analiza aiqbench
This research introduces CPLUS, a method that leverages self-probing gradients to mitigate catastrophic forgetting in language models during continual learning. By generating synthetic inputs from retained samples rather than relying on sparse replay data, the technique effectively preserves prior knowledge even when historical data is scarce. The approach demonstrates scalability across the Qwen3 model family, showing improved performance as model size increases.
Continual Learning via Self-Probe Gradients
arXiv:2609.32771v1 Announce Type: new Abstract: Adapting pretrained models to new data can cause catastrophic forgetting of previously learned behavior. When only a few past samples remain, they give continual learning methods sparse and narrow evidence about what to preserve. We show that language models can expand this evidence through self-probing, in which the frozen model generates new inputs from the retained samples and records its own predictions on them. Unlike prior work that replays such data as training examples, our method, CPLUS uses self-probe and past-sample gradients to scale down parameter updates that conflict with prior behavior. Experiments with five language models on four benchmarks show three results. First, the same probes preserve more prior behavior as gradient signals than as replay data. Second, CPLUS learns the new data while consistently reducing forgetting more than existing baselines, especially when past data are scarce, and this protection extends to benchmarks not used for training. Third, we observe that CPLUS also becomes more effective as models grow: within the Qwen3 model family, it recovers an increasing share of the forgetting caused by standard fine-tuning.
Abstract:Adapting pretrained models to new data can cause catastrophic forgetting of previously learned behavior. When only a few past samples remain, they give continual learning methods sparse and narrow evidence about what to preserve. We show that language models can expand this evidence through self-probing, in which the frozen model generates new inputs from the retained samples and records its own predictions on them. Unlike prior work that replays such data as training examples, our method, CPLUS uses self-probe and past-sample gradients to scale down parameter updates that conflict with prior behavior. Experiments with five language models on four benchmarks show three results. First, the same probes preserve more prior behavior as gradient signals than as replay data. Second, CPLUS learns the new data while consistently reducing forgetting more than existing baselines, especially when past data are scarce, and this protection extends to benchmarks not used for training. Third, we observe that CPLUS also becomes more effective as models grow: within the Qwen3 model family, it recovers an increasing share of the forgetting caused by standard fine-tuning.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2609.32771 [cs.LG]
(or arXiv:2609.32771v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2609.32771 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 274 cuvinte
Subiecte conexe
Research Note: Filtering Subversion-Relevant Information From Pretraining Data Is Feasible
LessWrong · 5 oct. 2026
Improving CoT Monitorability of Evaluation Awareness via Verbalization Training
LessWrong · 2 oct. 2026
Continual learning might make your blocking monitors nearly useless
AI Alignment Forum · 24 sept. 2026
Game Design Creativity evals
LessWrong · 12 oct. 2026