OrderNet: Ordering by Example
This addresses a fundamental ordering problem in machine learning, with potential applications in optimization and natural language processing, though it appears incremental as it builds on existing supervised techniques.
The paper tackles the problem of sorting unordered sequences where the correct order must be inferred from data, introducing OrderNet, a neural architecture that outperforms previous supervised techniques in generalizing to longer sequences for tasks like the Traveling Salesman Problem.
In this paper we introduce a new neural architecture for sorting unordered sequences where the correct sequence order is not easily defined but must rather be inferred from training data. We refer to this architecture as OrderNet and describe how it was constructed to be naturally permutation equivariant while still allowing for rich interactions of elements of the input set. We evaluate the capabilities of our architecture by training it to approximate solutions for the Traveling Salesman Problem and find that it outperforms previously studied supervised techniques in its ability to generalize to longer sequences than it was trained with. We further demonstrate the capability by reconstructing the order of sentences with scrambled word order.