Wind speed inference from environmental flow–structure interactions. Part 2. Leveraging unsteady kinematics

This work explores the relationship between wind speed and time-dependent structural motion response as a means of leveraging the rich information visible in flow–structure interactions for anemometry. We build on recent work by Cardona, Bouman and Dabiri (Flow, vol. 1, 2021, E4), which presented an approach using mean structural bending. Here, we present the amplitude of the dynamic structural sway as an alternative signal that can be used when mean bending is small or inconvenient to measure. A force balance relating the instantaneous loading and instantaneous deflection yields a relationship between the incident wind speed and the amplitude of structural sway. This physical model is applied to two field datasets comprising 13 trees of 4 different species exposed to ambient wind conditions. Model generalization to the diverse test structures is achieved through normalization with respect to a reference condition. The model agrees well with experimental measurements of the local wind speed, suggesting that tree sway amplitude can be used as an indirect measurement of mean wind speed, and is applicable to a broad variety of diverse trees.

Bearing Faulty Prognostic Approach Based on Multiscale Feature Extraction and Attention Learning Mechanism

Recently, researches on data-driven faulty identification have been achieving increasing attention due to the fast development of the modern conditional monitoring technology and the availability of the massive historical storage data. However, most industrial equipment is working under variable industrial operating conditions which can be a great challenge to the generalization ability of the normal data-driven model trained by the historical storage operating data whose distribution might be different from the current operating datasets. Moreover, the traditional data-driven faulty prognostic model trained on massive historical data can hardly meet the real-time requirement of the practical industry. Since the hierarchical feature extraction can enhance the model generalization ability and the attention learning mechanism can promote the prediction efficiency, this paper proposes a novel bearing faulty prognostic approach combining the U-net-based multiscale feature extraction network and the CBAM- (convolutional block attention module-) based attention learning network. First, time domain conditional monitoring signals are converted into the two-dimensional gray-scale image which can be applicable for the input of the CNN. Second, a CNN model based on the U-net structure is adopted as the feature extractor to hierarchically extract the multilevel features which can be very sensitive to the faulty information contained in the converted image. Finally, the extracted multilevel features containing different representations of the raw signals are sent to the designed CBAM-based attention learning network for high efficiency faulty classification with its unique emphasize discrimination characteristic. The effectiveness of the proposed approach is validated by two case studies offered by the CWRU (Case Western Reserved University) and the Paderborn University. The experimental result indicates that the proposed faulty prognostic approach outperforms other comparison models in terms of the generalization ability and the speed-up properties.

Robust Prediction of Personalized In Vivo Response to Unseen Drugs From In Vitro Screens Using a Novel Context-Aware Deconfounding Autoencoder

Accurate and robust prediction of patient-specific responses to drug treatments is critical for drug development and personalized medicine. However, patient data are often too scarce to train a generalized machine learning model. Although many methods have been developed to utilize cell line data, few of them can reliably predict individual patient clinical responses to new drugs due to data distribution shift and confounding factors. We have developed a novel Context-aware Deconfounding Autoencoder (CODE-AE) that can extract intrinsic biological signals masked by context-specific patterns and confounding factors. Extensive comparative studies demonstrated that CODE-AE effectively alleviated the out-of-distribution problem for the model generalization, significantly improved accuracy and robustness over state-of-the-art methods in predicting patient-specific in vivo drug responses purely from in vitro screens. Using CODE-AE, we screened 59 drugs for 9,808 cancer patients. Our results are consistent with existing clinical observations, suggesting the potential of CODE-AE in developing personalized anti-cancer therapies and drug-response biomarkers.

FLIP: Benchmark tasks in fitness landscape inference for proteins

Machine learning could enable an unprecedented level of control in protein engineering for therapeutic and industrial applications. Critical to its use in designing proteins with desired properties, machine learning models must capture the protein sequence-function relationship, often termed fitness landscape. Existing benchmarks like CASP or CAFA assess structure and function predictions of proteins, respectively, yet they do not target metrics relevant for protein engineering. In this work, we introduce Fitness Landscape Inference for Proteins (FLIP), a benchmark for function prediction to encourage rapid scoring of representation learning for protein engineering. Our curated tasks, baselines, and metrics probe model generalization in settings relevant for protein engineering, e.g. low-resource and extrapolative. Currently, FLIP encompasses experimental data across adeno-associated virus stability for gene therapy, protein domain B1 stability and immunoglobulin binding, and thermostability from multiple protein families. In order to enable ease of use and future expansion to new tasks, all data are presented in a standard format. FLIP scripts and data are freely accessible at https://benchmark.protein.properties/home.

Investigation of Improving The Pre-Training And Fine-Tuning of BERT Model For Biomedical Relation Extraction

Background: Recently, automatically extracting biomedical relations has been a significant subject in biomedical research due to the rapid growth of biomedical literature. Since the adaptation to the biomedical domain, the transformer-based BERT models have produced leading results on many biomedical natural language processing tasks. In this work, we will explore the approaches to improve the BERT model for relation extraction tasks in both the pre-training and fine-tuning stages of its applications. In the pre-training stage, we add another level of BERT adaptation on sub-domain data to bridge the gap between domain knowledge and task-specific knowledge. Also, we propose methods to incorporate the ignored knowledge in the last layer of BERT to improve its fine-tuning. Results: The experiment results demonstrate that our approaches for pre-training and fine-tuning can improve the BERT model performance. After combining the two proposed techniques, our approach outperforms the original BERT models with averaged F1 score improvement of 2.1% on relation extraction tasks. Moreover, our approach achieves state-of-the-art performance on three relation extraction benchmark datasets. Conclusions: The extra pre-training step on sub-domain data can help the BERT model generalization on specific tasks, and our proposed fine-tuning mechanism could utilize the knowledge in the last layer of BERT to boost the model performance. Furthermore, the combination of these two approaches further improves the performance of BERT model on the relation extraction tasks.

Characterisation and Classification of Foodborne Bacteria Using Reflectance FTIR Microscopic Imaging

This work investigates the application of reflectance Fourier transform infrared (FTIR) microscopic imaging for rapid, and non-invasive detection and classification between Bacillus subtilis and Escherichia coli cell suspensions dried onto metallic substrates (stainless steel (STS) and aluminium (Al) slides) in the optical density (OD) concentration range of 0.001 to 10. Results showed that reflectance FTIR of samples with OD lower than 0.1 did not present an acceptable spectral signal to enable classification. Two modelling strategies were devised to evaluate model performance, transferability and consistency among concentration levels. Modelling strategy 1 involves training the model with half of the sample set, consisting of all concentrations, and applying it to the remaining half. Using this approach, for the STS substrate, the best model was achieved using support vector machine (SVM) classification, providing an accuracy of 96% and Matthews correlation coefficient (MCC) of 0.93 for the independent test set. For the Al substrate, the best SVM model produced an accuracy and MCC of 91% and 0.82, respectively. Furthermore, the aforementioned best model built from one substrate was transferred to predict the bacterial samples deposited on the other substrate. Results revealed an acceptable predictive ability when transferring the STS model to samples on Al (accuracy = 82%). However, the Al model could not be adapted to bacterial samples deposited on STS (accuracy = 57%). For modelling strategy 2, models were developed using one concentration level and tested on the other concentrations for each substrate. Results proved that models built from samples with moderate (1 OD) concentration can be adapted to other concentrations with good model generalization. Prediction maps revealed the heterogeneous distribution of biomolecules due to the coffee ring effect. This work demonstrated the feasibility of applying FTIR to characterise spectroscopic fingerprints of dry bacterial cells on substrates of relevance for food processing.

Understanding Collections of Related Datasets Using Dependent MMD Coresets

Understanding how two datasets differ can help us determine whether one dataset under-represents certain sub-populations, and provides insights into how well models will generalize across datasets. Representative points selected by a maximum mean discrepancy (MMD) coreset can provide interpretable summaries of a single dataset, but are not easily compared across datasets. In this paper, we introduce dependent MMD coresets, a data summarization method for collections of datasets that facilitates comparison of distributions. We show that dependent MMD coresets are useful for understanding multiple related datasets and understanding model generalization between such datasets.

A Framework for Self-Supervised Federated Domain Adaptation

Unsupervised federated domain adaptation uses the knowledge from several distributed unlabelled source domains to complete the learning on the unlabelled target domain. Some of the existing methods have limited effectiveness and involve frequent communication. This paper proposes a framework to solve the distributed multi-source domain adaptation problem, referred as self-supervised federated domain adaptation (SFDA). Specifically, a multi-domain model generalization balance (MDMGB) is proposed to aggregate the models from multiple source domains in each round of communication. A weighted strategy based on centroid similarity is also designed for SFDA. SFDA conducts self-supervised training on the target domain to tackle domain shift. Compared with the classical federated adversarial domain adaptation algorithm, SFDA is not only strong in communication cost and privacy protection but also improves in the accuracy of the model.