Unverified50% confidenceFactExact time
In neural networks, parameters jointly encode knowledge of multiple tasks through gradient descent; when the model trains on a new task, gradient updates substantially change the weights critical to old tasks, while old task data no longer appears, causing the model to gradually forget the old tasks.
1
Sources
50%
Confidence
Long-term
Relevance
10/28/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe core mechanism of experience replay is to maintain a memory buffer of limited capacity that stores some samples from historical tasks, and when training on a new task, mix them with the current data to form mini-batches, so that the loss function always contains gradient signals from the old distribution.78% similarUnverifiedDuplicate content in training data can cause model memorization and performance degradation.78% similarUnverifiedRecency bias refers to the model repeatedly encountering the latest task data at the end of training, causing the decision boundary and output confidence to skew toward recent tasks, rendering threshold-based OOD discrimination ineffective, and this cannot be corrected by simple temperature scaling or post-processing.75% similarUnverifiedIn continual learning, standard confidence scores shift systematically as tasks progress: the model is overconfident on inputs from recent tasks and gives muddled confidence for old tasks and truly unknown inputs.74% similarUnverifiedContext compression causes previously injected system prompts and custom skill instructions to be diluted or even lost, leading to model behavior drift that requires reloading.74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1109275API
curl https://kongchang.com/api/v1/knowledge/claims/1109275MCP
get_claim(id=1109275)