zhangyuyi99
commented
2 years ago Finn, C., Abbeel, P., & Levine, S. (2017, July). Model-agnostic meta-learning for fast adaptation of deep networks. In International Conference on Machine Learning (pp. 1126-1135). PMLR. https://arxiv.org/pdf/1703.03400.pdf
Duan, Y., Andrychowicz, M., Stadie, B. C., Ho, J., Schneider, J., Sutskever, I., ... & Zaremba, W. (2017). One-shot imitation learning. arXiv preprint arXiv:1703.07326. https://sites.google.com/view/nips2017-one-shot-imitation/home https://arxiv.org/pdf/1703.07326.pdf
Key sentences The first problem is that of dexterity: robots should learn how to approach, grasp and pick up complex objects, and how to place or arrange them into a desired configuration. The second problem is that of communication: how to communicate the intent of the task at hand, so that the robot can replicate it in a broader set of initial conditions.
We made essential use of soft attention [6] for processing both the (potentially long) sequence of states and action that correspond to the demonstration, and for processing the components of the vector specifying the locations of the various blocks in our environment. The use of soft attention over both types of inputs made strong generalization possible.
Comments Soft attention mechanism should be used in piano playing. For the current MIDI we have the maximum attention. For previous and past MIDI we have some attention. For MIDI further away no attention is required. Depending on the playing speed and complexity of the music, the attention function may have changing shape.
My presentation slides on the use of imitation learning on piano playing robot for PhD interview QAQ https://docs.google.com/presentation/d/140rJhZfdgkvqXBR85hKvRVFT3WUQv_sv_0qqkXi-kqc/edit?usp=sharing