Reusing Neural Networks: Lessons learned and Suggestions for the future

<!--HTML-->I present the lessons learned as re-interpreters trying to reuse analyses centred on neural networks in the RIVET framework, using two recent ATLAS analyses -- SUSY and Exotics searches -- as examples. I survey the possible ways that an analysis team can preserve and publicise their neural network for future use, and provide a detailed examination of the ONNX and lwtnn preservation tools, describing their advantages and disadvantages for both the original analysis team and re-interpreters. I also comment on how thinking about re-use from the beginning could change how analyses design and use neural networks; and what supplementary data becomes even more important for validation.