Partial Label Learning for Emotion Recognition from EEG

TL;DR

This work tackles the problem of ambiguous EEG emotion labels by adapting Partial Label Learning (PLL) to emotion recognition, evaluating six PLL methods (DNPL, PRODEN, CAVL, LW, CR, PiCO) on SEED-IV and SEED-V under classical, circumplex, and real-world label-generation schemes. It formalizes the PLL setting for EEG, implements six algorithms with EEG-specific adaptations, and analyzes the role of label disambiguation, ground-truth reliability, and similarity-based candidate-label generation. Key findings show DNPL providing robust performance across setups, circumplex-based disambiguation often helping most methods, and real-world ambiguity (via delta scaling) challenging PLL methods, with PiCO maintaining some advantages due to contrastive learning. The results demonstrate the viability of PLL for affective EEG and offer guidance on candidate-label design, disambiguation strategies, and practical deployment in real-world emotion recognition tasks.

Abstract

Fully supervised learning has recently achieved promising performance in various electroencephalography (EEG) learning tasks by training on large datasets with ground truth labels. However, labeling EEG data for affective experiments is challenging, as it can be difficult for participants to accurately distinguish between similar emotions, resulting in ambiguous labeling (reporting multiple emotions for one EEG instance). This notion could cause model performance degradation, as the ground truth is hidden within multiple candidate labels. To address this issue, Partial Label Learning (PLL) has been proposed to identify the ground truth from candidate labels during the training phase, and has shown good performance in the computer vision domain. However, PLL methods have not yet been adopted for EEG representation learning or implemented for emotion recognition tasks. In this paper, we adapt and re-implement six state-of-the-art PLL approaches for emotion recognition from EEG on two large emotion datasets (SEED-IV and SEED-V). These datasets contain four and five categories of emotions, respectively. We evaluate the performance of all methods in classical, circumplex-based and real-world experiments. The results show that PLL methods can achieve strong results in affective computing from EEG and achieve comparable performance to fully supervised learning. We also investigate the effect of label disambiguation, a key step in many PLL methods. The results show that in most cases, label disambiguation would benefit the model when the candidate labels are generated based on their similarities to the ground truth rather than obeying a uniform distribution. This finding suggests the potential of using label disambiguation-based PLL methods for circumplex-based and real-world affective tasks.

Figures (13)

• Figure 1: A description of full-supervised learning, semi-supervised learning and partial label learning for emotion recognition is presented.
• Figure 2: A general EEG-based PLL framework for emotion recognition is presented.
• Figure 3: Wheel of emotions reproduced from russell1980circumplexzhong2019study.
• Figure 4: Normalized similarity score between two emotions as label ambiguity level in the circumplex PLL setup.
• Figure 5: Normalized similarity score ($\gamma_{\delta}$) between two emotions as label ambiguity level in the real-world PLL setup, with $\delta \in \{1, 2, 3, 4, 5\}$. The $\delta$ scaling controls the ambiguity level, affecting the similarity between emotion pairs based on their relative distances.