Sample Quality Evaluation using Reinforcement Learning for Lost Circulation Recognition
Bibliographic record
Abstract
Summary The quality of the training samples directly determines the performance of intelligent lost circulation recognition models. To mitigate the negative effect of low-quality training samples on lost circulation recognition accuracy, a reinforcement learning (RL)-based sample quality evaluation method, including a fully connected neural network (FCNN) for sample quality evaluation, a randomized trial-based sample selector, and a lost circulation recognition network based on bidirectional gated recurrent unit (Bi-GRU), is proposed. Initially, FCNN uses the Glorot uniform initializer to initialize the network weights, ensuring that each sample has a 50% probability of being selected. Next, the sample selector conducts a randomized trial on each sample using the obtained selection probability to label it as either 0 or 1, and the samples with Label 1 are included in the training set. Then, the lost circulation recognition model based on Bi-GRU is trained and verified using the samples in the training set to produce a recognition accuracy, which is used to calculate a reward value indicating the quality of samples in the training set. The obtained reward value is used to update the loss function of FCNN. Subsequently, the network parameters of FCNN are readjusted by minimizing the loss function using RL to reduce the selection probability of low-quality samples. The above cycle of sample selection, lost circulation recognition based on Bi-GRU, loss function update, and parameter optimization of FCNN network is repeated until the recognition accuracy of the Bi-GRU model ceases to increase. In addition, a reward value calculation method using the exponential moving average (EMA) is introduced, which allows the RL model to be trained more efficiently and stably. Sample quality evaluation experiments with 5,011 lost circulation samples from field data were conducted. The results demonstrate that compared with the Bi-GRU trained using all samples, the recognition accuracy of the Bi-GRU trained after excluding low-quality samples is improved by 22.7%, with its false negative rate (FNR) reduced from 15.8 to 4.7%. The model’s convergence speed is improved by 17.2% using the proposed reward value calculation method.
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.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.004 | 0.013 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 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 itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".