Difference between revisions of "Transfer Learning by Selection of Invariant Features"
(Created page with "{{StudentProjectTemplate |Summary=The project aims to develop novel methods to identify invariant features to transfer across multiple domains. |Keywords=Transfer Learning, F...") |
|||
| Line 1: | Line 1: | ||
{{StudentProjectTemplate | {{StudentProjectTemplate | ||
| − | |Summary=The project aims to develop novel methods to identify invariant features to transfer across multiple domains. | + | |Summary=The project aims to develop novel methods to identify invariant features to transfer across multiple domains. |
|Keywords=Transfer Learning, Feature Selection | |Keywords=Transfer Learning, Feature Selection | ||
|Supervisor=Mohammed Ghaith Altarabichi, Abdallah Alabdallah | |Supervisor=Mohammed Ghaith Altarabichi, Abdallah Alabdallah | ||
| + | |Level=Master | ||
| + | |Status=Open | ||
}} | }} | ||
Machine learning models often face a significant challenge in dynamically evolving environments. The conditions under which the model was trained (source domain) often vary from the testing conditions (the field conditions, or target domain). The change in conditions is often associated with a change in the conditional distribution of the target variable given some subset of covariates. A conventional feature selection method unaware of the change in distribution would fail in identifying predictive features in the target domain. | Machine learning models often face a significant challenge in dynamically evolving environments. The conditions under which the model was trained (source domain) often vary from the testing conditions (the field conditions, or target domain). The change in conditions is often associated with a change in the conditional distribution of the target variable given some subset of covariates. A conventional feature selection method unaware of the change in distribution would fail in identifying predictive features in the target domain. | ||
As a motivating example, we refer to our early work with modeling SoH of Li-Ion drive batteries. Our analysis showed that the deterioration processes of batteries in hybrid buses could vary significantly for different bus configurations and operating conditions and that many features were not useful (even harmful, leading to negative transfer) to transfer across different settings (e.g., different batteries). Therefore, we are looking to explore methods for selecting features that can be beneficial to transfer from the source domain (training setting) to the target domain (test setting). This project aims to develop a novel transfer learning method to select invariant features to transfer across multiple source domains | As a motivating example, we refer to our early work with modeling SoH of Li-Ion drive batteries. Our analysis showed that the deterioration processes of batteries in hybrid buses could vary significantly for different bus configurations and operating conditions and that many features were not useful (even harmful, leading to negative transfer) to transfer across different settings (e.g., different batteries). Therefore, we are looking to explore methods for selecting features that can be beneficial to transfer from the source domain (training setting) to the target domain (test setting). This project aims to develop a novel transfer learning method to select invariant features to transfer across multiple source domains | ||
Latest revision as of 12:25, 15 October 2020
| Title | Transfer Learning by Selection of Invariant Features |
|---|---|
| Summary | The project aims to develop novel methods to identify invariant features to transfer across multiple domains. |
| Keywords | Transfer Learning, Feature Selection |
| TimeFrame | |
| References | |
| Prerequisites | |
| Author | |
| Supervisor | Mohammed Ghaith Altarabichi, Abdallah Alabdallah |
| Level | Master |
| Status | Open |
Machine learning models often face a significant challenge in dynamically evolving environments. The conditions under which the model was trained (source domain) often vary from the testing conditions (the field conditions, or target domain). The change in conditions is often associated with a change in the conditional distribution of the target variable given some subset of covariates. A conventional feature selection method unaware of the change in distribution would fail in identifying predictive features in the target domain.
As a motivating example, we refer to our early work with modeling SoH of Li-Ion drive batteries. Our analysis showed that the deterioration processes of batteries in hybrid buses could vary significantly for different bus configurations and operating conditions and that many features were not useful (even harmful, leading to negative transfer) to transfer across different settings (e.g., different batteries). Therefore, we are looking to explore methods for selecting features that can be beneficial to transfer from the source domain (training setting) to the target domain (test setting). This project aims to develop a novel transfer learning method to select invariant features to transfer across multiple source domains
