Thank you for your outstanding work on this project. I have added an ICASSP paper that I believe could further enrich it.
The paper proposes a training algorithm called LRFR which optimizes network parameters in the null space of low-rank past tasks' feature representation matrix. LRFR outperforms other null space algorithms for continual learning, and the code is publicly available.
Thank you for your outstanding work on this project. I have added an ICASSP paper that I believe could further enrich it.
The paper proposes a training algorithm called LRFR which optimizes network parameters in the null space of low-rank past tasks' feature representation matrix. LRFR outperforms other null space algorithms for continual learning, and the code is publicly available.