Non-Uniform Stochastic Average Gradient Method for Training Conditional Random Fields
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
This paper explores using a stochastic average gradient (SAG) algorithm for train-ing conditional random fields (CRFs). The SAG algorithm is the first general stochastic gradient algorithm to have a linear convergence rate. However, despite its success on simple classification problems, when applied to CRFs the algorithm requires too much memory because it requires storing a previous gradient with respect to every training example. In this work we show that SAG algorithms can be tractably applied to large-scale CRFs by tracking the marginals over ver-tices and edges in the graphical model. We also incorporate a simple non-uniform adaptive sampling scheme that learns how often we should sample each training point. Our experimental results reveal that this method significantly outperforms existing methods. 1 Conditional Random Fields Conditional random fields (CRFs) [9] are a ubiquitous tool in natural language processing. They are used for part-of-speech tagging [12], semantic role labeling [1], topic modeling [27], information
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 | 0.003 | 0.001 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.000 | 0.000 |
| 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