Interesting Works

Summary: GiGaMAE investigated how to enhance the generalization capability of self-supervised graph generative models, by reconstructing graph information in the latent space. They proposed a nove self-supervised reconstruction loss.

Shi, Yucheng, et al. “Gigamae: Generalizable graph masked autoencoder via collaborative latent space reconstruction.” Proceedings of the 32nd ACM International Conference on Information and Knowledge Management. 2023.

title: “[Article] Masked graph auto-encoder constrained graph pooling.” date: 2024-03-13

Summary: MGAP is the novel node drop pooling method retaining sufficient effective graph information from both node attribute and network topology perspectives.

Liu, Chuang, et al. “Masked graph auto-encoder constrained graph pooling.” Joint European Conference on Machine Learning and Knowledge Discovery in Databases. Cham: Springer International Publishing, 2022.

Summary: The authors compared the theory-based RL model called Explore, Model, Plan Agent (EMPA) on Atari games with human performance and the double DQN model. The EMPA model showed compatible performances with human participants. Encoding analyses identified neural representation associated with theory-encoding and theory-updating.

Tomov, M. S., Tsividis, P. A., Pouncy, T., Tenenbaum, J. B., & Gershman, S. J. (2023). The neural architecture of theory-based reinforcement learning. Neuron.

Summary: They suggested BrainGSLs to capture more information in limited data and insufficient supervision. It incorporates a local topological-aware encoder, a node-edge bi-decoder, a signal representation learning module, and a classifier. They evaluated their model on ASD, BD, and MDD datasets.

Wen, G., Cao, P., Liu, L., Yang, J., Zhang, X., Wang, F., & Zaiane, O. R. (2023). Graph Self-supervised Learning with Application to Brain Networks Analysis. IEEE Journal of Biomedical and Health Informatics.

summary : Recently, Micro-Expression(ME) is one of the popular research interest for Facial Expression Recognition(FER) task. In particular, depth information is often utilized to analyze micro expressions. CAS(ME)3 offers around 80 hours of video dataset with manually labelled micro-expressions & macro-expressions. They also provide depth information and demonstrate effective way to process depth information for multimodal Mircro Expression Recognition(MER). CAS(ME)3 is currently one of the most well-known RGB-D dataset for emotion recognition.

Li, J., Dong, Z., Lu, S., Wang, S. J., Yan, W. J., Ma, Y., … & Fu, X. (2022). CAS (ME) 3: A third generation facial spontaneous micro-expression database with depth information and high ecological validity. IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(3), 2782-2800.

summary : Plants emit remotely detectable and informative airborne sounds under stress. Plants are not quite, human just cannot listen! With this experiments, these sound could be detected from a distance of 3–5m by many mammals and insects, which can make them interact with plant

Khait, I., Lewin-Epstein, O., Sharon, R., Saban, K., Goldstein, R., Anikster, Y., … & Hadany, L. (2023). Sounds emitted by plants under stress are airborne and informative. Cell, 186(7), 1328-1336.)

summary : This study investigates the perception of silence by employing the auditory illusion paradigm. The hypothesis posits that if silence can be perceived, then the same auditory illusion would exist. Therefore, silence was used instead of auditory stimuli to induce auditory illusions. Participants reported experiencing the same illusions as in the normal auditory illusion paradigm

Goh, R. Z., Phillips, I. B., & Firestone, C. (2023). The perception of silence. Proceedings of the National Academy of Sciences, 120(29), e2301463120.)

summary: They proposed a novel GCN based deep learning model to diagnose autism spectrum disorder. They used spatial-temporal features from fMRI data and used Graph convolution network with attention. They achieved SOTA performance.

Park, Kyoung-Won, and Sung-Bae Cho. “A residual graph convolutional network with spatio-temporal features for autism classification from fMRI brain images.” Applied Soft Computing 142 (2023): 110363.)

summary :Here they recorded simultaneous EEG-fMRI data from participants viewing affective pictures. Applying multivariate analyses including SVM and RSA. They found perceptual processing of affective pictures began ~100ms in the visual cortex, where affect-specific representation began to form ~200ms. The neural representation of affective scenes is sustained rather than dynamic.

Bo, Ke, et al. “Decoding the temporal dynamics of affective scene processing.” NeuroImage 261 (2022): 119532.)

summary : In this paper, they tried to diagnose ASD using semi-supervised autoencoder (SSAE). SSAE tries to reconstruct the features that are specifically important for downstream task (ASD vs HC), which is conceptually different than conventional AE. With sparsity constraints, the model can overcome SOTA performance in ABIDE classification. With a little bit of fine-tuning(ex. sparsity control), this method could be applied for our research using CHA data or p-factor related future research.

Yin, W., Li, L., & Wu, F. X. (2022). A semi-supervised autoencoder for autism disease diagnosis. Neurocomputing, 483, 140-147.)

summary : In this work, they utilize multimodal features of spoken words, speech prosody and facial expression from Aff-WIld2 dataset. They combine these features using a transformer-based fusion module which makes the output embedding features of sequences of images, audio and text. Integrated output feature is then processed in MLP layer for Action Unit (AU) detection and also facial expression recognition.

Zhang, Wei, et al. “Transformer-based multimodal information fusion for facial expression analysis.” Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2022.