness, both the response and the response time gradually increase, 1850% and 6 s for the 2 nm thick device. Moreover, both the low limit of detection (LOD) and the saturation cencentration decrease from 1.6 to 0.14 ppm and from 30,000 to 5000 ppm, respectively, with the gate thickness reduced from 100 to 2 nm, revealing that fine-tuning of the detection range could be achieved by adjusting the gate thickness. Finally, the response activation energy is also studied, 15.9, 19.7, and 42.8 kJ/mol for 2, 60, and 100 nm thick devices, respectively.
Display omitted
•Employment of AlGaN/GaN HEMT device as a H2 gas sensor.•Demonstration of a high performance H2 sensor with small LOD ∼0.14 ppm and ultrafast response ∼1.5 s.•Realization the fine-tuning of the detection range by simply adjusting the gate thickness.•Study the dependence of the response activation energy Ea on the gate thickness.
Balancing model complexity is a key challenge of modern computational ecology, particularly so since the spread of machine learning algorithms. Species distribution models are often implemented using ...a wide variety of machine learning algorithms that can be fine‐tuned to achieve the best model prediction while avoiding overfitting. We have released SDMtune, a new R package that aims to facilitate training, tuning, and evaluation of species distribution models in a unified framework. The main innovations of this package are its functions to perform data‐driven variable selection, and a novel genetic algorithm to tune model hyperparameters. Real‐time and interactive charts are displayed during the execution of several functions to help users understand the effect of removing a variable or varying model hyperparameters on model performance. SDMtune supports three different metrics to evaluate model performance: the area under the receiver operating characteristic curve, the true skill statistic, and Akaike's information criterion corrected for small sample sizes. It implements four statistical methods: artificial neural networks, boosted regression trees, maximum entropy modeling, and random forest. Moreover, it includes functions to display the outputs and create a final report. SDMtune therefore represents a new, unified and user‐friendly framework for the still‐growing field of species distribution modeling.
The main innovations of SDMtune are a novel genetic algorithm to tune the hyperparameters of a model and functions to perform data‐driven variable selection. Real‐time and interactive charts are displayed during the execution of several functions to help users understand the effect of removing a variable or varying model hyperparameters on model performance.
Rice (Oryza sativa) is a principal cereal crop in the world. It is consumed by greater than half of the world's population as a staple food for energy source. The yield production quantity and ...quality of the rice grain is affecting by abiotic and biotic factors such as precipitation, soil fertility, temperature, pests, bacteria, virus, etc. For disease management, farmers spending lot of time and resources and they detect the diseases through their penniless naked eye approach which leads to unhealthy farming. The advancement of technical support in agriculture greatly assists for automatic identification of infectious organisms in the rice plants leaves. The convolutional neural network algorithm (CNN) is one of the algorithms in deep learning has been triumphantly invoked for solving computer vision problems like image classification, object segmentation, image analysis, etc. In our work, InceptionResNetV2 is a type of CNN model utilized with transfer learning approach for recognizing diseases in rice leaf images. The parameters of the proposed model is optimized for the classification task and obtained a good accuracy of 95.67%.
•Rice leaf diseases are modelled by deep learning techniques.•Prediction accuracy of various diseases were compared.•Proposed deep neural networks with transfer learning detects rice leaf diseases with improved accuracy.
In mobile scenarios, there is a need for general user representations to solve multiple target tasks. However, there are some challenges in the related research (e.g., difficulty in learning a ...representation that satisfies both great generalization and performance). To address these problems, we proposed a network for downstream-adaptable mobile user modeling, which employed a novel fine-tuning strategy for optimizing the performance of several downstream tasks. Additionally, we designed a time-difference module to eliminate the impact of low-frequency and non-uniform app usage behavior over time. A parallel decoder structure was developed to incorporate multi-type features by minimizing information loss. We evaluated our method on a real-world dataset of 100,000 mobile users and three downstream tasks (i.e., age prediction, gender prediction, and app recommendation). The experimental results showed that our method could outperform existing methods significantly. It achieved 96.5% ACC on gender prediction, 68.1% ACC on age prediction, and 64.2% Recall@5 on app recommendation. These results imply that our method performs well on both generalization and performance. It could be anticipated promising to the unseen tasks inference.
The rapid development of big data leads to many researchers focusing on improving bearing fault classification accuracy using deep learning models. However, implementing a deep learning model on a ...limited resources platform such as the smartphone or STM32 includes two difficulties: making the model as lightweight as possible and reducing the dependence on large training samples. To this end, a self-attention ensemble lightweight model combined with the transfer learning (SLTL) method is proposed to solve these intractable problems, which are “small, light, and fast.” Firstly, the raw vibration signal is constructed into time–frequency images by continuous wavelet transform (CWT). Secondly, we build a self-attention lightweight convolutional neural network (SLCNN) model by integrating a self-attention mechanism (SAM) into the optimized SqueezeNet model. Then, based on a well-trained SLCNN in ImageNet, rich parameter knowledge is transferred from the pre-trained model to the target model. Finally, the fever training samples are used to fine-tune the target model. Experimental results on two bearing datasets validate the effectiveness of the SLTL method, which achieves 99.5% of classification accuracy with fewer training samples than other conventional CNN models. More importantly, the model parameters of SLTL are 0.95 M, and the floating-point operations (FLOPs) are 0.11 M, indicating that SLTL possesses high accuracy while maintaining lightweight, which benefits the platform with limited resources.
•A new transfer learning based approach for the classification of breast cancer in histopathological images is proposed.•It can handle both magnification dependent (MI) and independent (MI) binary ...and eight-class classifications simultaneously.•The residual CNN ResNet-18 is investigated as a backbone model with block-wise fine-tuning strategy.•GCN from the target’s task values and three-fold data augmentation are exploited to boost the classification performance.•The proposed approach outperforms 11 recent state-of-the-art MD and MI counterparts by a fair margin.
The visual analysis of histopathological images is the gold standard for diagnosing breast cancer, yet a strenuous and an intricate task that requires years of pathologist training. Therefore, automating this task using computer-aided diagnosis (CAD) is highly expected. This paper proposes a new transfer learning-based approach to automated classification of breast cancer from histopathological images, including magnification dependent (MD) and magnification independent (MI) binary and eight-class classifications. We apply the deep neural network ResNet-18 to this problem, which is pre-trained on ImageNet, a large dataset of common images. We then design our transfer learning method to refine the network on histopathological images. Our transfer learning method is based on block-wise fine-tuning strategy; in which we make the last two residual blocks of the deep network model more domain-specific to our target data. It substantially helps to avoid over-fitting and speed up the training. Furthermore, we strengthen the adaptability of the proposed approach by using global contrast normalization (GCN) based on the target’s data values and three-fold data augmentation on training data. The experimental results of MD and MI binary and eight-class classifications on the publicly available BreaKHis dataset demonstrate that our approach is promising and effective, outperforming recent state-of-the-art MD and MI counterparts by a fair margin.
Skin cancer is recognized as the most common kind of cancer in the world. It could be deadly if not identified at the primary stage, which makes early detection very crucial. It is possible to ...identify it with the naked eye, but high inter-class similarity and intra-class variations make it too hard to detect. Due to the prevalence of this disease around the world, so far many automated systems have been developed based on deep learning to assist the physician in the early detection of skin lesions.
In this study, we propose a weighted average ensemble learning-based model to classify seven types of skin lesions. We used five deep neural network models, namely, ResNeXt, SeResNeXt, ResNet, Xception, and DenseNet as the base of the ensemble. For the training and evaluation of our models, we used 18,730 dermoscopy images collected from the official Human Against Machine (HAM10000) and ISIC 2019 dataset together with class balancing, noise removal, and data augmentation technique. We found the best combinations of the base models in the ensemble using the grid search method and optimized the impact of each base model for the average recall score.
The five models performed excellent during evaluation and 88%, 89%, 91%, 88%, and 84% macro-average recall score were achieved by ResNeXt, SeResNeXt, DenseNet, Xception, and ResNet respectively. The simple average ensemble model boosted the result to 93% and the weighted average ensemble obtained a 94% recall score. The grid search method showed that the impact of all models are almost equal in the final model.
The average ensemble can improve the result by a significant amount. Our proposed system performed better than other existing systems and can support dermatologists for diagnosis.
•A deep learning based diagnosis system is proposed.•The proposed system identifies seven kinds of skin lesion.•Fine-tuning is applied to 5 different state-of-the-art convolutional networks.•The proposed system outperforms all the existing systems.
Large language models (LLMs) are proving to be very useful in many fields, especially chemistry and biology, because of their amazing capabilities. Biomolecular data is often represented ...sequentially, much like textual data used to train LLMs. However, developing LLMs from scratch requires a substantial amount of data and computational resources, which may not be feasible for most researchers. A more workable solution to this problem is to change the inputs or parameters so that the previously trained general LLMs can pick up the specific knowledge needed for biomolecular analysis. These adaption strategies lower the amount of data and hardware needed, providing a more affordable option. This review provides the introduction of two popular LLM adaptation techniques: fine-tuning and prompt engineering, along with their uses in the analysis of molecules, proteins, and genes. A thorough overview of current common datasets and pre-trained models is also provided. This review outlines the possible advantages and difficulties of LLMs for biomolecular analysis, opening the door for chemists and biologists to effectively utilize LLMs in their future studies.
•A thorough summary of the use of large language models in biomolecular analysis.•Lists of models and datasets in analyzing molecules, proteins, and genes.•Future directions on improving large language models for biomolecular analysis.
PDF
