Continual learning and data analysis of time series data
Time-series data finds wide usage among various field, making it a common and useful type of data. However, Catastrophic forgetting arise when model is optimizing for the new task, which in real-life case can be different data distribution. While most of the continual learning strategies focus...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project |
| Language: | English |
| Published: |
Nanyang Technological University
2023
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/167635 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | Time-series data finds wide usage among various field, making it a common and useful
type of data. However, Catastrophic forgetting arise when model is optimizing for the
new task, which in real-life case can be different data distribution.
While most of the continual learning strategies focus on static data, this project aims
to explore the feasibility of the most used strategy Elastic Weight Container (EWC) on
time-series data.
This project uses a Human Activity Recognition problem as a benchmark. EWC is then
implemented on the model to compare the loss of accuracy with and without it. The
result shows the extend of continual learning EWC can achieve on time series data. |
|---|