Unverified50% confidenceFactExact time
On the three geometric metrics of stable rank, spectral drift and principal weight overlap, Deletion Tuning is closer to full fine-tuning than LoRA and random sparse fine-tuning.
1
Sources
50%
Confidence
Long-term
Relevance
11/10/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe study analyzes the geometric properties of weight changes using three metrics: stable rank, spectral drift, and principal weight overlap.67% similarVerifiedWithout a unified testing environment specification, the comparability of acceleration data is limited, and voting results may be subject to sampling bias.64% similarUnverifiedJagannathan and Ma (2003) showed that the minimum-variance portfolio often outperforms full mean-variance optimization out of sample, and that a no-short-sale constraint has a regularization effect similar to shrinkage.61% similarUnverifiedIn the early stage of fine-tuning, LoFIT can already distinguish members, while the naive denoising loss baseline remains near random level.61% similarUnverifiedRanked-choice voting has voters rank candidates, and if no one exceeds 50%, the lowest is eliminated and votes are transferred; modern scanners and software make complex transfers faster and easier to manage, but the results are less intuitive.60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1157640API
curl https://kongchang.com/api/v1/knowledge/claims/1157640MCP
get_claim(id=1157640)