| Year-Phase | Objectives | Activities | Expected results | Achieved results |
|---|---|---|---|---|
| 2020 - phase 1: | Prepare transfer and semi-supervised learning algorithms | Identification of relevant results published from project submission | Research report (phase 1) |
[Report 2020] (romanian) Articles 2,3 |
| Identification of the appropriate databases | ||||
| Initialize knowledge transfer construction through dual training (prediction and grouping) | ||||
| Dissemination and report | ||||
| 2021 - phase 2: | Transfer learning algorithms | Baseline: Supervised learning | Research report (phase 2) 4 conference articles submitted/accepted, 1 article journal submitted/accepted |
[Report 2021 ] (romanian) Articles published: 4,5,6 and submitted R1,R2 from [Publications] |
| Transfer learning by combined clustering and prediction | ||||
| Semisupervised learning | Baseline: semisupervised learning | |||
| Self labelling techniques | ||||
| Self identification of noisy labels | Basic techniques of identification of noisy labels | |||
| Weight alteration by randomization | ||||
| Randomization injection | Database bias estimation | |||
| Weight alteration by randomization | ||||
| 2022 - phase 3: | Aesthetic Image Assesment | Aesthetic solution | Research report (phase 3) 5 conference articles submitted/accepted , 2 articles journal submitted/accepted |
[Report Final] (romanian) Journal articles published: 7 and submitted R3, R4, R5 [Publications] |
| Evaluation, conclusions | ||||
| Transfer Learning algorithms | Combined solution | |||
| Conclusion: efficiency applicability |
