Unverified50% confidenceFactExact time
Under the reinforcement learning setting of the continual learning scenario, Deletion Tuning's prior-task accuracy remains near the base model level, while full fine-tuning, LoRA and random sparse fine-tuning lose up to 5.4 percentage points.
1
Sources
50%
Confidence
Long-term
Relevance
11/10/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedRe-running Quiet-STaR's training recipe with an optimizer capable of applying every update caused model performance to diverge; the high-precision optimizer required lowering the learning rate by 3 to 10 times to obtain the same results and could not achieve better results.78% similarUnverifiedMAEM was validated on four class-incremental graph learning benchmarks, and the results show that recency bias is significantly alleviated, performance surpasses several strong continual graph learning baselines, and model capacity remains bounded under strictly replay-free conditions.77% similarUnverifiedEarly on, methods such as intermediate supervision, Sudoku-specific losses and architecture adjustments could accelerate the rise of accuracy in early training, but the final models mostly stalled near about 60% validation accuracy.77% similarVerifiedOn convergent tasks, there is a clear performance saturation phenomenon after models reach a certain scale, where continued increases in parameter count or training data yield rapidly diminishing marginal contributions to accuracy.77% similarUnverifiedExperiments show that even when what is learned is a time-independent static field, the proportion of off-mode samples generated by this method is far lower than that of score-matching generators with comparable parameter counts.76% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1157639API
curl https://kongchang.com/api/v1/knowledge/claims/1157639MCP
get_claim(id=1157639)