Correct and efficient search algorithms in the presence of repetitions
Bibliographic record
Abstract
AND/OR tree search has been a fundamental topic in Artificial Intelligence, because many tasks can be decomposed into subtasks, such that either all (AND) or one (OR) of them must be solved. Recent AND/OR tree search algorithms have become powerful, especially by using the notion of proof and disproof numbers. However, there are limitations of these algorithms if the search space involves repetitions. Repetitions cause a problem of efficiency versus correctness. Some approaches incorrectly deal with repetitions to preserve search efficiency. As a result, they occasionally return incorrect solutions. Other approaches compromise efficiency to guarantee correctness. However, they are not efficient enough to become satisfactory choices of practitioners. This thesis presents effective and correct methods for AND/OR tree search with repetitions. The one-eye problem in the game of Go, tsume-Go (life and death problem), and checkers are used as application domains to explore the new techniques. The thesis contains four research contributions. First of all, a solution to the Graph History Interaction (GHI) problem, which may cause a solver to return the incorrect outcome because of repetitions, is presented. Theoretical and empirical results show that the GHI solution is general, correct, and efficient. Secondly, a performance problem is presented when the depth-first proof number (df-pn) search algorithm, which is an effective algorithm using proof and disproof numbers, is adapted to domains involving repetitions. A solution to the problem is given and dramatical improvements over df-pn are empirically achieved. Thirdly, on top of these solutions, domain dependent enhancements are added to the programs that solve the one-eye and tsume-Go problems. These techniques are very promising, and contribute to surpass the performance of the best existing tsume-Go solver. Finally, a divide and conquer approach that can reduce the search space is presented. This approach further improves the performance of the one-eye solver.
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 | 0.008 | 0.058 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.002 | 0.002 |
| Bibliometrics | 0.002 | 0.002 |
| Science and technology studies | 0.001 | 0.004 |
| Scholarly communication | 0.004 | 0.011 |
| Open science | 0.003 | 0.005 |
| Research integrity | 0.003 | 0.004 |
| Insufficient payload (model declined to judge) | 0.004 | 0.002 |
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".