Maria Papadopouli

Aditi Aravind, Konstantinos Ladakis, Mario Alexios Savaglio et al.

We investigate how internal representations emerge across hierarchical processing systems by introducing a neuroscience-inspired framework for analyzing deep spiking neural networks (SNN) through the lens of functional connectivity. Drawing on concepts from systems neuroscience and information theory, we form the first-order functionally-connected (1FC) group of a neuron based on its statistically significant pairwise correlations with neurons from the previous layer of a trained SNN architecture. We then track its response properties during inference under various conditions. Our analysis shows that several principles of functional connectivity previously observed in biological cortex are preserved in spiking ResNet architectures. These 1FC ensembles display interesting properties: their aggregate cofiring reliably predicts downstream neuronal responses through a robust, ReLU-like input-output relationship, whose gain scales systematically with ensemble size. Reliable encoding of the presented class emerges only during high 1FC cofiring events, which themselves occur infrequently, indicating that informative representations are concentrated in rare but highly coordinated activity patterns. Under uniform random noise or adversarial perturbations, these response profiles are disrupted, particularly in early and intermediate layers. This enables a targeted high-resolution interrogation at specific nodes and pathways. We showed that the functional connectivity structure is shaped by learning and this structure breaks under weight permutation. These establish 1FC ensembles as a functionally meaningful substrate for input encoding and information transfer, with potential implications in designing targeted fine-grained diagnostics on the information flow.

Orestis Konstantaropoulos, Theodoris Mallios, Maria Papadopouli

While foundation AI models excel at tasks like classification and decision-making, their high energy consumption makes them unsuitable for energy-constrained applications. Inspired by the brain's efficiency, spiking neural networks (SNNs) have emerged as a viable alternative due to their event-driven nature and compatibility with neuromorphic chips. This work introduces a novel system that combines knowledge distillation and ensemble learning to bridge the performance gap between artificial neural networks (ANNs) and SNNs. A foundation AI model acts as a teacher network, guiding smaller student SNNs organized into an ensemble, called Spiking Neural Ensemble (SNE). SNE enables the disentanglement of the teacher's knowledge, allowing each student to specialize in predicting a distinct aspect of it, while processing the same input. The core innovation of SNE is the adaptive activation of a subset of SNN models of an ensemble, leveraging knowledge-distillation, enhanced with an informed-partitioning (disentanglement) of the teacher's feature space. By dynamically activating only a subset of these student SNNs, the system balances accuracy and energy efficiency, achieving substantial energy savings with minimal accuracy loss. Moreover, SNE is significantly more efficient than the teacher network, reducing computational requirements by up to 20x with only a 2% drop in accuracy on the CIFAR-10 dataset. This disentanglement procedure achieves an accuracy improvement of up to 2.4% on the CIFAR-10 dataset compared to other partitioning schemes. Finally, we comparatively analyze SNE performance under noisy conditions, demonstrating enhanced robustness compared to its ANN teacher. In summary, SNE offers a promising new direction for energy-constrained applications.

Orestis Konstantaropoulos, Stelios Manolis Smirnakis, Maria Papadopouli

The structure of biological neural circuits-modular, hierarchical, and sparsely interconnected-reflects an efficient trade-off between wiring cost, functional specialization, and robustness. These principles offer valuable insights for artificial neural network (ANN) design, especially as networks grow in depth and scale. Sparsity, in particular, has been widely explored for reducing memory and computation, improving speed, and enhancing generalization. Motivated by systems neuroscience findings, we explore how patterns of functional connectivity in the mouse visual cortex-specifically, ensemble-to-ensemble communication, can inform ANN design. We introduce G2GNet, a novel architecture that imposes sparse, modular connectivity across feedforward layers. Despite having significantly fewer parameters than fully connected models, G2GNet achieves superior accuracy on standard vision benchmarks. To our knowledge, this is the first architecture to incorporate biologically observed functional connectivity patterns as a structural bias in ANN design. We complement this static bias with a dynamic sparse training (DST) mechanism that prunes and regrows edges during training. We also propose a Hebbian-inspired rewiring rule based on activation correlations, drawing on principles of biological plasticity. G2GNet achieves up to 75% sparsity while improving accuracy by up to 4.3% on benchmarks, including Fashion-MNIST, CIFAR-10, and CIFAR-100, outperforming dense baselines with far fewer computations.

Eirini Troullinou, Grigorios Tsagkatakis, Ganna Palagina et al.

The field of neuroscience is experiencing rapid growth in the complexity and quantity of the recorded neural activity allowing us unprecedented access to its dynamics in different brain areas. The objective of this work is to discover directly from the experimental data rich and comprehensible models for brain function that will be concurrently robust to noise. Considering this task from the perspective of dimensionality reduction, we develop an innovative, robust to noise dictionary learning framework based on adversarial training methods for the identification of patterns of synchronous firing activity as well as within a time lag. We employ real-world binary datasets describing the spontaneous neuronal activity of laboratory mice over time, and we aim to their efficient low-dimensional representation. The results on the classification accuracy for the discrimination between the clean and the adversarial-noisy activation patterns obtained by an SVM classifier highlight the efficacy of the proposed scheme compared to other methods, and the visualization of the dictionary's distribution demonstrates the multifarious information that we obtain from it.

Vasileios Sideridis, Andrew Zacharakis, George Tzagkarakis et al.

This paper introduces and evaluates the GestureKeeper, a robust hand-gesture recognition system based on a wearable inertial measurements unit (IMU). The identification of the time windows where the gestures occur, without relying on an explicit user action or a special gesture marker, is a very challenging task. To address this problem, GestureKeeper identifies the start of a gesture by exploiting the underlying dynamics of the associated time series using a recurrence quantification analysis (RQA). RQA is a powerful method for nonlinear time-series analysis, which enables the detection of critical transitions in the system's dynamical behavior. Most importantly, it does not make any assumption about the underlying distribution or model that governs the data. Having estimated the gesture window, a support vector machine is employed to recognize the specific gesture. Our proposed method is evaluated by means of a small-scale pilot study at FORTH and demonstrated that GestureKeeper can identify correctly the start of a gesture with a 87\% mean balanced accuracy and classify correctly the specific hand-gesture with a mean accuracy of over 96\%. To the best of our knowledge, GestureKeeper is the first automatic hand-gesture identification system based only on accelerometer. The performance analysis reveals the predictive power of the features and the system's robustness in the presence of additive noise. We also performed a sensitivity analysis to examine the impact of various parameters and a comparative analysis of different classifiers (SVM, random forests). Most importantly, the system can be extended to incorporate a large dictionary of gestures and operate without further calibration for a new user.

Thomais Asvestopoulou, Victoria Manousaki, Antonis Psistakis et al.

Eye movements during text reading can provide insights about reading disorders. Via eye-trackers, we can measure when, where and how eyes move with relation to the words they read. Machine Learning (ML) algorithms can decode this information and provide differential analysis. This work developed DysLexML, a screening tool for developmental dyslexia that applies various ML algorithms to analyze fixation points recorded via eye-tracking during silent reading of children. It comparatively evaluated its performance using measurements collected in a systematic field study with 69 native Greek speakers, children, 32 of which were diagnosed as dyslexic by the official governmental agency for diagnosing learning and reading difficulties in Greece. We examined a large set of features based on statistical properties of fixations and saccadic movements and identified the ones with prominent predictive power, performing dimensionality reduction. Specifically, DysLexML achieves its best performance using linear SVM, with an a accuracy of 97 %, with a small feature set, namely saccade length, number of short forward movements, and number of multiply fixated words. Furthermore, we analyzed the impact of noise on the fixation positions and showed that DysLexML is accurate and robust in the presence of noise. These encouraging results set the basis for developing screening tools in less controlled, larger-scale environments, with inexpensive eye-trackers, potentially reaching a larger population for early intervention.

Evripidis Tzamousis, Maria Papadopouli

The recommendation systems aim to improve the user engagement by recommending appropriate personalized content to users, exploiting information about their preferences. We propose the enabler, a hybrid recommendation system which employs various machine-learning (ML) algorithms for learning an efficient combination of several recommendation algorithms and selects the best blending for a given input.Specifically, it integrates three layers, namely, the trainer which trains the underlying recommenders, the blender which determines the most efficient combination of the recommenders, and the tester for assessing the performance of the system. The enabler incorporates a variety of recommendation algorithms that span from collaborative filtering and content-based techniques to ones based on neural networks. It uses the nested cross validation for automatically selecting the best ML algorithm along with its hyper-parameter values for the given input, according to a specific metric. The enabler can be easily extended to include other recommenders and blenders. The enabler has been extensively evaluated in the context of video-streaming. It outperforms various other algorithms, when tested on the Movielens 1M benchmark dataset.encouraging results. Moreover For example, it achieves an RMSE of 0.8206, compared to the state-of-the-art performance of the AutoRec and SVD, 0.827 and 0.845, respectively. A pilot web-based recommendation system was developed and tested in the production environment of a large telecom operator in Greece. Volunteer customers of the video-streaming service provided by the telecom operator employed the system in the context of an out-in-the-wild field study with a post-analysis of the enabler, using the collected ratings of the pilot, demonstrated that it significantly outperforms several popular recommendation algorithms.