What recurrent networks can be applied to groups of related data?

Question

Neural networks

There is a set of labeled data with many features ( a, b, c, d ... n ). The set is divided into groups with a variable number of samples, where the samples are interconnected by dependencies unknown to us.
It is necessary to build a network that will display these dependencies in groups and learn to give the correct result. That is, the result of the sample must additionally be influenced by the features of all other samples of the group.
The number of samples in a group is variable, so it cannot be solved with a normal FNN. The problem is probably solved by LSTM, where, by analogy with the time series, the network must rely not only on past, but also on future data. Is there some sort of recurrent network that covers this problem? Or techniques/techniques that will help?

Answer 1

It does not follow from the task description that these groups represent a sequence of data.
If groups are just unordered sets, then you can try to sum their features in a cycle through attention