Continual Learning for Robotics
?
1 like?469 views
The document describes a deep learning model that uses experience replay to improve classification performance. The model contains a data layer, output layer, and hidden layers that extract low-level and class-specific features. It performs a forward pass on native patterns and stores rehearsal patterns externally, then does a backward pass on all patterns to update weights and prevent forgetting older information.
Continual Learning for Robotics
- 2. ? ? ? ? ? ?
- 3. ? ? ?
- 9. ? ?
- 17. ?
- 33. Data layer Output layer (classes) Low-level generic features Class specific discriminative features
- 34. Data layer Output layer (classes) Low-level generic features Class specific discriminative features
- 35. Data layer Output layer (classes) Low-level generic features Class specific discriminative features
- 36. Data layer Output layer (classes) Low-level generic features Class specific discriminative features
- 37. Data layer Output layer (classes) Low-level generic features Class specific discriminative features
- 41. Data layer Output layer (classes) Low-level generic features Class specific discriminative features Backwardpass (allpatterns) Forwardpass (allpatterns) Forwardpass (nativepatterns) Concat External storage (rehearsal patterns) (at minibatch level)
- 43. 3.