Эксперимент: Residual-Redundancy Adapter Clustering (#417)

{ "worked": true, "confidence": 9, "verdict": "Built a readable NumPy MVP implementing shrinkage covariance, Gaussian residual total correlation, merge-cost agglomerative clustering, and an equal-budget multitask regression comparison. The mechanism manifested: empirical TC matched the exact Gaussian formula across rho=0..0.8, small-rho behavior was approximately quadratic, and clustering recovered all three true residual pairs for rho>=0.2. At equal nominal adapter parameters (36), residual-TC routing achieved MSE 0.426 versus 5.375 for random routing and 8.808 for a shared low-rank adapter; this is a strong toy signal, not evidence of general Transformer/LoRA superiority.", "metrics": { "baseline": "Shared low-rank adapter: MSE 8.8078, 36 nominal parameters. Random 3-way partition: MSE 5.3748, 36 parameters.", "idea": "Residual-TC partition: MSE 0.4261, 36 parameters; recovered [[0,1],[2,3],[4,5]]. Gaussian TC prediction errors were <=0.0104 over rho=0..0.8; pairwise TC matched -0.5 log(1-rho^2) with observed versus predicted values .0216/.0204, .0858/.0872, .2259/.2231, and .5145/.5108 for rho=.2/.4/.6/.8. Exact block recovery was 4/5 sweep points, failing only at rho=0 as expected from no signal." }, "how_to_run": "/home/maxwelhelp/main/bin/python3 experiment.py", "files": [ "experiment.py", "results.json", "run_output.txt" ], "limitations": "Only a synthetic six-task linear regression was tested; no Transformer, LoRA implementation, GPU, periodic reassignment, hysteresis, mixed classification/regression, gradient-clustering baseline, or memory/speed measurement was evaluated. The equal parameter counts are nominal adapter counts and omit shared-head/base-model accounting." }