HIGHLIGHTS
- who: Kirsten Fischer from the Institut, Ju00fclich, Germany Aachen University have published the Article: Decomposing neural networks as mappings of correlation functions, in the Journal: (JOURNAL) of 28/Nov/2022
- what: The authors characterize this mapping as an iterated transformation of distributions where the nonlinearity in each transfers information between different orders of correlation functions. Applied to an XOR task and to MNIST the authors show that correlations up to second order predominantly capture the information processing in the internal layers while the input also extracts higher-order correlations from the data. This analysis provides . . .
If you want to have access to all the content you need to log in!
Thanks :)
If you don't have an account, you can create one here.