Interpreting Tangled Program Graphs Under Partially Observable Dota 2 Invoker Tasks
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
Interpretable machine learning (ML) implies that structural relationships to exist between different parts of the ML model. We demonstrate how the tangled program graph (TPG) framework is able to demonstrate structural relationships for a suite of partially observable reinforcement learning tasks. The tasks are defined in terms of developing spell casting behaviours for the Invoker hero under the Dota 2 game engine. We show that TPG is able to demonstrate all 4 of the properties used to define interpretable machine learning. Moreover, a unique form of feature engineering / modularity takes place between programs that define state for (indexed) memory versus programs defining actions. The full article appears as Smith and Heywood (2024) "Interpreting Tangled Program Graphs under Partially Observable Dota 2 Invoker tasks" in IEEE Transactions on Artificial Intelligence, 5 (4): 1511–1524. https://doi.org/10.1109/TAI.2023.3279057
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 | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.001 | 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