| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 200.68 KB | Adobe PDF |
Advisor(s)
Abstract(s)
The field of robotics is becoming continuously more important, due to
the impact it can bring to our everyday life. A long standing problem
with neural network learning is the catastrophic forgetting when one tries
to use the same network to learn more than one task. In this paper we
present results of the application of a method to avoid catastrophic forgetting
while using Convolutional Neural Networks (CNNs) to some visual
recognition tasks relevant to the field of robotics. The results show that
with this method a robot can learn new tasks without forgetting the previous
learned tasks. Results also showed that if we applied this method, the
performance on isolated tasks increases and it is better to use it than train
a CNN in an isolated way (single task). We use for our experiments two
well known data sets, namely, Olivetti Faces and Fashion-MNIST.