STAA: Spatio-Temporal Attention Attribution for Real-Time Interpreting Transformer-Based AI Video Models
Bibliographic record
Abstract
Transformer-based video models have performed SOTA in various action recognition and video understanding tasks. The limitation is the lack of explainability of these complex models. Current Explainable AI (XAI) methods focus on feature importance analysis of only one dimension, either spatial or temporal features. When applied to transformer-based video models, the challenges are two-fold. They fail to capture the integrated spatio-temporal nature of video data and incur prohibitive computational costs for real-time applications. This paper presents STAA (Spatio-Temporal Attention Attribution), an XAI method for interpreting video transformer models for action recognition tasks. STAA simultaneously extracts both spatial importance within frames and temporal relevance across the video sequence from attention values in transformers. This unified approach solves the challenge by capturing how transformer models integrate information across both dimensions for decision-making, providing comprehensive explanations that reflect the model’s actual reasoning process. We enhance STAA’s raw output through post-processing. The experiments on the Kinetics-400 dataset demonstrate superior faithfulness (0.844 ± 0.116) and monotonicity (0.850 ± 0.030). In terms of computational overhead, our STAA method is applicable for real-time video XAI analysis applications. We implement a cloud-based architecture that enables video explanations with 150 ms latency.
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 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 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.002 |
| Open science | 0.000 | 0.000 |
| 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 itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, 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".