I am trying to ascertain if my neural network is able to generalize or if it’s simply using memory/overfitting to solve a task. I would like my model to generalise.

Currently, I train the neural network on a randomly generated 3x3 frozen lake environment - with no holes. (The network simply chooses an action for each state it is presented.)

Then, I test the model on a much larger frozen lake environment. Still no holes. Still randomly generated. The test environment size is assigned by a random value of 5-15 for each axis (height/width), randomly generated.

Then I determine the "degree of generalization" by how many large environments the network is able to solve. At present, it solves 100/100 on the 3x3, and about 83/100 on the larger test environments.

When I track the solutions it generates, I can see that the network always takes the shortest route available, which is great.

Do you guys have any ideas, inputs or criticism on the method I use to determine the degree of generalization?

  • $\begingroup$ Do you have any working definition of "abstract thinking" that you want answers to apply? It is a hard concept to define at the level of maths and statistics where you are measuring it. Wiithout a working definition from you, answers will tend to be more philosophical than practical $\endgroup$ – Neil Slater Jul 12 '19 at 8:28
  • $\begingroup$ Hi Neil! No i don’t. I just want to avoid overfitting/memorisation, and want the network to be able to generalise and apply to similar environments as described. Does this make sense? $\endgroup$ – skywalkerdk Jul 12 '19 at 10:11
  • $\begingroup$ Yes. I was thinking that really you are asking about generalisation and not about "abstract thought" at all. Measuring generalisation is a very practical thing that you can do, and some discussion can be had about limits of this. "Abstract thought" - in the sense that it is used in natural English - is arguably not happening at all here, your agent has learned a kind of reflex behaviour a bit like riding a bicycle or running away from danger. $\endgroup$ – Neil Slater Jul 12 '19 at 10:28
  • $\begingroup$ Thanks for your input Neil. English is not my native language, so i will edit the post to reflect the correct meaning. - Do you happen to have any input on the issue at hand? $\endgroup$ – skywalkerdk Jul 12 '19 at 10:31
  • $\begingroup$ Yes, I do have some thoughts. I may get time to post an answer later today $\endgroup$ – Neil Slater Jul 12 '19 at 10:36

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.