The artificial intelligence (AI) is rapidly applied throughout the industry. But there is still an obstacle that interferes with AI introduction.
Typical difficulties are more than a skilled AI expert for building AI models. To build AI models, it is necessary to interpret the algorithm and require high mathematical knowledge to handle vast data precisely.
Most AIs do not disclose the resulting derivation process. It means that it is difficult to find the cause when an error occurred. As a result, the financial and medical fields are cautious to introduce AI.
MassWorks Korea Jang Ki-hwan was introduced to automated machine learning (Auto ML) and analytical ML using matter as a measure to solve this problem.
Jang Kwan-hwan, a recent Matt Matrap Machine Leaving Day 2021 Werp Machine Leading Day 2021 Wheelnati, introduced a method of implementing Auto ML and analytical ML.
■ Otto ML, non-professionals develop machine learning model
Auto ML is a fully automated machine learning model development process.
If you select an algorithm that is most suitable for data analysis, you can remove the repeated experimentation steps for selecting the optimal modeling technique to select the optimal modeling technique to help you optimize and deploy the model immediately.
The Matter optimized for the Auto ML implementation is a machine-based development process ▲ Feature engineering ▲ Modeling ▲ The hyperper parameter setting is automated to minimize resource waste during the development process.
Feature engineering provides an environment that easily supports a variety of feature engineering techniques, including a feature selection, or feature extraction, including wavelet conversion, high-speed Fourier transformation.
Modeling uses only one of the commands to help you automatically select a high-performance model that is appropriate for data in the middle of the many machine learning modeling techniques.
Auto ML can automate multiple stages that are difficult in workflow, and Automatically simplifies a lot of time to work in the workflow, said Jang Kwan-hwan, explained that domain experts can also implement optimized models without the help of AI experts..
The hyper parameter setting supports high-performance model development by automatically optimizing and applying hyper parameter values to various models in a mat lab environment. Compared to a manual process for optimization of hyper parameter, there is an advantage that the waste is removed and ensuring performance optimization.
As a result of the internal testing of MassWorks, the performance of human behavior and detection models developed by Mat Lap Otto ML and the performance of the cardiac abnormal detection model was highly compare with the performance of the two models developed by hand.
In addition, MathWorks supports augmented learning that supports learning and model updates to the amount of data that is updated from R2020b.
■ Remove black box phenomenon with ML explanatory possible
Currently, AI is a black box that can not explain the result value derivation process. This problem is also a fairness and ethical deflection issue.
Description ML is a technique that is presented to solve reliability problems of the Black Box AI model. It supports a description of the algorithm based on algorithm for specific models, and the overall data for the universal model and the description of the specific data range.
MassWorks assists to determine the set of features that plays a feature set that serves as an important role in the result of the model through various algorithm support, such as pitch selection and shapel values.
The feature selection is to extract only key features that affect the model and support the development of possible models that have excellent performance, while the scale is small and small.
Through a method of identifying the classification process, it is possible to find a feature that has a significant impact on the decision of the model and explain the classification of the model of the model.
Generally, the higher the predicted power, the lower the likelihood of explaining, and the low predictive model is likely to be explained. Explanation In the case of a high prerequisite, the specific data range can also implement the probability of the specific data and improve performance.
The shapel value is a method of applying game theory to determine the factors affecting the experimental results.
By applying a shapel value, it is possible to confirm that the element according to the importance is displayed as a graph and that any data affected AI.
The game theory is an analytical theory that distributes the benefits created through collaboration between participants of the game according to the contribution of each participant.
The Matter can be able to see if there is an error in analyzing the image as an interpretable ML.
Comments
Post a Comment