Wildest Dreams: Reproducible Research in Privacy-preserving Neural Network Training

The 'Wildest Dreams' project is primarily a research artifact accompanying an academic paper (arXiv:2403.03592). It scores low on defensibility (2) because it functions as a benchmarking suite rather than a production-ready tool, evidenced by its 0 stars and lack of community engagement over a 2-year period despite recent paper activity. Its primary value is diagnostic—identifying gaps in current PPNNT reproducibility—rather than providing a novel moat-building technology. Frontier labs and major cloud providers (Google via TensorFlow Privacy, Meta via Opacus, Microsoft via Seal/SEAL-Python) already dominate the privacy-preserving ML landscape with production-grade libraries. The risk of platform domination is high because these techniques are increasingly integrated directly into MLaaS offerings (e.g., AWS Clean Rooms, Azure Confidential Computing). For an investor, the project represents academic due diligence rather than a defensible software product. It is easily displaced by more integrated, actively maintained frameworks like PySyft (OpenMined) or the aforementioned platform-specific libraries.