IEEE Robotics & Automation Magazine - June 2020 - 15

●

●

●

Members: Zhengwei Wang, Eoin
Brophy, and Tomás E. Ward
Affiliations: Wang: V-SENSE, School
of Computer Science and Statistics,
Trinity College, Dublin, Ireland; Bro-
phy and Ward: Insight Center for
Data Analytics, School of Comput-
ing, Dublin City University, Ireland
Method: The core backend of the method
was LwF [7]. There was no replay of pre-
vious task images in this structure.

Neverforget Team
● Title: A Small Step to Remember:
Study of Single Model Versus Dy-
namic Model
● Members: Liguang Zhou, Tin Lun
Lam, and Yangsheng Xu
● Affiliation: The Chinese University of
Hong Kong, Shenzhen, China, and
Shenzhen Institute of Artificial Intel-
ligence and Robotics for Society,
China
● Method: This approach was based on
EWC [8] without a replay mechanism.
The team also found the fact that the
estimation of the Fisher information
matrix might be biasedly estimated.

created the replay memory for a task.
The replay memory was an efficient
representation of previous task data,
whose information was lost and
sampled from the validation set.
HYDRA-DI-ETRI Team
● Title: Selective Feature Learning with
Filtering Out Noisy Objects in Back-
ground Images
● Members: Soonyong Song, Heechul
Bae, Hyonyoung Han, and Young-
sung Son
● Affiliation: Electronics and Telecom-
munications Research Institute,
Korea
● Method: The team proposed a selec-
tive feature learning method to elim-
inate irrelevant objects in target
images. A single-shot multibox de-
tection (SSD) algorithm selected the
desired objects [23]. The SSD algo-
rithm alleviated performance degra-
dation by noisy objects. Then SSD

weights were trained with anno-
tated images in task 1 and the
refined data were fed into a classifi-
cation module.
NTU_LL Team
● Title: Lifelong Learning with
Regularization and Data Aug-
mentation
● Members: Duvindu Piyasena, Sathur-
san Kanagarajah, Siew-Kei Lam, and
Meiqing Wu
● Affiliation: Nanyang Technological
University, Singapore
● Method: The team utilized a combi-
nation of an SI-based regularization
method [9] and data augmentation
for each task.
Acknowledgments
This work was partially supported by a
grant from the Research Grants Council
of the Hong Kong Special Administrative
Region, China, under project CityU

SDU_BFA_PKU Team
● Title: SDKD: Saliency Detection with
Knowledge Distillation
● Members: Lin Yang and Baoquan
Chen
● Affiliation: Peking University, Beijing,
China; Shandong University, Qing-
dao, China; and Beijing Film Acad-
emy, Beijing, China
● Method: The approach disentangled
this problem with two aspects: back-
ground removal problem and classifica-
tion problem. The entrant used saliency
maps to implement background re-
moval and knowledge distillation to
address catastrophic forgetting.
Vidit98 Team
● Title: Intelligent Replay Sampling for
Lifelong Object Recognition
● Members: Vidit Goel, Debdoot Sheet,
and Somesh Kumar
● Affiliation: Indian Institute of Tech-
nology, Kharagpur, India
● Method: This approach sampled vali-
dation data from the buffer and used
them as replay data. It intelligently
JUNE 2020

*

IEEE ROBOTICS & AUTOMATION MAGAZINE

*

15
http://butterflyhaptics.com
IEEE Robotics & Automation Magazine - June 2020

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - June 2020
