Transforming healthcare with machine learning
-
M. Mahalakshmi
Abstract
Deep learning (DL) modelsdeep learning models in material science revolutionize the design and characterization of materials, but due to their complexity and the fact that no one knows exactly how the model works, they can be hard to understand. In this chapter, we introduce a critical issue of the current DL frameworks: interpretability, with a focus on post- and intermediate processing techniques that enhance the readability of the models’ predictions. The relationship between the input and the output is sometimes unclear due to the high complexity of DL methods, which require the use of high-dimensional nonlinear spaces, presenting one of the main challenges. All of these are separately discussed in the chapter: an interpretable model element, visualization methods, and local explanation methods such as Shapley values and local interpretable model-agnostic explanations (LIME). Thirdly, it compares factors such as model accuracy and interpretability, among others, because in each case, it is impossible to achieve the maximum increase in performance by introducing more complex and sophisticated models without sacrificing the interpretability of the results for end-users. Perspectives for further studies are described, with an emphasis on the ability to combine domain knowledge and create machine learningmachine learning models with domain-oriented interpretability protocols for material science fields. This proposed work should be able to close the gap between abstract DL schemes and tangible materials science applications, thus encouraging the general acceptance of AI technologiesAI technologies in materials science and engineering research and future application development.
Abstract
Deep learning (DL) modelsdeep learning models in material science revolutionize the design and characterization of materials, but due to their complexity and the fact that no one knows exactly how the model works, they can be hard to understand. In this chapter, we introduce a critical issue of the current DL frameworks: interpretability, with a focus on post- and intermediate processing techniques that enhance the readability of the models’ predictions. The relationship between the input and the output is sometimes unclear due to the high complexity of DL methods, which require the use of high-dimensional nonlinear spaces, presenting one of the main challenges. All of these are separately discussed in the chapter: an interpretable model element, visualization methods, and local explanation methods such as Shapley values and local interpretable model-agnostic explanations (LIME). Thirdly, it compares factors such as model accuracy and interpretability, among others, because in each case, it is impossible to achieve the maximum increase in performance by introducing more complex and sophisticated models without sacrificing the interpretability of the results for end-users. Perspectives for further studies are described, with an emphasis on the ability to combine domain knowledge and create machine learningmachine learning models with domain-oriented interpretability protocols for material science fields. This proposed work should be able to close the gap between abstract DL schemes and tangible materials science applications, thus encouraging the general acceptance of AI technologiesAI technologies in materials science and engineering research and future application development.
Kapitel in diesem Buch
- Frontmatter I
- Contents V
- List of contributors VII
- Blockchain technology to secure medical data sharing in machine learning applications ensure privacy and integrity 1
- AI-powered sensors and devices for sustained health tracking 39
- Development of AI-driven biomedical sensors and devices optimization for continuous health monitoring 89
- Design and development of AI-driven biomedical sensors and devices and their optimization for continuous health monitoring 131
- Machine learning-driven personalized medicine: customized drug delivery systems and patient-specific material applications 193
- Personalized medicine using customized drug delivery systems and patient-specific material solutions, enabled by machine learning algorithms 239
- AI-driven drug design exploring molecular docking and lead optimization using machine learning algorithms 297
- Machine learning models for predicting drug toxicity and side effects 335
- Machine learning innovations in biomedical materials from drug discovery to personalized medicine 395
- High-throughput screening for novel medical materials: machine learning-enabled approaches 445
- Automated materials characterization using machine learning for screening biocompatible materials 489
- Machine learning algorithms for enhanced medical image analysis and diagnostics 541
- Transforming healthcare with machine learning 585
- Revolutionizing healthcare 635
- Index 687
- De Gruyter Series in Advanced Mechanical Engineering
Kapitel in diesem Buch
- Frontmatter I
- Contents V
- List of contributors VII
- Blockchain technology to secure medical data sharing in machine learning applications ensure privacy and integrity 1
- AI-powered sensors and devices for sustained health tracking 39
- Development of AI-driven biomedical sensors and devices optimization for continuous health monitoring 89
- Design and development of AI-driven biomedical sensors and devices and their optimization for continuous health monitoring 131
- Machine learning-driven personalized medicine: customized drug delivery systems and patient-specific material applications 193
- Personalized medicine using customized drug delivery systems and patient-specific material solutions, enabled by machine learning algorithms 239
- AI-driven drug design exploring molecular docking and lead optimization using machine learning algorithms 297
- Machine learning models for predicting drug toxicity and side effects 335
- Machine learning innovations in biomedical materials from drug discovery to personalized medicine 395
- High-throughput screening for novel medical materials: machine learning-enabled approaches 445
- Automated materials characterization using machine learning for screening biocompatible materials 489
- Machine learning algorithms for enhanced medical image analysis and diagnostics 541
- Transforming healthcare with machine learning 585
- Revolutionizing healthcare 635
- Index 687
- De Gruyter Series in Advanced Mechanical Engineering
