Incremental Learning
- 0 Collaborators
This research mainly concerns with addressing the issues which lie in retaining information in neural nets when trained on multiple tasks (more than 2). This research aims to solve the problem of overcatastrophic forgetting in neural nets and as well as to adapt to more tasks which may have varied domain shifts. There are many problems that lie in doing considerably well in multiple tasks. This work address those issues. ...learn more
Project status: Under Development
Groups
Student Developers for AI
Intel Technologies
AI DevCloud / Xeon,
MKL,
Intel Opt ML/DL Framework,
Intel Python
Overview / Usage
The following research is being performed for publication, and as of now we are trying with two tasks. In both the tasks, State of the art results have been achieved but still facing issues with domain adaptation. Both the sources of dataset are completely different. These issue arise due to varied reasons like orientation, lighting conditions, camera types, background and much more. Later on this can further will be extended to different domains of computer vision. The source code will be released once the paper gets accepted.
Methodology / Approach
We are using a custom architecture. Details would be released soon.
Technologies Used
Pytorch, Intel DevCloud, Intel MKL, Intel optimized Python