.png){kind=icon}
.png){kind=icon}
Return
DLRM-FER: Self-evolving dual-loop neuro-symbolic learning via dynamic rule memory for explainable facial expression recognition
.png){kind=icon}
OA
AI
T
K
DOI：10.1016/j.knosys.2026.116727.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Deep learning-based facial expression recognition (FER) models have made significant progress, but most still lack interpretability, and existing explainable models often sacrifice accuracy in pursuit of explainability. To address this bottleneck, we propose an adaptive attention-driven dynamic rule bank closed-loop explainable learning framework. This framework introduces an Adaptive Bi-Stream Emotion-Selective Attention (ABESA) mechanism, which captures both global emotional semantics and local muscle activations through adaptive weighting. By leveraging attention saliency, the framework constructs a self-evolving symbolic rule bank without relying on Action Unit (AU) annotations or prior knowledge. The rules are continuously updated in real time based on attention-category consistency, either strengthening or discarding them. These explainable rules serve both as explanatory signals and corrective feedback to guide the model’s decision-making, constructing the first autonomous neuro-symbolic system with two explicit loops: a training-time learning loop (rule-consistency supervision reshapes attention via back-propagation) and an inference-time self-correction loop (uncertainty-triggered rule lookup and correction). Our approach is evaluated on major FER datasets. Our results compares favorably to state-of-the-art methods on explainable FER models, and achieves competitive performance compared with the mainstream non-interpretable FER deep learning methods, thus validating its ability to balance high accuracy with strong self-explainability in affective computing.
Keywords:
Facial expression recognition (FER)
Explainable artificial intelligence (XAI)
Neuro-symbolic learning
Dual-loop closed-loop learning
Dynamic rule memory
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
K
IF:
7.6
Papers:
1.2W
Citations:
4.5W
Organization
No organization information available
Cited Papers