Unverified50% confidenceFactExact time
At the representation level, adversarially trained robust models are more easily decomposed by SAEs and invoke fewer SAE features in task attribution.
1
Sources
50%
Confidence
Long-term
Relevance
9/30/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedTask-domain-specific training can reduce inference compute resources in exchange for a moderate trade-off in general-purpose capability.71% similarVerifiedLoRA reduces fine-tuning costs by training only a small number of adapter layers (low-rank matrix bypass training).70% similarUnverified在过参数化区间存在无数个能完美拟合训练数据的解，优化算法如SGD倾向于选择范数最小的解，这种隐式正则化使模型获得更好泛化性能70% similarUnverifiedSparse autoencoders (SAEs) are unsupervised learning tools trained to decompose a neural network's activation vectors into a set of sparse, interpretable feature directions.69% similarUnverified扩散模型的训练目标为简单的去噪均方误差，相比GAN避免了对抗训练的不稳定性，降低了模式崩塌的风险69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/961073API
curl https://kongchang.com/api/v1/knowledge/claims/961073MCP
get_claim(id=961073)