Mother-Infant EEG Attractor States Predict Early Social Decision Making
Poster No:
1338
Submission Type:
Abstract Submission
Authors:
Angshuk Dutta1, Lorena Santamaria2, George Rafael Domenikos3, Victoria Leong3
Institutions:
1CLIC, Nanyang Technological University, Singapore, Singapore, 2School of Social Sciences, Nanyang Technological University, Singapore, Singapore, 3EMPOWER, Nanyang Technological University, Singapore, Singapore
First Author:
Co-Author(s):
Introduction:
Infants learn vicariously through social observation of their caregiver's actions and affective states, but the neural mechanisms by which the infant's brain encodes such information remain poorly understood. Here, we adopt a dynamical systems perspective to model attractor states derived from the infant's neural activity conditioned on the mother's during a social decision-making task.Our contribution is twofold: 1) we develop a piecewise linear recurrent neural network to infer the latent dynamics in infant-mother EEG, and 2) we provide evidence for line attractors states that are associated with systematic biases in infants' affective decision-making.
Methods:
Participants. A total of 44 British mother-infant dyads (mean age = 29.84m, SD = 3.21) participated in this study.
Task: In this social decision-making task, the mother demonstrates a pair of novel objects to her infant, labelling one with positive facial affect and tone of voice and the other with negative affect and tone. The infant is then allowed to interact with both objects, and we observe which object is selected first. Infants completed up to 16 trials (average 8 trials). We classified infants into three typologies based on their decision bias: positive, negative or volatile (choices were ambivalent).
EEG: Data was concurrently collected from mother and infant using a 32-channel gel-based system and manually pre-processed to remove motion artifacts. Data from the maternal demonstration phase was used for this analysis.
Modelling: Our model was trained using EEG data from 39 mother-infant dyads (infant age: 29.26m ± 4.92 s.d.) and tested on 5 dyads. We developed a Dendritic Piecewise Linear Recurrent Neural Network (PLRNN) (Kramer et al., 2021) to infer latent infant EEG dynamics conditioned on maternal EEG. Given the chaotic nature of EEG dynamics and the sensitivity of RNN learning to Lyapunov exponents (Mikhaeil, Monfared, & Durstewitz, 2022), we addressed potential divergence by jointly training a four-layer Convolutional Neural Network (CNN) with the PLRNN using teacher forcing (Hess et al., 2023). The CNN corrected PLRNN trajectories during divergence, ensuring convergence and capturing the underlying dynamics.
The model was evaluated by reconstructing infant EEG from latent states and predicting trial-level choices, comparing predictions to ground truth. Attractor states were analysed by reducing latent trajectories to two dimensions using PCA, and the L2 norm of these dimensions was computed to quantify latent state energy.
Task: In this social decision-making task, the mother demonstrates a pair of novel objects to her infant, labelling one with positive facial affect and tone of voice and the other with negative affect and tone. The infant is then allowed to interact with both objects, and we observe which object is selected first. Infants completed up to 16 trials (average 8 trials). We classified infants into three typologies based on their decision bias: positive, negative or volatile (choices were ambivalent).
EEG: Data was concurrently collected from mother and infant using a 32-channel gel-based system and manually pre-processed to remove motion artifacts. Data from the maternal demonstration phase was used for this analysis.
Modelling: Our model was trained using EEG data from 39 mother-infant dyads (infant age: 29.26m ± 4.92 s.d.) and tested on 5 dyads. We developed a Dendritic Piecewise Linear Recurrent Neural Network (PLRNN) (Kramer et al., 2021) to infer latent infant EEG dynamics conditioned on maternal EEG. Given the chaotic nature of EEG dynamics and the sensitivity of RNN learning to Lyapunov exponents (Mikhaeil, Monfared, & Durstewitz, 2022), we addressed potential divergence by jointly training a four-layer Convolutional Neural Network (CNN) with the PLRNN using teacher forcing (Hess et al., 2023). The CNN corrected PLRNN trajectories during divergence, ensuring convergence and capturing the underlying dynamics.
The model was evaluated by reconstructing infant EEG from latent states and predicting trial-level choices, comparing predictions to ground truth. Attractor states were analysed by reducing latent trajectories to two dimensions using PCA, and the L2 norm of these dimensions was computed to quantify latent state energy.
Results:
Infant Object Choice: Overall, 15 infants (35%) showed a stable positive bias, 11 infants (23.2%) showed a stable negative bias, and 18 infants (41.8%) showed volatile bias.
Model Validation: The RNN latent trajectories achieved an F1 score of 0.865 ± 0.023 s.d in predicting infant choices and a cosine similarity of 0.802 ± 0.011 s.d between reconstructed and observed infant EEG.
Attractor States: Visualizing the energy landscape highlighted attractor basins for line attractors (Figure 2a, b). Line attractor scores, calculated using the two greatest Lyapunov exponents (>1 indicating presence), showed evidence for line attractors in infants with positive (2.08 ± 0.42) and negative (2.56 ± 1.021) decision biases but not in volatile infants (1.002 ± 1.03). Significant differences were observed between groups: volatile-positive (p = 0.00329**) and volatile-negative (p = 0.000841***) using Welch's t-test.
Model Validation: The RNN latent trajectories achieved an F1 score of 0.865 ± 0.023 s.d in predicting infant choices and a cosine similarity of 0.802 ± 0.011 s.d between reconstructed and observed infant EEG.
Attractor States: Visualizing the energy landscape highlighted attractor basins for line attractors (Figure 2a, b). Line attractor scores, calculated using the two greatest Lyapunov exponents (>1 indicating presence), showed evidence for line attractors in infants with positive (2.08 ± 0.42) and negative (2.56 ± 1.021) decision biases but not in volatile infants (1.002 ± 1.03). Significant differences were observed between groups: volatile-positive (p = 0.00329**) and volatile-negative (p = 0.000841***) using Welch's t-test.
Conclusions:
Here, we demonstrate the existence of a line attractor in mother-infant EEG latent states, which corresponds to persistent decision-making affective biases by infants. Line attractors in EEG reflect persistent neural activity, linked to working memory (Brennan & Proekt, 2023; Seung, 1996) and recall of demonstrated objects. Further investigation is warranted to demonstrate social behaviours from the mother that modulate or tune this attractor.
Emotion, Motivation and Social Neuroscience:
Social Interaction 2
Modeling and Analysis Methods:
EEG/MEG Modeling and Analysis 1
Keywords:
Computational Neuroscience
Development
Electroencephaolography (EEG)
Emotions
Modeling
Social Interactions
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
Provide references using APA citation style.
Reconstructing nonlinear dynamical systems from multi-modal time series.
arXiv preprint arXiv:2111.02922 .
Mikhaeil, J., Monfared, Z., & Durstewitz, D. (2022). On the difficulty of learning
chaotic dynamics with rnns. Advances in Neural Information Processing
Systems, 35 , 11297–11312.
Hess, F., Monfared, Z., Brenner, M., & Durstewitz, D. (2023). Gener-
alized teacher forcing for learning chaotic dynamics. arXiv preprint
arXiv:2306.04406 .
Brennan, C., & Proekt, A. (2023). Attractor dynamics with activity-dependent
plasticity capture human working memory across time scales. Communi-
cations psychology, 1 (1), 28.
Seung, H. S. (1996). How the brain keeps the eyes still. Proceedings of the
National Academy of Sciences, 93 (23), 13339–13344.