Differentially Private Model Training on DP Noise Correlation Utility benchmark leaderboard · SOTA2 Research

Differentially Private Model Training on DP Noise Correlation Utility

7.77RMSE (eps=8, w/o Amp)BandMF

Evaluation Results

Method	Links
BandMF2026.01		7.77	5.84	10.42	19.16	35.94	68.02	128.79
BandInvMF2026.01		7.87	5.66	9.86	17.76	33.25	62.84	119.2
BLT2026.01		8.14	5.87	10.67	19.5	35.97	69.41	127.69
BandInvMF2026.01		8.15	6.39	11.36	20.87	39.13	74.06	140.38
BSR2026.01		8.15	5.49	9.31	17.06	31.95	60.43	114.52
BISR2026.01		8.45	6.34	11.3	20.77	38.96	73.7	139.68
BandInvMF2026.01		8.54	6.56	11.23	20.17	37.09	71.08	130.9
BandInvMF2026.01		10.27	8.07	12.48	20.76	37.37	70.9	130.48
BISR2026.01		10.5	6.81	9.87	15.92	28.8	53.89	101.9
BandMF2026.01		12.58	8.77	13.25	19.76	31.4	56.07	104.68
BandInvMF2026.01		12.69	9.7	14.9	24.41	43.66	82.77	125.87
DP-λCGD2026.01		12.73	9.33	14.03	21.45	36.4	68.37	125.69
DP-λCGD2026.01		14.74	10.27	15.28	21.45	31.7	56.75	103.44
BSR2026.01		14.89	9.62	13.32	18.88	29.34	52.16	97.51
BISR2026.01		17.95	11.5	15.75	21.1	20.25	52.81	96.45
DP-λCGD2026.01		19.72	13.25	18.42	24.73	33.66	54.33	97.73
BandMF2026.01		22.05	15	21.2	28.13	37.06	56.29	99.33
BSR2026.01		26.27	16.88	22.41	28.59	36.78	55.03	97.09
BISR2026.01		33.47	21	27.97	34.87	42.5	58.27	98.52
BandMF2026.01		42.29	27	36.99	45.9	55.35	68.64	104.42
BSR2026.01		46.8	29.29	39.32	48.13	57.03	70.47	104.9
BISR2026.01		48.45	30.52	40.58	49.44	58.16	71.2	105.09
BandMF2026.01		59.44	37.67	49.08	59.95	71.17	86.25	114.94
BSR2026.01		62.51	39.65	51.87	62.97	74.14	88.33	116.43
DP-SGD2026.01		83.85	21.82	26.27	31.68	40.1	59.17	100.27