Train Accurate Models Research Articles

Automatic segmentation of pericardial adipose tissue from cardiac MR images via semi-supervised method with difference-guided consistency.

Accurate and automatic segmentation of pericardial adipose tissue (PEAT) in cardiac magnetic resonance (MR) images is essential for the diagnosis and treatment of cardiovascular diseases. Precise segmentation is challenging due to high costs and the need for specialized knowledge, as a large amount of accurately annotated data is required, demanding significant time and medicalresources. In order to reduce the burden of data annotation while maintaining the high accuracy of segmentation tasks, this paper introduces a semi-supervised learning method to solve the limitations of current PEAT segmentationmethods. In this paper, we propose a difference-guided collaborative mean teacher (DCMT) semi-supervised method, designed for the segmentation of PEAT from DCMT consists of two main components: a semi-supervised framework with a difference fusion strategy and a backbone network MCM-UNet using Mamba-CNN mixture (MCM) blocks. The differential fusion strategy effectively utilizes the uncertain areas in unlabeled data, encouraging the model to reach a consensus in predictions across these difficult-to-segment yet information-rich areas. In addition, considering the sparse and scattered distribution of PEAT in cardiac MR images, which makes it challenging to segment, we propose MCM-UNet as the backbone network in our semi-supervised framework. This not only enhances the processing ability of global information, but also accurately captures the detailed local features of the image, which greatly improves the accuracy of PEATsegmentation. Our experiments conducted on the MRPEAT dataset show that our DCMT method outperforms existing state-of-the-art semi-supervised methods in terms of segmentation accuracy. These findings underscore the effectiveness of our approach in handling the specific challenges associated with PEATsegmentation. The DCMT method significantly improves the accuracy of PEAT segmentation in cardiac MR images. By effectively utilizing uncertain areas in the data and enhancing feature capture with the MCM-UNet, our approach demonstrates superior performance and offers a promising solution for semi-supervised learning in medical image segmentation. This method can alleviate the extensive annotation requirements typically necessary for training accurate segmentation models in medicalimaging.

Advancing anomaly detection in computational workflows with active learning

A computational workflow, also known as workflow, consists of tasks that are executed in a certain order to attain a specific computational campaign. Computational workflows are commonly employed in science domains, such as physics, chemistry, genomics, to complete large-scale experiments in distributed and heterogeneous computing environments. However, running computations at such a large scale makes the workflow applications prone to failures and performance degradation, which can slowdown, stall, and ultimately lead to workflow failure. Learning how these workflows behave under normal and anomalous conditions can help us identify the causes of degraded performance and subsequently trigger appropriate actions to resolve them. However, learning in such circumstances is a challenging task because of the large volume of high-quality historical data needed to train accurate and reliable models. Generating such datasets not only takes a lot of time and effort but it also requires a lot of resources to be devoted to data generation for training purposes. Active learning is a promising approach to this problem. It is an approach where the data is generated as required by the machine learning model and thus it can potentially reduce the training data needed to derive accurate models. In this work, we present an active learning approach that is supported by an experimental framework, Poseidon-X, that utilizes a modern workflow management system and two cloud testbeds. We evaluate our approach using three computational workflows. For one workflow we run an end-to-end live active learning experiment, for the other two we evaluate our active learning algorithms using pre-captured data traces provided by the Flow-Bench benchmark. Our findings indicate that active learning not only saves resources, but it also improves the accuracy of the detection of anomalies.

Deep Learning-Based In Situ Micrograph Synthesis and Augmentation for Crystallization Process Image Analysis

Deep learning-based in situ imaging and analysis for crystallization process are essential for optimizing product qualities, reducing experimental costs through real-time monitoring, and controlling the process. However, large and high-quality annotated datasets are required to train accurate models, which are time consuming. Therefore, we proposed a novel methodology that applied image synthesis neural networks to generate virtual information-rich images, enabling efficient and rapid dataset expansion while simultaneously reducing annotation costs. Experiments were conducted on the L-alanine crystallization process to obtain process images and to validate the proposed workflow. The proposed method, aided by interpolation augmentation and data warping augmentation to enhance data richness, utilized only 25% of the training annotations, consistently segmenting crystallization process images comparable to those models utilizing 100% of the training data annotations, achieving an average precision of nearly 98%. Additionally, based on the analysis of Kullback–Leibler divergence, the proposed method demonstrated excellent performance in extracting in situ information regarding aspect ratios and crystal size distributions during the crystallization process. Moreover, its ability to leverage expert labels with a four-fold enhanced efficiency holds great potential for advancing various applications in crystallization processes.

Immunohistochemistry annotations enhance AI identification of lymphocytes and neutrophils in digitized H&E slides from inflammatory bowel disease

Background and ObjectiveHistologic assessment of the immune infiltrate in H&E slides is vital in diagnosing and managing inflammatory bowel diseases, but these assessments are subjective and time-consuming even for those with expertise. The development of deep learning models to aid in these assessments has been limited by the paucity of image data with reliably annotated immune cells available for training. MethodsTo address these challenges, we developed a pipeline that automates the neutrophil and lymphocyte labeling in ROIs from digital H&E slides. The data included ROIs extracted from 19 digitized H&E slides and the same slides restained with immunohistochemistry. Our pipeline first delineates each nucleus in H&E ROIs. Using the colorimetric features of the immunohistochemical stains (red: neutrophils, green: lymphocytes) in the immunohistochemistry ROIs, each cell was labeled as a neutrophil, a lymphocyte, or another cell. The labels were then transferred to the corresponding H&E ROIs by image registration, and the ROI registration accuracy was assessed by the median target registration error resulting in a labeled dataset. The newly formed dataset (NeuLy-IHC) comprising 519 ROIs with 235,256 labeled cells (74,339 lymphocytes, 16,326 neutrophils and 144,591 other cells) was used to train the HoVer-Net(NeuLy) model. The performance of HoVer-Net(NeuLy) measured by DICE coefficient (segmentation accuracy) and F1-scores (classification accuracy), was compared to those achieved by HoVer-Net(MoNuSAC) and SMILE(MoNuSAC) publicly available models trained on cancer-containing ROIs from the MoNuSAC dataset with manual cell labeling and pathologists’ annotations. ResultsThe 1.0 μm median target registration error of ROIs observed was low demonstrating robust transferring of cellular labels from immunohistochemistry ROIs to H&E ROIs. In the test set comprising 76 NeuLy-IHC and 78 MoNuSAC ROIs, the HoVer-Net(NeuLy) achieved a DICE coefficient of 0.861 and F1-sores of 0.827, 0.838, and 0.828, for neutrophils, lymphocytes, and other cells, respectively, outperforming the HoVer-Net(MoNuSAC)'s and SMILE(MoNuSAC)’s DICE coefficient and F1 scores for each cell category. ConclusionsWe attribute the improved performance of HoVer-Net(NeuLy) to the larger number of immune cells in the NeuLy-IHC dataset (in total 5x more, including 21x more neutrophils) than in the MoNuSAC dataset. Despite being trained on data from inflammatory bowel disease specimens, our model maintained robust performance when tested on previously unseen data derived from cancer specimens. The NeuLy-IHC set provides opportunities for training accurate models to quantify the inflammatory infiltrate in digital histologic slides.

Learning three-dimensional aortic root assessment based on sparse annotations.

Analyzing the anatomy of the aorta and left ventricular outflow tract (LVOT) is crucial for risk assessment and planning of transcatheter aortic valve implantation (TAVI). A comprehensive analysis of the aortic root and LVOT requires the extraction of the patient-individual anatomy via segmentation. Deep learning has shown good performance on various segmentation tasks. If this is formulated as a supervised problem, large amounts of annotated data are required for training. Therefore, minimizing the annotation complexity is desirable. We propose two-dimensional (2D) cross-sectional annotation and point cloud-based surface reconstruction to train a fully automatic 3D segmentation network for the aortic root and the LVOT. Our sparse annotation scheme enables easy and fast training data generation for tubular structures such as the aortic root. From the segmentation results, we derive clinically relevant parameters for TAVI planning. The proposed 2D cross-sectional annotation results in high inter-observer agreement [Dice similarity coefficient (DSC): 0.94]. The segmentation model achieves a DSC of 0.90 and an average surface distance of 0.96mm. Our approach achieves an aortic annulus maximum diameter difference between prediction and annotation of 0.45mm (inter-observer variance: 0.25mm). The presented approach facilitates reproducible annotations. The annotations allow for training accurate segmentation models of the aortic root and LVOT. The segmentation results facilitate reproducible and quantifiable measurements for TAVI planning.

DeepCRISTL: deep transfer learning to predict CRISPR/Cas9 on-target editing efficiency in specific cellular contexts.

CRISPR/Cas9 technology has been revolutionizing the field of gene editing. Guide RNAs (gRNAs) enable Cas9 proteins to target specific genomic loci for editing. However, editing efficiency varies between gRNAs and so computational methods were developed to predict editing efficiency for any gRNA of interest. High-throughput datasets of Cas9 editing efficiencies were produced to train machine-learning models to predict editing efficiency. However, these high-throughput datasets have a low correlation with functional and endogenous datasets, which are too small to train accurate machine-learning models on. We developed DeepCRISTL, a deep-learning model to predict the editing efficiency in a specific cellular context. DeepCRISTL takes advantage of high-throughput datasets to learn general patterns of gRNA editing efficiency and then fine-tunes the model on functional or endogenous data to fit a specific cellular context. We tested two state-of-the-art models trained on high-throughput datasets for editing efficiency prediction, our newly improved DeepHF and CRISPRon, combined with various transfer-learning approaches. The combination of CRISPRon and fine-tuning all model weights was the overall best performer. DeepCRISTL outperformed state-of-the-art methods in predicting editing efficiency in a specific cellular context on functional and endogenous datasets. Using saliency maps, we identified and compared the important features learned by DeepCRISTL across cellular contexts. We believe DeepCRISTL will improve prediction performance in many other CRISPR/Cas9 editing contexts by leveraging transfer learning to utilize both high-throughput datasets and smaller and more biologically relevant datasets. DeepCRISTL is available via https://github.com/OrensteinLab/DeepCRISTL.

An Identification Method for Anomaly Types of Active Distribution Network Based on Data Mining

With the increasing penetration of distributed generators (DGs) and the growing demand for reliable power sources, it has become imperative to promptly identify anomalies in active distribution networks (ADNs). Additionally, anomaly identification is also crucial to assisting in targeting maintenance to handle failures after the action of relay protections. This study presents a highly accurate and rapidly responsive approach to identifying anomalies in ADNs. The proposed method uses three layers of intricately interconnected modules. In the first module, an iterative clustering layer is devised to process the original data. In the second module, an association rule layer is designed to identify hidden patterns based on the types of anomalies and the post-fault voltage. In the third module, a regression training layer is employed to train accurate identification models. The case study utilizes data samples collected by an IEEE 33-bus system with distributed generations as well as post-fault data from a practical distribution network. The results of comparative tests demonstrate that the proposed data-mining structure achieves reliable performance with high identification accuracy and short reaction times, indicating its ability to be applied to the state awareness system of ADNs.

Enhancing automated inspection in metal industries: zero-shot segmentation of surface defects using bounding box prompts

Surface defect detection poses a challenging problem that has been addressed for decades. Many of actual solutions are based on deep learning algorithms. However, these algorithms require a large amount of data to train accurate models. This becomes especially problematic for semantic segmentation algorithms, which need labeled datasets at the pixel level, a laborious and time-consuming task. Therefore, this paper proposes a semi-supervised method in which the predictions of an object detector are combined with the segmentation of a zero-shot model, eliminating the need to label a dataset for semantic segmentation. Results are compared with relevant supervised semantic segmentation models, such as UNet and DeepLabv3+. UNet achieves an F1 score of 0.824, while DeepLabv3+ achieves 0.847. The proposed method combining YOLOv8n and Segment Anything Model achieves an F1 score of 0.804. This slight loss of F1 is compensated by the significant time savings, unlike UNet and DeepLabv3+, where labeling of the dataset is necessary.

Synthetic Data for Deep Learning in Computer Vision & Medical Imaging: A Means to Reduce Data Bias

Deep-learning (DL) performs well in computer-vision and medical-imaging automated decision-making applications. A bottleneck of DL stems from the large amount of labelled data required to train accurate models that generalise well. Data scarcity and imbalance are common problems in imaging applications that can lead DL models towards biased decision making. A solution to this problem is synthetic data. Synthetic data is an inexpensive substitute to real data for improved accuracy and generalisability of DL models. This survey reviews the recent methods published in relation to the creation and use of synthetic data for computer-vision and medical-imaging DL applications. The focus will be on applications that utilised synthetic data to improve DL models by either incorporating an increased diversity of data that is difficult to obtain in real life, or by reducing a bias caused by class imbalance. Computer-graphics software and generative networks are the most popular data generation techniques encountered in the literature. We highlight their suitability for typical computer-vision and medical-imaging applications, and present promising avenues for research to overcome their computational and theoretical limitations.

A High-Quality Sample Generation Method for Improving Steel Surface Defect Inspection.

Defect inspection is a critical task in ensuring the surface quality of steel plates. Deep neural networks have the potential to achieve excellent inspection accuracy if defect samples are sufficient. Nevertheless, it is very different to collect enough samples using cameras alone. To a certain extent, generative models can alleviate this problem but poor sample quality can greatly affect the final inspection performance. A sample generation method, which employs a generative adversarial network (GAN), is proposed to generate high-quality defect samples for training accurate inspection models. To improve generation quality, we propose a production-and-elimination, two-stage sample generation process by simulating the formation of defects on the surface of steel plates. The production stage learns to generate defects on defect-free background samples, and the elimination stage learns to erase defects on defective samples. By minimizing the differences between the samples at both stages, the proposed model can make generated background samples close to real ones while guiding the generated defect samples to be more realistic. Experimental results show that the proposed method has the ability to generate high-quality samples that can help train powerful inspection models and thereby improve inspection performance.

Optimizing Facial Expression Recognition with Image Augmentation Techniques: VGG19 Approach on FERC Dataset

In the field of facial expression recognition (FER), the availability of balanced and representative datasets is key to success in training accurate models. However, Facial Expression Recognition Challenge (FERC) datasets often face the challenge of class imbalance, where some facial expressions have a much smaller number of samples compared to others. This issue can result in biased and unsatisfactory model performance, especially in recognizing less common facial expressions. Data augmentation techniques are becoming an important strategy as they can expand the dataset by creating new variations of existing samples, thus increasing the variety and diversity of the data. Data augmentation can be used to increase the number of samples for less common facial expression classes, thus improving the model's ability to recognize and understand diverse facial expressions. The augmentation results are then combined with balancing techniques such as SMOTE coupled with undersampling to improve model performance. In this study, VGG19 is used to support better model performance. This will provide valuable guidelines for optimizing more advanced CNN models in the future and may encourage further research in creating more innovative augmentation techniques.

Improving Semantic Segmentation via Efficient Self-Training.

Starting from the seminal work of Fully Convolutional Networks (FCN), there has been significant progress on semantic segmentation. However, deep learning models often require large amounts of pixelwise annotations to train accurate and robust models. Given the prohibitively expensive annotation cost of segmentation masks, we introduce a self-training framework in this paper to leverage pseudo labels generated from unlabeled data. In order to handle the data imbalance problem of semantic segmentation, we propose a centroid sampling strategy to uniformly select training samples from every class within each epoch. We also introduce a fast training schedule to alleviate the computational burden. This enables us to explore the usage of large amounts of pseudo labels. Our Centroid Sampling based Self-Training framework (CSST) achieves state-of-the-art results on Cityscapes and CamVid datasets. On PASCAL VOC 2012 test set, our models trained with the original train set even outperform the same models trained on the much bigger augmented train set. This indicates the effectiveness of CSST when there are fewer annotations. We also demonstrate promising few-shot generalization capability from Cityscapes to BDD100K and from Cityscapes to Mapillary datasets.

Analysis of domain shift in whole prostate gland, zonal and lesions segmentation and detection, using multicentric retrospective data

Despite being one of the most prevalent forms of cancer, prostate cancer (PCa) shows a significantly high survival rate, provided there is timely detection and treatment. Computational methods can help make this detection process considerably faster and more robust. However, some modern machine-learning approaches require accurate segmentation of the prostate gland and the index lesion. Since performing manual segmentations is a very time-consuming task, and highly prone to inter-observer variability, there is a need to develop robust semi-automatic segmentation models. In this work, we leverage the large and highly diverse ProstateNet dataset, which includes 638 whole gland and 461 lesion segmentation masks, from 3 different scanner manufacturers provided by 14 institutions, in addition to other 3 independent public datasets, to train accurate and robust segmentation models for the whole prostate gland, zones and lesions. We show that models trained on large amounts of diverse data are better at generalizing to data from other institutions and obtained with other manufacturers, outperforming models trained on single-institution single-manufacturer datasets in all segmentation tasks. Furthermore, we show that lesion segmentation models trained on ProstateNet can be reliably used as lesion detection models.

Discovering novel halide perovskite alloys using multi-fidelity machine learning and genetic algorithm.

Expanding the pool of stable halide perovskites with attractive optoelectronic properties is crucial to addressing current limitations in their performance as photovoltaic (PV) absorbers. In this article, we demonstrate how a high-throughput density functional theory (DFT) dataset of halide perovskite alloys can be used to train accurate surrogate models for property prediction and subsequently perform inverse design using genetic algorithm (GA). Our dataset consists of decomposition energies, bandgaps, and photovoltaic efficiencies of nearly 800 pure and mixed composition ABX3 compounds from both the GGA-PBE and HSE06 functionals, and are combined with ∼100 experimental data points collected from the literature. Multi-fidelity random forest regression models are trained on the DFT + experimental dataset for each property using descriptors that one-hot encode composition, phase, and fidelity, and additionally include well-known elemental or molecular properties of species at the A, B, and X sites. Rigorously optimized models are deployed for experiment-level prediction over >150 000 hypothetical compounds, leading to thousands of promising materials with low decomposition energy, band gap between 1 and 2eV, and efficiency of >15%. Surrogate models are further combined with GA using an objective function to maintain chemical feasibility, minimize decomposition energy, maximize PV efficiency, and keep bandgap between 1 and 2eV; thus, hundreds more optimal compositions and phases are discovered. We present an analysis of the screened and inverse-designed materials, visualize ternary phase diagrams generated for many systems of interest using machine learning predictions, and suggest strategies for further improvement and expansion in the future.

Weakly supervised deep learning-based concrete aggregates automatic segmentation for assessing separation degree

Vision-based automatic concrete aggregate segmentation helps to facilitate fast quantitative analysis of concrete aggregate shape, gradation, and uniformity. However, existing methods are mainly based on supervised deep learning. Since concrete surface images often contain numerous aggregates with irregular sizes and boundaries, this leads to a time-consuming and labor-intensive data annotation process. To reduce annotation costs, this study proposes a weakly supervised deep learning (WSDL) concrete aggregates segmentation method, which can train accurate pixel-level concrete aggregates segmentation models using only low-cost bounding box-level annotation. A concrete aggregate segmentation dataset containing 1400 images is created and experiments are conducted. The results show that the proposed method can achieve pixel-level concrete aggregate segmentation with an AP50 of 95.27 %. Compared with the supervised deep learning-based method, the annotation cost is reduced by 86.24 %. This study also verifies the feasibility of the proposed method for quantitative evaluation of concrete aggregate segregation degree. Expert systems for automatic analysis of concrete aggregate gradation and uniformity can be developed based on this method in the future.

High-Quality and Reproducible Automatic Drum Transcription from Crowdsourced Data

Within the broad problem known as automatic music transcription, we considered the specific task of automatic drum transcription (ADT). This is a complex task that has recently shown significant advances thanks to deep learning (DL) techniques. Most notably, massive amounts of labeled data obtained from crowds of annotators have made it possible to implement large-scale supervised learning architectures for ADT. In this study, we explored the untapped potential of these new datasets by addressing three key points: First, we reviewed recent trends in DL architectures and focused on two techniques, self-attention mechanisms and tatum-synchronous convolutions. Then, to mitigate the noise and bias that are inherent in crowdsourced data, we extended the training data with additional annotations. Finally, to quantify the potential of the data, we compared many training scenarios by combining up to six different datasets, including zero-shot evaluations. Our findings revealed that crowdsourced datasets outperform previously utilized datasets, and regardless of the DL architecture employed, they are sufficient in size and quality to train accurate models. By fully exploiting this data source, our models produced high-quality drum transcriptions, achieving state-of-the-art results. Thanks to this accuracy, our work can be more successfully used by musicians (e.g., to learn new musical pieces by reading, or to convert their performances to MIDI) and researchers in music information retrieval (e.g., to retrieve information from the notes instead of audio, such as the rhythm or structure of a piece).

Genomic prediction of pig growth traits based on machine learning.

This study aimed to assess and compare the performance of different machine learning models in predicting selected pig growth traits and genomic estimated breeding values (GEBV) using automated machine learning, with the goal of optimizing whole-genome evaluation methods in pig breeding. The research employed genomic information, pedigree matrices, fixed effects, and phenotype data from 9968 pigs across multiple companies to derive four optimal machine learning models: deep learning (DL), random forest (RF), gradient boosting machine (GBM), and extreme gradient boosting (XGB). Through 10-fold cross-validation, predictions were made for GEBV and phenotypes of pigs reaching weight milestones (100 kg and 115 kg) with adjustments for backfat and days to weight. The findings indicated that machine learning models exhibited higher accuracy in predicting GEBV compared to phenotypic traits. Notably, GBM demonstrated superior GEBV prediction accuracy, with values of 0.683, 0.710, 0.866, and 0.871 for B100, B115, D100, and D115, respectively, slightly outperforming other methods. In phenotype prediction, GBM emerged as the best-performing model for pigs with B100, B115, D100, and D115 traits, achieving prediction accuracies of 0.547, followed by DL at 0.547, and then XGB with accuracies of 0.672 and 0.670. In terms of model training time, RF required the most time, while GBM and DL fell in between, and XGB demonstrated the shortest training time. In summary, machine learning models obtained through automated techniques exhibited higher GEBV prediction accuracy compared to phenotypic traits. GBM emerged as the overall top performer in terms of prediction accuracy and training time efficiency, while XGB demonstrated the ability to train accurate prediction models within a short timeframe. RF, on the other hand, had longer training times and insufficient accuracy, rendering it unsuitable for predicting pig growth traits and GEBV.

Machine learning small molecule properties in drug discovery

Machine learning (ML) is a promising approach for predicting small molecule properties in drug discovery. Here, we provide a comprehensive overview of various ML methods introduced for this purpose in recent years. We review a wide range of properties, including binding affinities, solubility, and ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity). We discuss existing popular datasets and molecular descriptors and embeddings, such as chemical fingerprints and graph-based neural networks. We highlight also challenges of predicting and optimizing multiple properties during hit-to-lead and lead optimization stages of drug discovery and explore briefly possible multi-objective optimization techniques that can be used to balance diverse properties while optimizing lead candidates. Finally, techniques to provide an understanding of model predictions, especially for critical decision-making in drug discovery are assessed. Overall, this review provides insights into the landscape of ML models for small molecule property predictions in drug discovery. So far, there are multiple diverse approaches, but their performances are often comparable. Neural networks, while more flexible, do not always outperform simpler models. This shows that the availability of high-quality training data remains crucial for training accurate models and there is a need for standardized benchmarks, additional performance metrics, and best practices to enable richer comparisons between the different techniques and models that can shed a better light on the differences between the many techniques.

An effective correlation-based data modeling framework for automatic diabetes prediction using machine and deep learning techniques

The rising risk of diabetes, particularly in emerging countries, highlights the importance of early detection. Manual prediction can be a challenging task, leading to the need for automatic approaches. The major challenge with biomedical datasets is data scarcity. Biomedical data is often difficult to obtain in large quantities, which can limit the ability to train deep learning models effectively. Biomedical data can be noisy and inconsistent, which can make it difficult to train accurate models. To overcome the above-mentioned challenges, this work presents a new framework for data modeling that is based on correlation measures between features and can be used to process data effectively for predicting diabetes. The standard, publicly available Pima Indians Medical Diabetes (PIMA) dataset is utilized to verify the effectiveness of the proposed techniques. Experiments using the PIMA dataset showed that the proposed data modeling method improved the accuracy of machine learning models by an average of 9%, with deep convolutional neural network models achieving an accuracy of 96.13%. Overall, this study demonstrates the effectiveness of the proposed strategy in the early and reliable prediction of diabetes.

Efficient dynamic sampling of batch processes through operation recipes

The complexity of dynamic phenomena present in chemical processes often results in high evaluation costs of accurate first-principles models. This limits their real-time applicability, e.g. for advanced process control. A common solution is the derivation of simpler but faster data-driven surrogate models trained on simulated time series generated from dynamic samplings of a mechanistic model. For batch processes, known non-adaptive dynamic sampling methods lead to unrealistic or even infeasible operation cycles, raising the cost of generating simulated datasets with sufficient information content to train accurate surrogate models. An alternative sampling strategy is developed and analyzed, where sampled input trajectories are constrained to process knowledge in the form of parametrized operation recipes. The proposed methodology is tested for the case studies of full batch cycles of a crystallizer and a batch distillation column, showing that it is more efficient in terms of convergent simulations compared to an established dynamic sampling strategy.