Foundation Models have demonstrated significant success across various domains in Artificial Intelligence (AI), yet their capabilities for brainwave modeling remain unclear. In this paper, we comprehensively evaluate current Large Brainwave Foundation Models (LBMs) through systematic fine-tuning experiments across multiple Brain-Computer Interface (BCI) benchmark tasks, including memory tasks and sleep stage classification. Our extensive analysis shows that state-of-the-art LBMs achieve only marginal improvements (0.5\%) over traditional deep architectures while requiring significantly more parameters (millions vs thousands), raising important questions about their efficiency and applicability in BCI contexts. Moreover, through detailed ablation studies and Low-Rank Adaptation (LoRA), we significantly reduce trainable parameters without performance degradation, while demonstrating that architectural and training inefficiencies limit LBMs' current capabilities. Our experiments span both full model fine-tuning and parameter-efficient adaptation techniques, providing insights into optimal training strategies for BCI applications. We pioneer the application of LoRA to LBMs, revealing that performance benefits generally emerge when adapting multiple neural network components simultaneously. These findings highlight the critical need for domain-specific development strategies to advance LBMs, suggesting that current architectures may require redesign to fully leverage the potential of foundation models in brainwave analysis.

Error-Related Potentials (ErrPs) have been used lately in order to improve several existing Brain-Computer Interface (BCI) applications. In our study we investigate the contribution of ErrPs in a Steady State Visual Evoked Potential (SSVEP) based BCI. An extensive study is presented in order to discover the limitations of the proposed scheme. Using Common Spatial Patterns and Random Forests we manage to show encouraging results regarding the incorporation of ErrPs in a SSVEP system. Finally, we provide a novel methodology (Inverse Correct Response Time) that can measure the gain of a BCI system by incorporating ErrPs in terms of time efficiency.

Dynamic recordings of functional activity maps can naturally and efficiently be represented in the form of functional/effective connectivity networks. New methods for mapping synaptic connections and recording neural signals generate rich and complex data about the structure and dynamics of brain networks. To study the most complex network in nature, the brain, there is need to integrate a huge amount of brain networks collected from laboratories over the world in large databases. Human Brain Project (Europe and USA) aims to explore brain functionality in various ways. Brain networks are central to achieving the goals of this ambitious plan. However, the immense amount of thousands of brain networks, prevent an easy way to utilizable knowledge. In this paper, we demonstrate a data-driven approach that discovers consistent patterns from a collection of brain networks via a querying approach: formulating a query of “finding an increasing or a decreasing consistent subgraph over an amount of subjects” after taking the difference between two sets of graphs referred as two conditions (an active and a baseline). Experiments demonstrated that our data-driven approach allows identifying frequency-dependent selective spatial pattern changes of the EEG functional connectivity network during a mental task. This is the first time that a method fully exploits the connectivity weights of a brain network to discover consistent subgraph patterns.

The Web as a graph can be embedded in a low-dimensional space where its geometry can be visualized and studied in order to mine interesting patterns such as web communities. The existing algorithms operate on small-to-medium-scale graphs; thus, we propose a close to linear time algorithm called Mani-Web suitable for large-scale graphs. The result is similar to the one produced by the manifold-learning technique Laplacian eigenmap that is tested on artificial manifolds and real web-graphs. Mani-Web can also be used as a general-purpose manifold-learning/dimensionality-reduction technique as long as the data can be represented as a graph.

In this paper, we propose novel algorithms for low dimensionality nonnegative embedding of vectorial and/or relational data, as well as nonnegative projections for dimensionality reduction. We start by introducing a novel algorithm for Metric Multidimensional Scaling (MMS). We propose algorithms for Nonnegative Locally Linear Embedding (NLLE) and Nonnegative Laplacian Eigenmaps (NLE). By reformulating the problem of MMS, NLLE and NLE for finding projections we propose algorithms for Nonnegative Principal Component Analysis (NPCA), for Nonnegative Orthogonal Neighbourhood Preserving Projections (NONPP) and Nonnegative Orthogonal Locality Preserving Projections (NOLPP). We demonstrate some first preliminary results of the proposed methods in data visualization.

In this letter, we suggest a novel method for clustering, based on finding the smallest enclosing hyperellipse in arbitrary Hilbert spaces. In particular, we show that the one class support vector method that finds the minimum bounding hypersphere, under the whitening transform, becomes a method for finding the minimum bounding hyperellipse. Afterwards, we generalize the method in order to find the minimum bounding hyperellipse in arbitrary Hilbert spaces. We illustrate the power of the proposed methods in clustering applications.

In this work, the idea of key frames extraction from single shots in video sequences is presented. The method is implemented by an efficient two-step algorithm, which is classified neither to clustering nor to temporal variations based techniques. In the first step, an MST (minimal spanning tree) graph is constructed, where each node is associated to a single frame of the shot. In the second step, extracts key frames based on the principle of their maximum spread, are extracted. The number of the selected key frames is controlled by an adaptively defined threshold, while the validity of the results is evaluated by the fidelity measure.

A novel strategy for color-based image retrieval is introduced. Initially, a vector quantization technique is adopted, based on the application of self-organizing neural networks. The color content in each image is summarized by representative RGB-vectors extracted using the Neural-Gas network, an efficient way to extract faithful representations from multivariate distributions. The similarity between two images is then assessed as commonality between the corresponding representative color distributions and quantified using the multivariate Wald-Wolfowitz test, a nonparametric statistical test dealing with the "multivariate two-sample problem". Experimental results drawn from a diverse collection of color images show a significantly improved performance relative to the popular approach of color histogram.

In this study, a conceptually simple, yet flexible and extendable strategy to contrast two different color images is introduced. The proposed approach is based on the multivariate Wald-Wolfowitz test, a nonparametric test that assesses the commonality between two different sets of multivariate observations. It provides an aggregate gauge of the match between color images, taking into consideration all the (selected) low-level characteristics, while alleviating correspondence issues. We show that a powerful measure of similarity between two color images can emerge from the statistical comparison of their representations in a properly formed feature space. For the sake of simplicity, the RGB-space is selected as the feature space, while we are experimenting with different ways to represent the images within this space. By altering the feature-extraction implementation, complementary ways to portray the image content appear. The reported results, from the application on a diverse collection of images, clearly demonstrate the effectiveness of our method, its superiority over previous methods, and suggest that even further improvements can be achieved along the same line of research. It is not only the unifying character that makes our strategy appealing, but also the fact that the retrieval performance does not increase continuously with the amount of details in the image representation. The latter sets an upper limit to the computational demands and reminds of performance plateaus reached by novel approaches in information retrieval.

In this paper, a new filter that performs color image enhancement is presented. The filter achieves this through the minimization of a weighted cost function. The weights are determined using potential functions which are calculated in such a way as to convey spatial information. Application of the proposed filter on a real blurred and noisy color image is performed to verify its enhancement capabilities.

This letter deals with the color image edge enhancement issue using clustering ideas and based on the use of potential functions (Parzen windows). Two algorithms are proposed. The first uses potential functions (PF's) and selects the output as the vector maximizing the PF. The second one elaborates further by employing the mountain clustering method and modifying it appropriately. Both algorithms are robust in the presence of noise, Gaussian and impulsive.

Vector filtering processes are of special importance when correlated multichannel data are of interest. In this case the data sequence is taken as a whole. The idea of using vectors instead of scalers has been used extensively. This work presents a method where data dependent coefficients are derived at each time instant, upon which the filter output is computed. This nonlinear filter has properties in between the vector median and other nonlinear vector filters and similar to linear ones.>