Skip to main content
Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

< Back to Article

Fig 1.

DGASA architecture: lightweight subspace construction and drift compensation.

More »

Fig 1 Expand

Fig 2.

Inference pipeline of the proposed DGASA framework.

GA modules are inserted in each Transformer block for task-specific adaptation.

More »

Fig 2 Expand

Fig 3.

Architecture of the Gated Adapter (GA) module.

The gating mechanism uses global pooling followed by a linear layer with sigmoid activation to produce a scalar gate g(x), which modulates the adapter residual branch.

More »

Fig 3 Expand

Table 1.

Comparison of different methods on five benchmark datasets. denotes average accuracy over incremental stages, and is the accuracy after the last stage. The best performance is shown in bold. All methods are implemented without using exemplars.

More »

Table 1 Expand

Table 2.

Comparison of different methods with exemplar usage and accuracies on benchmark datasets.

More »

Table 2 Expand

Table 3.

Ablation Study Results on CIFAR-100 Inc5 Task.

More »

Table 3 Expand

Table 4.

Ablation study on the number and positions of Attention-Gated Adapters.

More »

Table 4 Expand

Table 5.

Ablation study on the weighting coefficient .

More »

Table 5 Expand

Fig 4.

t-SNE visualization comparison between (a) Adaptive Weighting by Instance-level Significance and (b) Fixed Weight Fusion by Subspace Prototypes.

More »

Fig 4 Expand

Table 6.

Parameter growth analysis across incremental tasks.

More »

Table 6 Expand

Table 7.

Prototype memory cost across incremental tasks (CIFAR-100, 100 total classes).

More »

Table 7 Expand

Table 8.

Inference efficiency analysis across tasks (batch size = 64).

More »

Table 8 Expand

Table 9.

Break-even analysis: efficiency vs. accuracy trade-off. All results are reported on CIFAR-100 Inc5.

More »

Table 9 Expand

Table 10.

Forgetting and backward transfer analysis across benchmark datasets. Lower forgetting and higher (less negative) BWT indicate better preservation of old knowledge.

More »

Table 10 Expand

Fig 5.

Old vs. new class accuracy across incremental sessions on CIFAR-100 Inc5.

More »

Fig 5 Expand

Table 11.

Statistical significance analysis (paired t-test, p-values). Values < 0.05 indicate significant improvement.

More »

Table 11 Expand

Fig 6.

Per-task accuracy matrix on CIFAR-100 Inc5.

Diagonal entries show performance on the current task; off-diagonal entries show retention of previous tasks.

More »

Fig 6 Expand

Table 12.

Old vs. new class accuracy on CUB Inc10 (final stage).

More »

Table 12 Expand

Fig 7.

Confusion matrix comparison on CUB Inc10: (a) ADAM+Adapter, (b) DGASA.

More »

Fig 7 Expand

Fig 8.

Distribution of adaptive weights: (a) samples from old classes, (b) samples from new classes.

More »

Fig 8 Expand

Table 13.

Representative failure cases on CUB Inc10.

More »

Table 13 Expand

Table 14.

Calibration analysis on CUB Inc10.

More »

Table 14 Expand

Table 15.

Conditional analysis of adaptive weighting effectiveness.

More »

Table 15 Expand

Table 16.

Computational overhead of adaptive weighting (20 tasks).

More »

Table 16 Expand

Table 17.

Sensitivity analysis of the regularization parameter on CIFAR-100 Inc5 (20 tasks).

More »

Table 17 Expand

Table 18.

Impact of increment size and conditioning on drift compensation performance. Experiments conducted on CIFAR-100 with varying per-task class increments.

More »

Table 18 Expand

Table 19.

Stability analysis across long task sequences on CIFAR-100 (Inc5, 20 tasks total).

More »

Table 19 Expand