图灵学习实现步骤的理解
Turing Learning Implementation steps:
- Training data. Genuine data samples from actual hardware (e.g. data comes from experiment with e-puck).
- Model representation. In any form, while neural networks are applied widely. Objective is to produce data (good or bad).
Input: training data.
Output: new data needs to be inferred (e.g. Velocities of two wheels are 10cm/s in actual hardware. the same data (10cm/s for both wheels) as real object. which means good model, and different data (10cm/s for left wheel while 5cm/s for right wheel) which means bad model.). - Discriminator representation. In any form, neural network is a great choice. Objective is to distinguish between model and real object.
Input: training data and model data