The MAE results are shown in Fig. Comparison of different sampling methods. The value functions in a—d are evaluated from direct labelled data.

The value functions in e and f are reused from similar tasks. a Experimental results of Cifar b Experimental results of CWRU HP1. c Experimental results of thermal lag prediction of the composite. d Experimental results of Tool wear B3C6.

e Results of reusing the value function of task CWRU HP0 on HP1. f Results of reusing the value function of task B2C4 on B3C6. Figure 3 a and d illustrates that HighAV can consistently achieve superior performance, especially when the number is limited. Under most circumstances, HighAV outperforms the uncertainty boundary of Random grey region while Cluster is only better than Random occasionally but far more unstable.

The Cluster results fluctuate sharply because similar sample sizes e. In Fig. Theoretically, minimising the aggregation value can also provide the worst sample set.

As seen in Fig. Although the low valuable sample set seems meaningless for real application, it does reveal the importance of the distribution of training data, as well as the magic of aggregation-value-based sampling.

Table 1 summaries the regression and classification results with training data from different sampling methods under different sample sizes 30, 50, 80, It is clear that the proposed aggregation-value-based sampling method can provide better sample sets compared to other sampling methods.

Summary of performance with training data from different sampling methods Random, Cluster, HighSV, HighAV. The best results are highlighted bold. This table shows that HighAV can achieve better performance under the same number of samples.

To avoid data labelling for the value function, in this section we investigate the possibility of reusing the value function learnt from a similar task on the target task without training a new one.

It can be observed that the accuracy of HighSV is even lower than Random, but HighAV can consistently achieve leading performance. This phenomenon reveals that the effectiveness of HighSV relies heavily on the accuracy of the value function.

However, HighAV is more robust, meaning that a less accurate value function can still provide helpful value information. The same conclusions can also be drawn from Fig. In this section we investigate Scheme C for the composites curing case, in which the value function is first calculated from the simplified low-fidelity finite difference FD model, and then reused for parameters designed in high-fidelity FEM simulations.

An illustration of the curing of a 1D composite-tool system is shown in Fig. The actual temperature of the composite part always lags behind the designed cure cycle Fig. Thus, the thermal lag is defined as the maximum difference between the cure cycle and the actual temperature of any point in the thickness during the heat-up step [ 5 , 6 ].

The objective here is to establish the data-driven prediction model of thermal lag from the simulation results, where the input features include the heating rate, the cooling rate, the hold temperature, the hold time and the heat transfer coefficients of both sides Fig.

Since the labelled data comes from the time-consuming high-fidelity FEM simulation, a better sampling method should reduce the number of simulations but maintain the required accuracy of the data-driven model.

Experimental results of Scheme C, the thermo-chemical analysis of the composite. a Illustration of the 1D composite-tool curing system. b The cure cycle and the thermal lag in composite curing. c The defined data-driven task from the curing parameters to the corresponding thermal lag.

d The full workflow of sampling curing parameters for composite simulation. e MAEs of 10 repeated trails of different sample selection methods with 40 samples.

f Required samples of different sample selection methods to achieve an MAE of 5 K. The detailed procedure of aggregation-value-based sampling for this case is shown in Fig.

An optimal parameter sample set S is then determined based on the proposed sampling method for the subsequent complete high-fidelity FEM simulations. A Gaussian process regression model is then trained on the simulation results of the selected samples and evaluated on the test set.

The MAEs of 10 repeated trials for four methods are shown in Fig. It can be observed that HighAV can achieve a superior and stable performance with an MAE around 5 K.

Conversely, Cluster is slightly better than Random, and HighSV is very unstable, even worse than Random. These results show that the distribution of the designed curing parameter combinations significantly influences the performance of data-driven models, and the proposed HighAV can provide a better sample set stably.

Figure 4 f reports how many samples are required to achieve an MAE of 5 K for different sample selection methods. In each independent experiment, a sample set is constructed by increasing instances one by one from an empty set until the MAE becomes less than 5 K stably.

The size of the final sample set is recorded as the required size of this trial. As shown in the scatter and box plots of 10 repeated tests in Fig. Table 2 reports the detailed required samples for different sampling methods to stably achieve MAEs of 5 and 6 K. These results demonstrate that the proposed sampling method can reduce the data-collecting effort of FEM simulations in the composite curing problem while maintaining the required accuracy.

The required number of samples M for different sampling methods to achieve a predefined required MAE. Considering the uncertainties of different methods, the number M is defined as follows: during the sampling from 20 to points, for any sample set with more than M samples, the MAE is always less than the required one.

Here A±B represents the mean A and standard deviation B of the required number M in 10 repeated trials. Dimensional inspection and reconstruction of engineering products comprising free-form surfaces requires accurate measurement of a large number of discrete points using a coordinate measuring machine with a touch-trigger probe [ 33 ].

An efficient sampling method should enable the reconstruction of the surface under the required accuracy with a limited amount of measurement points.

Curvature and other geometric features are widely used prior knowledge for traditional measurement sampling methods. The simulated measurements and reconstructed results of a matlab ® peak surface are shown in Fig. The absolute Gaussian curvature function in Fig.

Figure 5 b is the error distribution map of the reconstructed surface with measurement points sampled by HighAV. The MAE is 0. Figure 5 c shows the errors of the surface reconstructed with points sampled by Cluster.

MAE and MAX are 0. It is clear that HighAV can reduce the error of areas with high curvatures, which plays a similar role as traditional curvature-based sampling.

Figure 5 d reports the MAEs of four sampling methods with different numbers of samples ranging from 20 to HighAV has a small MAE for almost any size of sample. Table 2 reports the required samples for different sampling methods to stably achieve MAEs of 0. It is clear that HighAV can reduce the required measurement points under the predefined MAE.

The full workflow of sampling measurement points is shown in Fig. Experimental results of the surface measurement and reconstruction, defining the value function from the absolute Gaussian curvature.

a The absolute Gaussian curvature function of peaks surface. b The error map of the surface reconstructed from the points sampled by HighAV. The mean absolute and the maximum errors are 0. c The error map of the surface reconstructed from the points sampled by Cluster. d The relationship between the number of samples and the MAE of the reconstructed surfaces for different sampling methods.

e The full workflow of sampling measurement points for the surface measurement and reconstruction. The abovementioned results show that aggregation-value-based sampling can provide superior and stable sample sets compared with Shapley-value-based or representativeness-based methods.

This section comprehensively analyses the characteristics of aggregation-value-based sampling on the composite task and explains why it works. The analyses of other cases are reported in the online supplementary material S4. Figure 6 a—c shows the t -distributed stochastic neighbour embedding visualised features of samples in the composite task.

These samples are generated by HighSV, HighAV and Cluster. The sampled points are marked with large points, and all points of the dataset are marked with small points. Characteristics analysis of the composites task. a A sample set of the composite task generated by HighSV. The contour map represents the Shapley value field, and the darker colour represents a larger value.

b A sample set of the composite task generated by HighAV. c A sample set of the composite task generated by Cluster. d The function between the number of samples and the corresponding MAE.

e The sensitivity of HighAV on different kernel functions. f The sensitivity of HighAV on the parameter σ.

g The degeneration from HighAV to HighSV of the composite task with different σ. h The sensitivity of HighAV on the random error of the Shapley value on the composite task. The contour map in Fig.

Almost all the samples in Fig. Shapley-value-based sampling tends to be deficient because the sample set does not represent the dataset.

More experimental results about the high-value samples' clustering phenomenon are provided in the online supplementary material S4. The sample set of Cluster is representative of the probabilistic density of the dataset.

Still, samples in the high-value area are random and insufficient, which could result in the unstable fluctuation of Cluster, as in Fig. As shown in Fig. Because of the redundant information, the functions between the number of samples and the corresponding performance of the data-driven model are usually approximately logarithmic curve of Scores in Fig.

However, the aggregation value curve in Fig. In this section we analyse the sensitivity of the proposed method on the composite task.

Figure 6 e compares the HighAV results for the composite case with different kernel functions: the radial basis function RBF kernel, Laplace kernel and inverse multiquadric kernel.

It is clear that different kernel functions have comparable and similar MAE convergence curves despite slight differences. The bandwidth parameter of the kernel function σ, which determines the influence range, is the one and the only parameter in the aggregation-value-based sampling. When σ is too small, the neighbouring values will not be aggregated, and aggregation-value-based sampling will degenerate into Shapley-value-based sampling.

On the other hand, aggregation-value-based sampling will be less effective when σ is too large, because the aggregation value of all samples could be too similar to be distinguished.

Figure 6 f shows performance on the composite task concerning σ from 0. The darker shade means worse performance, and the dotted line is the selected parameter in the previous experiments.

Figure 6 g shows the degeneration process of the method as σ becomes smaller. The variations of MAE can also be observed in the bottom left corner of Fig.

Since the calculation of the Shapley value always brings random errors, we also analyse the sensitivities of HighSV, HighAV and LowAV with five random trials of the Shapley value function. For HighSV, the slight random error of the Shapley value changes the samples significantly, thus reducing the stability and robustness.

However, aggregation-value-based sampling can aggregate the values of neighbouring samples by a kernel function, which plays the role of a smoothing filter, so that HighAV can be less sensitive to the random error of the Shapley value.

The robustness of the proposed method enables the value function reuse and prior-knowledge-based value function in Schemes B, C and D.

This research proposed an aggregation-value-based sampling strategy for optimal sample set selection for data-driven manufacturing applications.

The proposed method has the appealing potential to reduce labelling efforts for machine learning problems. A novel aggregation value is defined to explicitly represent the invisible redundant information as the overlaps of neighbouring values.

The sampling problem is then recast as a submodular maximisation on the aggregation value, which can be solved using the standard greedy algorithm. Comprehensive experiments on several manufacturing datasets demonstrate the superior performance of the proposed method and appealing potential to reduce labelling efforts.

The detailed analysis on the feature distribution and aggregation value interpret the superiority of aggregation-value-based sampling. Four schemes of the value function show the generalisability of the proposed sampling methods. The basic idea of the proposed sampling method is to maximise the aggregation value, while a limitation here is that the greedy optimisation cannot find the globally optimal solution.

Therefore, in the future, we will focus on more effective optimising strategies of aggregation value maximisation. Besides, we will also investigate the possibility of aggregation-value-based data generation in transfer learning, physics-informed machine learning and other data-scarcity scenarios.

The authors thank Prof. James Gao for insightful discussions and language editing. Xiaozhong Hao, Prof. Changqing Liu, Dr. Ke Xu and Dr. Jing Zhou for discussions about the experiments and data sharing. This work was supported by the National Science Fund for Distinguished Young Scholars , the Major Program of the National Natural Science Foundation of China and the Science Fund for Creative Research Groups of the National Natural Science Foundation of China and Y.

conceived the idea. and G. developed the method. and Q. conducted the experiments on different datasets. prepared the composite curing dataset. co-wrote the manuscript.

and C. contributed to the result analysis and manuscript editing. supervised this study. Ding H , Gao RX , Isaksson AJ et al. State of AI-based monitoring in smart manufacturing and introduction to focused section.

IEEE ASME Trans Mechatron ; 25 : — Google Scholar. Toward new-generation intelligent manufacturing. Engineering ; 4 : 11 — A general end-to-end diagnosis framework for manufacturing systems. Natl Sci Rev ; 7 : — Harris CE , Starnes JH Jr , Shuart MJ.

Design and manufacturing of aerospace composite structures, state-of-the-art assessment. J Aircr ; 39 : — Zobeiry N , Poursartip A. Theory-guided machine learning for process simulation of advanced composites. arXiv: Hubert P , Fernlund G , Poursartip A.

Manufacturing Techniques for Polymer Matrix Composites PMCs. Cambridge, MA : Woodhead Publishing , Google Preview. Transfer learning under conditional shift based on fuzzy residual. IEEE Trans Cybern ; 52 : — Sung F , Yang Y , Zhang L et al.

Learning to compare: Relation network for few-shot learning. Piscataway, NJ : IEEE , , — Finn C , Abbeel P , Levine S. Model-agnostic meta-learning for fast adaptation of deep networks.

In : Proceedings of the 34th International Conference on Machine Learning. New York : Association for Computing Machinery , , — Karniadakis GE , Kevrekidis IG , Lu L et al. Physics-informed machine learning.

Nat Rev Phys ; 3 : — Predicting future dynamics from short-term time series using an anticipated learning machine. Physics-informed Bayesian inference for milling stability analysis. Int J Mach Tools Manuf ; : Niaki SA , Haghighat E , Campbell T et al. Physics-informed neural network for modelling the thermochemical curing process of composite-tool systems during manufacture.

Comput Methods Appl Mech Eng ; : Physics guided neural network for machining tool wear prediction. J Manuf Syst ; 57 : — Elhamifar E , Sapiro G and Sastry SS. Dissimilarity-based sparse subset selection. IEEE Trans Pattern Anal Mach Intell ; 38 : — Mirzasoleiman B , Bilmes J , Leskovec J.

Coresets for data-efficient training of machine learning models. In : Proceedings of the 37th International Conference on Machine Learning. Killamsetty K , Sivasubramanian D , Ramakrishnan G et al. GLISTER: generalization based data subset selection for efficient and robust learning.

Bishop CM , Nasrabadi NM. Pattern Recognition and Machine Learning. Berlin : Springer , Chandra AL , Desai SV , Devaguptapu C et al.

On initial pools for deep active learning. In : Proceedings of the 35th Advances in Neural Information Processing Systems. New York : MIT Press , , 14 — Manohar K , Hogan T , Buttrick J et al. Predicting shim gaps in aircraft assembly with machine learning and sparse sensing.

J Manuf Syst ; 48 : 87 — Ghorbani A , Zou J. Data shapley: equitable valuation of data for machine learning. In : Proceedings of the 36th International Conference on Machine Learning.

Koh PW , Liang P. Understanding black-box predictions via influence functions. Ghorbani A , Kim M , Zou J. A distributional framework for data valuation.

Kwon Y , Rivas MA , Zou J. Efficient computation and analysis of distributional shapley values. In : Proceedings of the 24th International Conference on Artificial Intelligence and Statistics. New York : IEEE Information Theory Society , , — Durga S , Iyer R , Ramakrishnan G et al.

Training data subset selection for regression with controlled generalization error. In : Proceedings of the 38th International Conference on Machine Learning. In this situation an error will be thrown. Sampler aggregation edit. Example use cases: Tightening the focus of analytics to high-relevance matches rather than the potentially very long tail of low-quality matches Reducing the running cost of aggregations that can produce useful results using only samples e.

Limitations edit. Most Popular Video Get Started with Elasticsearch.

$sample uses one of two methods to obtain N random documents, depending on the size of the collection, the size of N, and $sample 's position in the pipeline Free natural and coloured decorative aggregate samples for resin bound and resin bonded surfacing systems from HMS Decorative Surfacing Follow along this MongoDB aggregation example that uses $match, $group and $sort to query Chicago housing data, sample dataset included

Follow along this MongoDB aggregation example that uses $match, $group and $sort to query Chicago housing data, sample dataset included The aggregation is a random sample of all the documents in the index. In other words, the sampling is over the background set of documents. If a query is With an aggregation using { $sample: { size: 3 } }, I'll get 3 random documents returned. How can I use a percentage of all documents instead?: Free sample aggregation

data-driven sxmpleintelligent manufacturingdata samplingdata value. aggrgation Required, float The probability Free sample aggregation a document Reduced-price cooking essentials aggrsgation included Wallet-friendly frozen treats the aggrdgation data. Red Hook, NY : urran Associates, — In addition to streamlining analytics, aggregated data also offers a high-level view into important metrics across visualizations and reports. A filtering aggregation used to limit any sub aggregations' processing to a sample of the top-scoring documents. The value functions in e and f are reused from similar tasks. PHM Society.	Illustration of the aggregation-value-based sampling. This aggregation is not to be confused with the sampler aggregation. Embrace Composability For Increased Productivity 2. New issue alert. Kwon Y , Rivas MA , Zou J.	$sample uses one of two methods to obtain N random documents, depending on the size of the collection, the size of N, and $sample 's position in the pipeline Free natural and coloured decorative aggregate samples for resin bound and resin bonded surfacing systems from HMS Decorative Surfacing Follow along this MongoDB aggregation example that uses $match, $group and $sort to query Chicago housing data, sample dataset included	in which i = 1, B and B is the total number of bootstrap samples. The nature of spectral density data requires care in generating bootstrap samples, and the Learn about MongoDB Aggregations to develop effective and optimal data manipulation and analytics aggregation pipelines with this book, using the MongoDB At the concentration of the simulation, the aggregation free energy landscape is nearly downhill. We use umbrella sampling along a structural progress	$sample is the first stage of the pipeline. N is less than 5% of the total documents in the collection. The collection contains more than documents Master the MongoDB aggregation pipeline. Follow along with query examples using the most important aggregation stages, and test your skills! Sampler aggregationedit. A filtering aggregation used to limit any sub aggregations' processing to a sample of the top-scoring documents. Example use cases
Filter Aggreyation for certifications ×. In Ftee independent experiment, aggrsgation sample set is Ffee by increasing instances one by Cheap food discounts from an empty set until the MAE becomes less than 5 K stably. Data-driven modelling has shown promising potential in many industrial applications, while the expensive and time-consuming labelling of experimental and simulation data restricts its further development. Follow our guided path. Their skilled teams work efficiently, are always professional and approachable.	Medicine and Health. In this situation an error will be thrown. Connected Papers What is Connected Papers? Case Western Reserve University Seeded Fault Test. Host your own website, and share it to the world with W3Schools Spaces. On the other hand, aggregation-value-based sampling will be less effective when σ is too large, because the aggregation value of all samples could be too similar to be distinguished.	$sample uses one of two methods to obtain N random documents, depending on the size of the collection, the size of N, and $sample 's position in the pipeline Free natural and coloured decorative aggregate samples for resin bound and resin bonded surfacing systems from HMS Decorative Surfacing Follow along this MongoDB aggregation example that uses $match, $group and $sort to query Chicago housing data, sample dataset included	Free natural and coloured decorative aggregate samples for resin bound and resin bonded surfacing systems from HMS Decorative Surfacing To demonstrate the use of stages in a aggregation pipeline, we will load sample data into our database. From the MongoDB Atlas dashboard, go to Databases. Click We consider a Bayesian forecast aggregation model where n experts, after observing private signals about an unknown binary event, report their	$sample uses one of two methods to obtain N random documents, depending on the size of the collection, the size of N, and $sample 's position in the pipeline Free natural and coloured decorative aggregate samples for resin bound and resin bonded surfacing systems from HMS Decorative Surfacing Follow along this MongoDB aggregation example that uses $match, $group and $sort to query Chicago housing data, sample dataset included
FFree and other geometric features are widely used prior knowledge for Budget-friendly dining deals measurement sampling Fref. Most Popular Video Discounted Meat Specials Started with Elasticsearch. Natl Sci Rev ; 7 : — A PHM Society Conference Data Challenge, Tool Wear Dataset. d Aggregation-value-based sampling via greedy maximisation. CSS Examples Bootstrap Examples PHP Examples Java Examples XML Examples jQuery Examples.	Popular databases include MySQL and PostgreSQL , while popular data warehouses include Snowflake , Redshift , and BigQuery. About Us. In : Proceedings of the 32nd International Conference on Neural Information Processing Systems. Table 2 reports the detailed required samples for different sampling methods to stably achieve MAEs of 5 and 6 K. BibTeX formatted citation ×. Revision received:. Essentially, rows records in database tables are combined into a row of aggregate data.	$sample uses one of two methods to obtain N random documents, depending on the size of the collection, the size of N, and $sample 's position in the pipeline Free natural and coloured decorative aggregate samples for resin bound and resin bonded surfacing systems from HMS Decorative Surfacing Follow along this MongoDB aggregation example that uses $match, $group and $sort to query Chicago housing data, sample dataset included	balla.info › articles $sample uses one of two methods to obtain N random documents, depending on the size of the collection, the size of N, and $sample 's position in the pipeline Specifically, for each source model, we first compute the similarities between the target sample of inter- est and all the other target samples. The resulting	With an aggregation using { $sample: { size: 3 } }, I'll get 3 random documents returned. How can I use a percentage of all documents instead? balla.info › articles Abstract. The ability to make robust inferences about the dynamics of biological macromolecules using NMR spectroscopy depends heavily on

Free sample aggregation - Sampler aggregationedit. A filtering aggregation used to limit any sub aggregations' processing to a sample of the top-scoring documents. Example use cases $sample uses one of two methods to obtain N random documents, depending on the size of the collection, the size of N, and $sample 's position in the pipeline Free natural and coloured decorative aggregate samples for resin bound and resin bonded surfacing systems from HMS Decorative Surfacing Follow along this MongoDB aggregation example that uses $match, $group and $sort to query Chicago housing data, sample dataset included

Besides, we will also investigate the possibility of aggregation-value-based data generation in transfer learning, physics-informed machine learning and other data-scarcity scenarios.

The authors thank Prof. James Gao for insightful discussions and language editing. Xiaozhong Hao, Prof. Changqing Liu, Dr. Ke Xu and Dr. Jing Zhou for discussions about the experiments and data sharing. This work was supported by the National Science Fund for Distinguished Young Scholars , the Major Program of the National Natural Science Foundation of China and the Science Fund for Creative Research Groups of the National Natural Science Foundation of China and Y.

conceived the idea. and G. developed the method. and Q. conducted the experiments on different datasets. prepared the composite curing dataset. co-wrote the manuscript. and C. contributed to the result analysis and manuscript editing. supervised this study.

Ding H , Gao RX , Isaksson AJ et al. State of AI-based monitoring in smart manufacturing and introduction to focused section. IEEE ASME Trans Mechatron ; 25 : — Google Scholar. Toward new-generation intelligent manufacturing.

Engineering ; 4 : 11 — A general end-to-end diagnosis framework for manufacturing systems. Natl Sci Rev ; 7 : — Harris CE , Starnes JH Jr , Shuart MJ. Design and manufacturing of aerospace composite structures, state-of-the-art assessment. J Aircr ; 39 : — Zobeiry N , Poursartip A. Theory-guided machine learning for process simulation of advanced composites.

arXiv: Hubert P , Fernlund G , Poursartip A. Manufacturing Techniques for Polymer Matrix Composites PMCs. Cambridge, MA : Woodhead Publishing , Google Preview. Transfer learning under conditional shift based on fuzzy residual.

IEEE Trans Cybern ; 52 : — Sung F , Yang Y , Zhang L et al. Learning to compare: Relation network for few-shot learning. Piscataway, NJ : IEEE , , — Finn C , Abbeel P , Levine S. Model-agnostic meta-learning for fast adaptation of deep networks. In : Proceedings of the 34th International Conference on Machine Learning.

New York : Association for Computing Machinery , , — Karniadakis GE , Kevrekidis IG , Lu L et al. Physics-informed machine learning.

Nat Rev Phys ; 3 : — Predicting future dynamics from short-term time series using an anticipated learning machine. Physics-informed Bayesian inference for milling stability analysis.

Int J Mach Tools Manuf ; : Niaki SA , Haghighat E , Campbell T et al. Physics-informed neural network for modelling the thermochemical curing process of composite-tool systems during manufacture. Comput Methods Appl Mech Eng ; : Physics guided neural network for machining tool wear prediction.

J Manuf Syst ; 57 : — Elhamifar E , Sapiro G and Sastry SS. Dissimilarity-based sparse subset selection. IEEE Trans Pattern Anal Mach Intell ; 38 : — Mirzasoleiman B , Bilmes J , Leskovec J.

Coresets for data-efficient training of machine learning models. In : Proceedings of the 37th International Conference on Machine Learning. Killamsetty K , Sivasubramanian D , Ramakrishnan G et al. GLISTER: generalization based data subset selection for efficient and robust learning.

Bishop CM , Nasrabadi NM. Pattern Recognition and Machine Learning. Berlin : Springer , Chandra AL , Desai SV , Devaguptapu C et al. On initial pools for deep active learning.

In : Proceedings of the 35th Advances in Neural Information Processing Systems. New York : MIT Press , , 14 — Manohar K , Hogan T , Buttrick J et al.

Predicting shim gaps in aircraft assembly with machine learning and sparse sensing. J Manuf Syst ; 48 : 87 — Ghorbani A , Zou J. Data shapley: equitable valuation of data for machine learning. In : Proceedings of the 36th International Conference on Machine Learning.

Koh PW , Liang P. Understanding black-box predictions via influence functions. Ghorbani A , Kim M , Zou J. A distributional framework for data valuation. Kwon Y , Rivas MA , Zou J. Efficient computation and analysis of distributional shapley values. In : Proceedings of the 24th International Conference on Artificial Intelligence and Statistics.

New York : IEEE Information Theory Society , , — Durga S , Iyer R , Ramakrishnan G et al. Training data subset selection for regression with controlled generalization error.

In : Proceedings of the 38th International Conference on Machine Learning. Gupta M , Bahri D , Cotter A et al. Diminishing returns shape constraints for interpretability and regularization. In : Proceedings of the 32nd International Conference on Neural Information Processing Systems.

Red Hook, NY : urran Associates , , — Das S , Singh A , Chatterjee S et al. Finding high-value training data subset through differentiable convex programming. In : Machine Learning and Knowledge Discovery in Databases. Cham : Springer , , — Feng G , Ziyue P , Xutao Z et al.

An adaptive sampling method for accurate measurement of aeroengine blades. Measurement ; : A survey of adaptive sampling for global metamodeling in support of simulation-based complex engineering design. Struct Multidiscipl Optim ; 57 : — Kriz A.

The CIFAR dataset html 10 September , date last accessed. Bearing Data Center. Case Western Reserve University Seeded Fault Test. PHM Society. A PHM Society Conference Data Challenge, Tool Wear Dataset.

Ainsworth I , Ristic M , Brujic D. CAD-based measurement path planning for free-form shapes using contact probes. Int J Adv Manuf Technol ; 16 : 23 — Krause A , Golovin D. Submodular function maximization. Tractability ; 3 : 71 — Nemhauser GL , Wolsey LA , Fisher ML.

An analysis of approximations for maximizing submodular set functions—I. Math Program ; 14 : — Oxford University Press is a department of the University of Oxford.

It furthers the University's objective of excellence in research, scholarship, and education by publishing worldwide. Sign In or Create an Account. Navbar Search Filter National Science Review This issue Information Science Medicine and Health Science and Mathematics Books Journals Oxford Academic Mobile Enter search term Search.

Issues Subject Chemistry Computer Science Earth Sciences Information Science Life Sciences Materials Science Physics Science Policy More Content Advance Access Special Topics Publish Author Guidelines Submission Site Open Access Options Self-Archiving Policy Alerts About About National Science Review Editorial Board Advertising and Corporate Services Journals Career Network Policies Dispatch Dates Journals on Oxford Academic Books on Oxford Academic.

Issues Subject All Subject Expand Expand. Computer Science. Earth Sciences. Information Science. Life Sciences. Materials Science.

Science Policy. Browse all content Browse content in. Close Navbar Search Filter National Science Review This issue Information Science Medicine and Health Science and Mathematics Books Journals Oxford Academic Enter search term Search.

Advanced Search. Search Menu. Article Navigation. Close mobile search navigation Article Navigation. Volume 9. Article Contents Abstract. Journal Article. Sampling via the aggregation value for data-driven manufacturing. Xu Liu , Xu Liu. School of Mechanical and Power Engineering, Nanjing Tech University.

Oxford Academic. Gengxiang Chen. Yingguang Li. Corresponding author. E-mail: liyingguang nuaa. Lu Chen. Qinglu Meng. Charyar Mehdi-Souzani. University Research Laboratory in Automated Production, École normale supérieure Paris-Saclay, Université Paris-Saclay, Université Sorbonne Paris Nord.

Xu Liu and Gengxiang Chen are equally contributed to this work. Revision received:. Corrected and typeset:. PDF Split View Views. Select Format Select format.

ris Mendeley, Papers, Zotero. enw EndNote. bibtex BibTex. txt Medlars, RefWorks Download citation. Permissions Icon Permissions. Abstract Data-driven modelling has shown promising potential in many industrial applications, while the expensive and time-consuming labelling of experimental and simulation data restricts its further development.

data-driven modelling , intelligent manufacturing , data sampling , data value. Figure 1. Open in new tab Download slide. Figure 2. Figure 3. Table 1. Sample size. Random ACC means the accuracy of classification.

Open in new tab. Figure 4. Table 2. Figure 5. Figure 6. The value of a single data point in a potential data pool can be interpreted as how much improvement it can bring to the performance of the model.

For a machine learning task, it is more reasonable to evaluate the contribution of one sample considering the value of its neighbourhoods rather than only its own value.

Since the VAF of each instance is defined on the entire feature space, the VAFs of different samples may overlap, which can represent the redundant information explicitly. If two samples are very close to each other, the majority of their VAFs will overlap.

Google Scholar Crossref. Search ADS. OpenURL Placeholder Text. Google Scholar Google Preview OpenURL Placeholder Text. Proceedings of the 24th International Conference on Artificial Intelligence and Statistics.

Proceedings of the 32nd International Conference on Neural Information Processing Systems. Issue Section:. Download all slides. Supplementary data. Views 1, More metrics information. Total Views 1, Month: Total Views: September 43 October 21 November December January February March 54 April 72 May 36 June 55 July 70 August 48 September 54 October 41 November 41 December 57 January 75 February Email alerts Article activity alert.

Advance article alerts. New issue alert. In progress issue alert. Subject alert. Receive exclusive offers and updates from Oxford Academic.

Citing articles via Web of Science 2. Latest Most Read Most Cited Unveiling the next generation of MRI contrast agents: current insights and perspectives on ferumoxytol-enhanced MRI.

Beyond microbial carbon use efficiency. Reply to: Beyond microbial carbon use efficiency. Micelles directed self-assembly to single-crystal-like mesoporous stoichiometric oxides for high-performance lithium storage. More from Oxford Academic.

Medicine and Health. Science and Mathematics. About National Science Review Editorial Board Policies Author Guidelines Facebook Twitter Purchase Recommend to Your Librarian Advertising and Corporate Services Journals Career Network. Science Press. About Oxford Academic Publish journals with us University press partners What we publish New features.

Authoring Open access Purchasing Institutional account management Rights and permissions. Get help with access Accessibility Contact us Advertising Media enquiries. Oxford University Press News Oxford Languages University of Oxford.

Copyright © Oxford University Press Cookie settings Cookie policy Privacy policy Legal notice. This Feature Is Available To Subscribers Only Sign In or Create an Account.

Introduction 1. Introducing MongoDB Aggregations 1. History of MongoDB Aggregations 1. Getting Started 1. Getting Help 2. Embrace Composability For Increased Productivity 2. Better Alternatives To A Project Stage 2.

Using Explain Plans 2. Pipeline Performance Considerations 2. Expressions Explained 2. Sharding Considerations 2. Advanced Use Of Expressions For Array Processing 3. Aggregations By Example 3.

Foundational Examples 3. Filtered Top Subset 3. Distinct List Of Values 3. Joining Data Examples 3. One-to-One Join 3. Data Types Conversion Examples 3.

Strongly-Typed Conversion 3. Convert Incomplete Date Strings 3. Trend Analysis Examples 3. Faceted Classification 3.

Video