⏱
UTC Weighted Average AI Optimizer (LSTM model)
12-Lab metrology · All weights non-zero · Target: wt. avg = 0.000000 ns per MJD
Developer: Tamal Maity
sis = Σ 1/uᵢ²w_trad = (1/uᵢ²)/sis → wt.avg ≠ 0w_AI = min (Σ UTC_i·wᵢ)² s.t. Σwᵢ=1, wᵢ ≥ ε_min → wt.avg = 0
📂
Drop clean_UTC_average.csv here
or click to browse
HOW IT WORKS
→ Parse CSVReads MJD, UTC−UTC(Lab), and uncertainties u₁…u₁₂
→ Traditional weightsw_trad = (1/uᵢ²) / sis → wt.avg ≠ 0 (avg ~1 ns error)
→ AI OptimizationConstrained PGD: all wᵢ ≥ ε_min, Σwᵢ=1, minimizes (Σ UTC·w)²
→ No zero weightsSet ε_min via slider — every lab gets meaningful weight
→ Iteration gapOptimizer runs until |wt.avg| ≤ target tolerance → exact 0