AdaProb: Efficient Machine Unlearning via Adaptive Probability

AdaProb addresses a high-value niche in machine learning—privacy and the 'right to be forgotten' (GDPR compliance). While the project has 4 forks (indicating immediate academic interest), its 0 stars and 8-day age reflect its status as a fresh research publication rather than a production-ready tool. The defensibility is low (3) because the code serves as a reference implementation for an algorithm; the 'moat' is purely the mathematical approach, which is now public. Frontier labs (Google, OpenAI) have a high risk of displacing this because they are heavily invested in machine unlearning to handle copyright and privacy disputes; they are more likely to implement their own proprietary versions of adaptive probability or gradient ascent methods than adopt this specific repository. Competitors include the SISA (Sharded, Isolated, Sliced, and Aggregated) framework and other 'selective forgetting' techniques. The displacement horizon is short (1-2 years) because the unlearning field is moving rapidly with frequent state-of-the-art shifts.