Electroencephalogram (EEG)-based brain-computer interfaces face challenges in data insufficiency to train neural networks for tasks and generalizability for multiple subjects. To address the issue of data scarcity in EEG research, transfer learning using EEG foundation models (EFMs) has recently gained attention for its ability to leverage prior knowledge. Although transfer learning with EFMs enables tasks to be performed with limited training data, their increasing size presents significant computational challenges. Parameter-efficient transfer learning (PETL) methods address this computational issue by tuning only a small subset of parameters from the pre-trained model. However, existing PETL methods mostly fail to account for the high-dimensional nature of EEG data, which limits their ability to fully leverage the prior knowledge of the EFM when applied to downstream tasks. To address these challenges, we propose a novel PETL method with a TASk-relevanT fEature Focusing modULe (TASTEFUL) to transfer EFMs efficiently. TASTEFUL is designed to focus on task-relevant features and efficiently learn representations tailored for downstream tasks. We evaluated our proposed TASTEFUL on tasks using publicly available EEG datasets, demonstrating its superior performance. Finally, our work highlights TASTEFUL's potential to enhance the practical application of EFMs, marking a significant advancement in PETL for EFMs.

Time series data, widely used in fields such as climate studies, finance, and healthcare, often face scarcity in rare scenarios and privacy concerns, prompting growing interest in time series synthesis. Diffusion models have shown strong potential for generating high-quality data, but challenges remain in capturing long-range dependencies and complex patterns. We propose a novel diffusion model that integrates time-domain information with rich frequency-domain features, accounting for differences in noise decay rates across frequencies. Instead of arbitrary frequency splits used in prior works, we partition components based on spectral density, model them separately within the denoising backbone, and fuse them with time-domain features. This enables effective capture of both global and local patterns, enhancing representation of high- and low-frequency information. Extensive experiments on multiple public datasets show promising performance, and analyses including long-term generation and ablation studies demonstrate the model's ability to learn and represent complex time series distributions.

Causal inference involves determining how interventions affect outcomes and explaining the underlying mechanisms, and it holds critical importance across various fields. A key assumption in causal inference is that the measured covariates form a sufficient adjustment set. However, this assumption often fails due to unobserved confounders, as confounding mechanisms are rarely fully captured by measured covariates alone. Recent research has attempted to address this challenge using variational autoencoders (VAEs), but these approaches face practical limitations, including unidentifiability and bias toward proxy variables. To overcome these issues, we propose a novel method that incorporates quantized factor identifiability into VAEs for causal effect estimation. This integration mitigates unidentifiability and reduces the dominance of proxy variables, thereby enhancing consistency and accuracy in causal inference. Extensive experiments on both simulated and real-world datasets demonstrate the robustness and effectiveness of our method, establishing a new benchmark in deep causal modeling.

Face age transformation aims to convert reference images into synthesized images so that they portray the specified target ages. The crux of this task is to change only age-related areas of the given image while maintaining the age-irrelevant areas unchanged. Nevertheless, a common limitation among most existing models is the struggle to generate high-quality aging images that effectively consider both crucial properties. To address this problem, we propose a novel GAN-based face-aging framework that utilizes age-aware Guidance via Masking-Based Attention (GMBA). Specifically, we devise an age-aware guidance module to adjust age-relevant and age-irrelevant attributes within the image seamlessly. By virtue of its capability, it enables the model to produce realistic age-transformed images that certainly preserve the input's identities while delicately imposing age-related properties. Experimental results show that our proposed GMBA outperformed other state-of-the-art methods in terms of identity preservation and accurate age conversion, as well as providing superior visual quality for age-transformed images.

Recently, deep learning-based electroencephalogram (EEG) analysis and decoding have gained widespread attention to monitor a user's clinical condition or identify his/her intention/emotion. Nevertheless, the existing methods mostly model EEG signals with limited viewpoints or restricted concerns about the characteristics of the EEG signals, thus suffering from representing complex spatio-spectro-temporal patterns as well as inter-subject variability. In this work, we propose novel EEG-oriented self-supervised learning methods to discover complex and diverse patterns of spatio-spectral characteristics and spatio-temporal dynamics. Combined with the proposed self-supervised representation learning, we also devise a feature normalization strategy to resolve an inter-subject variability problem via clustering. We demonstrated the validity of the proposed framework on three publicly available datasets by comparing with state-of-the-art methods.

Alzheimer’s disease (AD) is widely aware as a neurodegenerative disease that is characterized as a leading cause of irreversible progressive dementia. From a clinical perspective, it is vital to forecast a patient’s progression over time. In that regard, there have been rigorous researches on AD progression modeling with structural magnetic resonance imaging (MRI). Methodologically, there are three major aspects of MRI modeling: (i) variability over time, (ii) sparseness in observations, and (iii) geometrical properties in temporal dynamics. While the existing deep-learning-based methods have addressed variability or sparsity in data, there is still a need to take into account the inherent geometrical properties. Recently, geometric modeling based on ordinary differential equations (ODE-RGRU) has shown its ability in various time-series data by combining an RNN and an ordinary differential equation (ODE) in symmetric positive definite (SPD) space. Despite the success of ODE-RGRU in time-series data modeling, it is limited to estimating the SPD matrix from sparse data with missing values. To this end, we propose a novel geometric learning framework for AD progression modeling to tackle the aforementioned issues simultaneously. And, we also propose training algorithms for manifold mapping on irregular and incomplete MRI and cognitive scores observations. Our proposed framework efficiently learns three major aspects of longitudinal MRI biomarker and cognitive scores by the manifold transformation module, ODE-RGRU, and missing value estimation module. We demonstrate the effectiveness of our method in experiments that forecast multi-class classification and cognitive scores over time. Additionally, we provide a multi-faceted analysis of the proposed method through an ablation study.

Recently, deep learning, a branch of machine learning and data mining, has gained widespread acceptance in many applications thanks to its unprecedented successes. In this regard, pioneering studies employed deep learning frameworks for imaging genetics in virtue of their own representation caliber. But, existing approaches suffer from some limitations: (i) exploiting a simple concatenation strategy for joint analysis, (ii) a lack of extension to biomedical applications, and (iii) insufficient and inappropriate interpretations in the viewpoint of both data science and bio-neuroscience. In this work, we propose a novel deep learning framework to tackle the aforementioned issues simultaneously. Our proposed framework learns to effectively represent the neuroimaging and the genetic data jointly, and achieves state-of-the-art performance in its use for Alzheimer’s disease and mild cognitive impairment identification. Further, unlike the existing methods in the literature, the framework allows learning the relation between imaging phenotypes and genotypes in a nonlinear way without any prior neuroscientific knowledge. To demonstrate the validity of our proposed framework, we conducted experiments on a publicly available dataset and analyzed the results from diverse perspectives. Based on our experimental results, we believe that the proposed framework has a great potential to give new insights and perspectives in deep learning-based imaging genetics studies.