Concluding Remarks

The basic idea of implementing curiosity and boredom is not limited to the particular algorithm described in the first section. Every model-dependent on-line algorithm for learning goal directed behavior might be augmented by a similar implementation of `the desire to improve the world model'. The basic motivation is: Instead of using some separate mechanism for improving the world model, we want to make use of the capabilities of the goal-directed learning algorithm itself.

The interesting side effect is: Since the learning algorithm depends on the model network, the model network has to make a prediction about its own current prediction capabilities. The activations of the model network are (partly) interpreted as a statement about the current weights of the model network. Note that this is already a rudimentary form of self-introspective behavior! The author believes that extensions of these rudimentary forms of introspective neural algorithms will be the key to learning systems which are much more sophisticated than the ones we know so far.

Back to Active Learning - Exploration - Curiosity page