Unverified50% confidenceBenchmarkExact time
In the class-incremental drift scenario, the accuracy drop of EWC and naive fine-tuning reaches as high as 16.2%–28.5%, far exceeding the 1.8%–5.8% of experience replay.
1
Sources
50%
Confidence
Long-term
Relevance
10/27/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedEigenCAM's overall faithfulness is lower than D-RISE, but it similarly exhibits score coupling.70% similarUnverifiedCompared to the historical mean baseline, entry-level and turn-level predictions reduced mean absolute error (MAE) by 4.03% and 3.92%, respectively.69% similarUnverifiedWhen models were provided with complete company records, strict accuracy actually dropped from 41% to 18%, while recall rose and precision collapsed.69% similarUnverifiedAt moderate noise levels and the middle layers of the model, 2 learnable null slots absorbed approximately 90% of the cross-attention mass, while the EOS token's share dropped sharply to approximately 4%.69% similarVerifiedINT8 quantization loses approximately 0.5% accuracy while halving VRAM usage; INT4 quantization compresses VRAM to 1/4 of the original with an accuracy loss of approximately 1–3%.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1102414API
curl https://kongchang.com/api/v1/knowledge/claims/1102414MCP
get_claim(id=1102414)