Music is a universal phenomenon that influences human experiences across cultures. We investigate whether music can be decoded from human brain activity measured with fMRI, by modeling mappings between neural data and latent representations of musical stimuli. Our approach integrates functional and anatomical alignment techniques to facilitate cross-subject decoding. Starting from the GTZan fMRI dataset, where five participants listened to 540 musical tracks from 10 genres, we used the CLAP model to extract latent representations of the musical stimuli and developed voxel-wise encoding models to identify brain regions responsive to these stimuli, by applying a threshold to the correlation between predicted and actual brain activity. Our decoding pipeline, primarily retrieval-based, employs a linear map to project back brain activity to the corresponding CLAP features. This enables us to retrieve the musical stimuli most similar to those that originated the fMRI data. Our results demonstrate state-of-the-art identification accuracy, outperforming existing approaches.

Mixed-signal implementations of Spiking Neural Networks (SNNs) offer a promising solution to edge computing applications that require low-power and compact embedded processing systems. However, device mismatch in the analog circuits of these neuromorphic processors poses a significant challenge to the deployment of robust processing in these systems. Here we introduce a novel architectural solution inspired by biological development to address this issue. Specifically we propose to implement architectures that incorporate network motifs found in developed brains through a differentiable re-parameterization of weight matrices based on gene expression patterns and genetic rules. Thanks to the gradient descent optimization compatibility of the method proposed, we can apply the robustness of biological neural development to neuromorphic computing.We benchmark this approach using the Yin-Yang classification dataset, and compare its performance with that of standard multilayer perceptrons trained with state-of-the-art hardware-aware training method. Our results demonstrate that the proposed method mitigates mismatch-induced noise without requiring precise device mismatch measurements, effectively outperforming alternative hardware-aware techniques proposed in the literature, and providing a more general solution for improving the robustness of SNNs in neuromorphic hardware.

Decoding of visual stimuli from noninvasive neuroimaging techniques such as functional magnetic resonance (fMRI) has advanced rapidly in the last years; yet, most high-performing brain decoding models rely on complicated, non-interpretable latent spaces. In this study we present an interpretable brain decoding framework that inserts a semantic bottleneck into BrainDiffuser, a well established, simple and linear decoding pipeline. We firstly produce a $214-\text{dimensional}$ binary interpretable space $\mathcal{L}$ for images, in which each dimension answers to a specific question about the image (e.g., "Is there a person?", "Is it outdoors?"). A first ridge regression maps voxel activity to this semantic space. Because this mapping is linear, its weight matrix can be visualized as maps of voxel importance for each dimension of $\mathcal{L}$, revealing which cortical regions influence mostly each semantic dimension. A second regression then transforms these concept vectors into CLIP embeddings required to produce the final decoded image, conditioning the BrainDiffuser model. We found that voxel-wise weight maps for individual questions are highly consistent with canonical category-selective regions in the visual cortex (face, bodies, places, words), simultaneously revealing that activation distributions, not merely location, bear semantic meaning in the brain. Visual brain decoding performances are only slightly lower compared to the original BrainDiffuser metrics (e.g., the CLIP similarity is decreased by $\leq 4$% for the four subjects), yet offering substantial gains in interpretability and neuroscientific insights. These results show that our interpretable brain decoding pipeline enables voxel-level analysis of semantic representations in the human brain without sacrificing decoding accuracy.

Thanks to their ease of implementation, multilayer perceptrons (MLPs) have become ubiquitous in deep learning applications. The graph underlying an MLP is indeed multipartite, i.e. each layer of neurons only connects to neurons belonging to the adjacent layer. In contrast, in vivo brain connectomes at the level of individual synapses suggest that biological neuronal networks are characterized by scale-free degree distributions or exponentially truncated power law strength distributions, hinting at potentially novel avenues for the exploitation of evolution-derived neuronal networks. In this paper, we present “4Ward”, a method and Python library capable of generating flexible and efficient neural networks (NNs) from arbitrarily complex directed acyclic graphs. 4Ward is inspired by layering algorithms drawn from the graph drawing discipline to implement efficient forward passes, and provides significant time gains in computational experiments with various Erdős-Rényi graphs. 4Ward not only overcomes the sequential nature of the learning matrix method, by parallelizing the computation of activations, but also addresses the scalability issues encountered in the current state-of-the-art and provides the designer with freedom to customize weight initialization and activation functions. Our algorithm can be of aid for any investigator seeking to exploit complex topologies in a NN design framework at the microscale.

This study delves into the crucial aspect of network topology in artificial neural networks (NNs) and its impact on model performance. Addressing the need to comprehend how network structures influence learning capabilities, the research contrasts traditional multilayer perceptrons (MLPs) with models built on various complex topologies using novel network generation techniques. Drawing insights from synthetic datasets, the study reveals the remarkable accuracy of complex NNs, particularly in high-difficulty scenarios, outperforming MLPs. Our exploration extends to real-world datasets, highlighting the task-specific nature of optimal network topologies and unveiling trade-offs, including increased computational demands and reduced robustness to graph damage in complex NNs compared to MLPs. This research underscores the pivotal role of complex topologies in addressing challenging learning tasks. However, it also signals the necessity for deeper insights into the complex interplay among topological attributes influencing NN performance. By shedding light on the advantages and limitations of complex topologies, this study provides valuable guidance for practitioners and paves the way for future endeavors to design more efficient and adaptable neural architectures across various applications.

In medical image datasets, discrete labels are often used to describe a continuous spectrum of conditions, making unsupervised image stratification a challenging task. In this work, we propose VAESim, an architecture for image stratification based on a conditional variational autoencoder. VAESim learns a set of prototypical vectors during training, each associated with a cluster in a continuous latent space. We perform a soft assignment of each data sample to the clusters and reconstruct the sample based on a similarity measure between the sample embedding and the prototypical vectors. To update the prototypical embeddings, we use an exponential moving average of the most similar representations between actual prototypes and samples in the batch size. We test our approach on the MNIST handwritten digit dataset and the PneumoniaMNIST medical benchmark dataset, where we show that our method outperforms baselines in terms of kNN accuracy (up to +15% improvement in performance) and performs at par with classification models trained in a fully supervised way. Our model also outperforms current end-to-end models for unsupervised stratification.

Resting-state functional magnetic resonance imaging (rs-fMRI) has been successfully employed to understand the organisation of the human brain. Typically, the brain is parcellated into regions of interest (ROIs) and modelled as a graph where each ROI represents a node and association measures between ROI-specific blood-oxygen-level-dependent (BOLD) time series are edges. Recently, graph neural networks (GNNs) have seen a surge in popularity due to their success in modelling unstructured relational data. The latest developments with GNNs, however, have not yet been fully exploited for the analysis of rs-fMRI data, particularly with regards to its spatio-temporal dynamics. In this paper, we present a novel deep neural network architecture which combines both GNNs and temporal convolutional networks (TCNs) in order to learn from both the spatial and temporal components of rs-fMRI data in an end-to-end fashion. In particular, this corresponds to intra-feature learning (i.e., learning temporal dynamics with TCNs) as well as inter-feature learning (i.e., leveraging interactions between ROI-wise dynamics with GNNs). We evaluate our model with an ablation study using 35,159 samples from the UK Biobank rs-fMRI database, as well as in the smaller Human Connectome Project (HCP) dataset, both in a unimodal and in a multimodal fashion. We also demonstrate that out architecture contains explainability-related features which easily map to realistic neurobiological insights. We suggest that this model could lay the groundwork for future deep learning architectures focused on leveraging the inherently and inextricably spatio-temporal nature of rs-fMRI data.

Background and Significance: Mechanical alternans (MA) is a biomarker associated with mortality and life-threatening arrhythmias in heart failure patients. Despite showing prognostic value, its use is limited by the requirement of measuring the continuous blood pressure (BP), which is costly and impractical. Objective: To develop and test, for the first time, non-invasive MA surrogates based on photoplethysmography (PPG). Methods: Continuous BP and PPG were recorded during clinical procedures and tests in 35 patients. MA was induced either by ventricular pacing (Group A, N = 19) or exercise (Group B, N = 16). MA was categorized as sustained or intermittent if MA episodes were observed in at least 20 or between 12 and 20 consecutive beats, respectively. Eight features characterizing the pulse morphology were derived from the PPG, and MA surrogates were evaluated. Results: Sustained alternans was observed in 9 patients (47%) from Group A, whereas intermittent alternans was observed in 13 patients (68%) from Group A and in 10 patients (63%) from Group B. The PPG-based MA surrogate showing the highest accuracy, V1M, was based on the maximum of the first derivative of the PPG pulse. It detected both sustained and intermittent MA with 100% sensitivity and 100% specificity in Group A and intermittent MA with 100% sensitivity and 83% specificity in Group B. Furthermore, the magnitudes of MA and its PPG-based surrogate were linearly correlated (R2= 0.83, p <; 0.001). Conclusion: MA can be accurately identified non-invasively through PPG analysis. This may have important clinical implications for risk stratification and remote monitoring.

Mild cognitive impairment (MCI) is a common condition in patients with diffuse hyperintensities of cerebral white matter (WM) in T2-weighted magnetic resonance images and cerebral small vessel disease (SVD). In MCI due to SVD, the most prominent feature of cognitive impairment lies in degradation of executive functions, i.e., of processes that supervise the organization and execution of complex behavior. The trail making test is a widely employed test sensitive to cognitive processing speed and executive functioning. MCI due to SVD has been hypothesized to be the effect of WM damage, and diffusion tensor imaging (DTI) is a well-established technique for in vivo characterization of WM. We propose a machine learning scheme tailored to 1) predicting the impairment in executive functions in patients with MCI and SVD, and 2) examining the brain substrates of this impairment. We employed data from 40 MCI patients with SVD and created feature vectors by averaging mean diffusivity (MD) and fractional anisotropy maps within 50 WM regions of interest. We trained support vector machines (SVMs) with polynomial as well as radial basis function kernels using different DTI-derived features while simultaneously optimizing parameters in leave-one-out nested cross validation. The best performance was obtained using MD features only and linear kernel SVMs, which were able to distinguish an impaired performance with high sensitivity (72.7%-89.5%), specificity (71.4%-83.3%), and accuracy (77.5%-80.0%). While brain substrates of executive functions are still debated, feature ranking confirm that MD in several WM regions, not limited to the frontal lobes, are truly predictive of executive functions.