method
active
method:forward-backward-training-passForward-Backward Training Pass
Standard training procedure used for DLGN, updating gate probability distributions via backpropagation
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Model selects future output targets and works backward from them.
- Used to update pain beliefs online from observations of happiness
- The post-training approach used by frontier AI labs to shape the assistant persona, introduced as open-source in this paper
- The phase after pre-training where models are further tuned with techniques like DPO; the period where the studied behavior emerged.
- Fine-tuning on reflection turns to implant principles into workspace.
- Broader research area: methods to align model behavior after initial training, where undesired behaviors can emerge.
- Training models to answer Yes/No whether a prompt edit changes their behavior.
- Training regime where random subsets of cells update per step, improving robustness of learned circuits