A new regression method called Direct Inference on Admissible Sets (Diona) has been introduced, which bypasses extensive global training by directly using samples to estimate state and derivative values along with their admissible bounds. This approach frames values and derivatives as latent variables in a constrained optimization problem, allowing for on-demand local solves instead of lengthy training loops. Diona demonstrates theoretical properties like finite-sample coverage and convergence to true states and derivatives under dense sampling in noiseless conditions. Benchmarks show Diona to be more accurate than Gaussian processes on noiseless data, especially with denser datasets, and it provides uncertainty bounds that contain the true value at over 95% of query points for several functions. Crucially, Diona offers significantly faster prediction times after data updates compared to Gaussian processes and neural networks, though this speed comes at the cost of on-demand predictions. AI
IMPACT This new method could offer faster inference for certain regression tasks compared to existing models like Gaussian processes.
RANK_REASON The cluster contains a research paper detailing a new method. [lever_c_demoted from research: ic=1 ai=0.7]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
