5-Use different datasets

Annotations:

•   Assume that we are given some labelled data points in a source domain. The goal of transfer learning is to reduce the number of labelled data needed in the target domain. The source and target domains are related but not identical. It learns and transfers a model based on the labelled data from the source domain and unlabelled data from the target domain \cite{wang2014active}. There are two types of transfer learning: 1) static transfer learning 2) active transfer learning \cite{wang2014active}. \begin{itemize} \item Static transfer learning: the goal is to learn a predictive model using all the given data according to a certain performance metric \item active transfer learning: The performance metric is the same but the active transfer learning algorithm chooses the points rather than being given a randomly chosen set. \end{itemize}

1. 7-Use pose estimatin dataset to improve action recognition
2. 8-Transfer learning so far has worked based on retraining a network on the target task. Now we want to solve how to combine networks trained with different datastets to get the best performance? -> emkan sanji
3. 9-How to recognize datasets with negative transfer effects
   1. 11-Definition for negative knowledge transfer
   2. 12-Relational knowledge transfer
4. 10-Can we exploit sub architectures of networks trained with other datasets to both improve accuracy and computational efficiency?
5. 1-What is knowledge transfer?
6. 2-Benefits of knowledge transfer
7. 3-Types of knowledge transfer: different domains-different models-different targets-different datsets
8. 6-Use diferent action recognition dataset
9. 4-Literature Survey
10. Literature Survey
11. Literature survey
12. Active learning
13. Literature
14. Use Object Recognition in transfer learning