Online Continual Learning on Permuted MNIST 1000-tasks (last 5 tasks)
95.2Mean Accuracy (5 Tasks)BiMU
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BiMUWeight precision=Binary, Network capacity=2000 neurons, Monte Carlo inference samples=52026.05 | 95.2 | 862.09 | |
| MESUWeight precision=Real-valued, Network capacity=2000 neurons, Monte Carlo inference samples=102026.05 | 92.99 | 171.82 | |
| SIWeight precision=Real-valued, Network capacity=2000 neurons, Monte Carlo inference samples=Deterministic2026.05 | 92.82 | 89.77 | |
| MESUWeight Type=Real-valued, Task Bounds=NO, Architecture=100-unit MLP2026.05 | 91.69 | 261.1 | |
| EWC O.Weight precision=Real-valued, Network capacity=2000 neurons, Monte Carlo inference samples=Deterministic2026.05 | 91.47 | 139.66 | |
| SGDWeight precision=Real-valued, Network capacity=2000 neurons, Monte Carlo inference samples=Deterministic2026.05 | 90.61 | 117.65 | |
| BiMUWeight Type=Binary, Task Bounds=NO, Architecture=100-unit MLP2026.05 | 90.3 | 139.47 | |
| BAYESBINNWeight precision=Binary, Network capacity=2000 neurons, Monte Carlo inference samples=52026.05 | 86.61 | 9.86 | |
| EWC O.Weight Type=Real-valued, Task Bounds=YES, Architecture=100-unit MLP2026.05 | 81.78 | 6.63 | |
| SIWeight Type=Real-valued, Task Bounds=YES, Architecture=100-unit MLP2026.05 | 74.41 | 5.11 | |
| SGDWeight Type=Real-valued, Task Bounds=NO, Architecture=100-unit MLP2026.05 | 66.64 | 43.5 | |
| STEWeight precision=Binary, Network capacity=2000 neurons, Monte Carlo inference samples=Deterministic2026.05 | 47.39 | 39.12 | |
| BayesBiNNWeight Type=Binary, Task Bounds=YES, Architecture=100-unit MLP2026.05 | 41.12 | 2.04 | |
| STEWeight Type=Binary, Task Bounds=NO, Architecture=100-unit MLP2026.05 | 29.35 | 9.32 | |
| SYN. META.Weight precision=Binary, Network capacity=2000 neurons, Monte Carlo inference samples=Deterministic2026.05 | 10.34 | 1.33 | |
| Syn. Meta.Weight Type=Binary, Task Bounds=YES, Architecture=100-unit MLP2026.05 | 10.27 | 1.64 |