Hindsight: Posterior-guided training of retrievers for improved\n open-ended generation
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Many text generation systems benefit from using a retriever to retrieve\npassages from a textual knowledge corpus (e.g., Wikipedia) which are then\nprovided as additional context to the generator. For open-ended generation\ntasks (like generating informative utterances in conversations) many varied\npassages may be equally relevant and we find that existing methods that jointly\ntrain the retriever and generator underperform: the retriever may not find\nrelevant passages even amongst the top-10 and hence the generator may not learn\na preference to ground its generated output in them. We propose using an\nadditional guide retriever that is allowed to use the target output and "in\nhindsight" retrieve relevant passages during training. We model the guide\nretriever after the posterior distribution Q of passages given the input and\nthe target output and train it jointly with the standard retriever and the\ngenerator by maximizing the evidence lower bound (ELBo) in expectation over Q.\nFor informative conversations from the Wizard of Wikipedia dataset, with\nposterior-guided training, the retriever finds passages with higher relevance\nin the top-10 (23% relative improvement), the generator's responses are more\ngrounded in the retrieved passage (19% relative improvement) and the end-to-end\nsystem produces better overall output (6.4% relative improvement).\n
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.005 | 0.005 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it