Sparse Autoencoder Concept Alignment on CUB benchmark leaderboard · SOTA2 Research

Sparse Autoencoder Concept Alignment on CUB

0.997SparsityMP

Evaluation Results

Method	Links
MP2025.12		0.997	0.129	0.071	0.133	0.778	6.37	6.2	11.5	8.1
MP2025.12		0.997	0.174	0.053	0.193	0.87	4.13	13.8	13.5	13.5
TopK2025.12		0.995	0.099	0.062	0.241	0.968	6.89	18.2	29.7	22.4
TopK2025.12		0.995	0.208	0.066	0.26	0.736	6.95	13.3	23.7	17
TopK2025.12		0.995	0.331	0.076	0.275	0.848	6.71	13.7	22.8	17
Archetypal2025.12		0.994	0.132	0.079	0.395	0.893	6.85	14.4	22.1	17.2
Archetypal2025.12		0.991	0.1	0.082	0.55	0.967	6.07	20.7	33	25.3
Vanilla2025.12		0.99	0.127	0.099	0.152	0.752	6.86	10.8	16.6	13
BatchTopK2025.12		0.988	0.098	0.075	0.657	0.973	5.77	23.1	36.6	28.2
MP2025.12		0.988	0.1	0.088	0.363	0.968	6.51	18.5	26.4	21.5
Vanilla2025.12		0.985	0.101	0.088	0.35	0.968	6.55	19.9	28	23
BatchTopK2025.12		0.982	0.127	0.077	0.632	0.835	5.47	15	26.7	19.1
Jump2025.12		0.98	0.098	0.095	0.419	0.974	6.34	19.7	28.5	23
Archetypal2025.12		0.974	0.115	0.092	0.23	0.797	5.655	17.4	29.4	21.7
BatchTopK2025.12		0.969	0.114	0.089	0.607	0.768	5.03	15.9	27.6	20.1
Jump2025.12		0.967	0.118	0.137	0.362	0.807	6.59	15	23.5	18.2
Vanilla2025.12		0.963	0.134	0.153	0.676	0.928	6.79	22.6	28.1	24.5
Jump2025.12		0.949	0.131	0.156	0.706	0.938	6.81	24.2	32.3	27.2