Neural networks and motor memories: new insights for Parkinson's treatment

New comment in Nature reviews neuro by Sadra Sadeh and Claudia Clopath, “The emergence of NeuroAI: bridging neuroscience and artificial intelligence.” (https://lnkd.in/eaYAfcr4). This short piece describes the future roadmap of NeuroAI. It shows how progress will focus on better ways to organise and analyse data, on modelling and predicting neural activity, and on generating (as well as prioritising) testable hypotheses. It also highlights the usual AI challenges of interpretability and reproducibility, and, something that resonates with my background, the need for neuroscientists that truly understand both fields. I would add to their list the promise for drug discovery. When machine learning is applied to pathology-centred big data, it can integrate terabyte- to petabyte-scale, multi-modal readouts in disease contexts, from neural activity to behaviour and molecular readouts. From this, one can extract disease signatures that act as an ultimate biomarker both in preclinical animal models and in humans. With that signal, we can propose and rank therapeutic hypotheses, prioritise targets, predict responders and design studies. It suggests a way of redefining drug discovery in neuroscience, where data integration drives the questions rather than following them.

TL;DR - educational neuroscience is fascinating and important, but at the moment there's still a big gap between analyses of the brain's physical structure and how cognition functions. Be wary of overly simplistic "the brain lights up during this cognitive process, so we should teach this way" recommendations and explanations. https://lnkd.in/eMuHnwei

Brain-HGCN: A Hyperbolic Graph Convolutional Network for Brain Functional Network Analysis https://lnkd.in/e-f_VTfR Functional magnetic resonance imaging (fMRI) provides a powerful non-invasive window into the brain's functional organization by generating complex functional networks, typically modeled as graphs. These brain networks exhibit a hierarchical topology that is crucial for cognitive processing. However, due to inherent spatial constraints, standard Euclidean GNNs struggle to represent these hierarchical structures without high distortion, limiting their clinical performance. To address this limitation, we propose Brain-HGCN, a geometric deep learning framework based on hyperbolic geometry, which leverages the intrinsic property of negatively curved space to model the brain's network hierarchy with high fidelity. Grounded in the Lorentz model, our model employs a novel hyperbolic graph attention layer with a signed aggregation mechanism to distinctly process excitatory and inhibitory connections, ultimately learning robust graph-level representations via a geometrically sound Fréchet mean for graph readout. Experiments on two large-scale fMRI datasets for psychiatric disorder classification demonstrate that our approach significantly outperforms a wide range of state-of-the-art Euclidean baselines. This work pioneers a new geometric deep learning paradigm for fMRI analysis, highlighting the immense potential of hyperbolic GNNs in the field of computational psychiatry. --- Newsletter https://lnkd.in/emCkRuA More story https://lnkd.in/enY7VpM LinkedIn https://lnkd.in/ehrfPYQ6 #AINewsClips #AI #ML #ArtificialIntelligence #MachineLearning #ComputerVision

Your Brain Thinks Faster Than You Realize… And New Research Proves It! Computational Neuroscience researchers have discovered that our brain doesn’t just solve problems—it predicts the future at the neural level! Advanced computational models show that neural networks in our brain exchange information faster than any supercomputer, helping us understand learning, memory, and decision-making in a completely new way. This breakthrough could revolutionize AI, human-computer interaction, and treatments for neurological disorders. If our brains can truly predict the future… what would you love to see this applied to first? #Neuroscience #BrainResearch #ComputationalNeuroscience #AI #NeuroTech #Innovation

🚨 Big step forward for neuroscience. For the first time, scientists have mapped the activity of single neurons across the entire brain during decision-making. That means recording from 600,000+ neurons in 279 brain areas — covering about 95% of the mouse brain volume. An incredible scale that just a few years ago would have sounded impossible. This achievement gives us a first real glimpse of how distributed brain circuits work together to guide behaviour. Exciting times ahead for neuroscience, network science, and computational modeling! https://lnkd.in/dh3McUwu

Recurrence affects the geometry of visual representations across the ventral visual stream in the human brain https://lnkd.in/emRZYC4R by Siying Xie et al. "We find that recurrent processing substantially shapes visual representations across the ventral visual stream, starting early on at around 100 ms in early visual cortex (EVC) and two later phases of around 175 and 300 ms in lateral occipital cortex (LOC), adding persistent rather than transient neural dynamics to visual processing." Could (neuromodulation of) recurrence extend to auditory areas? https://lnkd.in/eKrTTzK3 (Shape conveyed by visual-to-auditory sensory substitution activates the lateral occipital complex, 2007.) #neuroscience #recurrence #neuromodulation #vision

It's a mind-boggling fact – the human brain generates enough electrical energy to power a small light bulb! This incredible statistic highlights the immense electrical activity occurring within our neural networks every second. But what does this mean for neuroscience research? By leveraging advanced EEG

📣 The next GDR NeuralNet meeting will take place 𝗶𝗻 𝗕𝗼𝗿𝗱𝗲𝗮𝘂𝘅 on 𝗡𝗼𝘃 𝟮𝟲-𝟮𝟴, 𝟮𝟬𝟮𝟱. The deadline for early registration has been postponed to Sept 22, 2025. 📄 The objective of the NeuralNet Research Cluster (GDR CNRS) is to develop and promote synergies between researchers, engineers, and other stakeholders engaged in measuring, manipulating, analyzing and interpreting neuronal activity at the level of networks. ℹ️ This year's meeting is revolving around 𝟰 𝘁𝗿𝗮𝗻𝘀𝘃𝗲𝗿𝘀𝗮𝗹 𝘁𝗵𝗲𝗺𝗲𝘀 𝗧𝗶𝗺𝗲𝘀𝗰𝗮𝗹𝗲𝘀: 𝗕𝗲𝗵𝗮𝘃𝗶𝗼𝗿𝗮𝗹 𝘃𝘀. 𝗻𝗲𝘂𝗿𝗮𝗹 with Fanny Cazettes (INT, FR), Sami El-Boustani (UNIGE,CH), Nikolas Karalis (ICM, FR) and more. 𝗡𝗮𝘁𝘂𝗿𝗮𝗹 𝗯𝗲𝗵𝗮𝘃𝗶𝗼𝗿: 𝗘𝗰𝗼-𝗲𝘁𝗵𝗼-𝗹𝗼𝗴𝗶𝗰𝗮𝗹 𝗮𝗽𝗽𝗿𝗼𝗮𝗰𝗵𝗲𝘀 𝘃𝘀. 𝗻𝗲𝘂𝗿𝗮𝗹 𝗰𝗼𝗻𝘁𝗿𝗼𝗹. with Richard Hahnloser (INI, CH), Julia Sliwa (ICM, FR), Heike Stein (ISIR, FR) and more. 𝗚𝗲𝗻𝗲𝗿𝗮𝗹 𝗣𝗿𝗶𝗻𝗰𝗶𝗽𝗹𝗲𝘀 𝗼𝗳 𝗮𝗰𝘁𝗶𝘃𝗲 𝘀𝗲𝗻𝘀𝗶𝗻𝗴 with Guy Bouvier (Neuropsi,FR), Emmanuelle Courtiol (CRNL, FR), Mathew Diamond (SISSA, IT) and more. 𝗦𝘆𝘀𝘁𝗲𝗺𝗮𝘁𝗶𝗰 𝗮𝗽𝗽𝗿𝗼𝗮𝗰𝗵: 𝗠𝗼𝗱𝘂𝗹𝗮𝗿 𝘃𝘀 𝗱𝗶𝘀𝘁𝗿𝗶𝗯𝘂𝘁𝗲𝗱 𝗰𝗼𝗱𝗶𝗻𝗴 with David Dupret(Oxford, UK),Stephanie Forkel(Radboud, NL), McKenzie Mathis (EPFL, CH) and more. Organizing committee: Julien Courtin (Neurocentre Magendie), Xavier Hinaut (Institute of Neurodegenerative Diseases (IMN)), Frédéric Lanore (Interdisciplinary Institute for Neuroscience), Catherine Le Moine (Institut de Neurosciences Cognitives et Intégratives d'Aquitaine - INCIA) 👉 Program, registration, abstract submission: https://lnkd.in/gtm94RWv

AI’s struggle to read analog clocks is striking, and this is one of the first cognitive functions to decline in human dementia. That parallel makes ClockBench more than just a quirky test of visual reasoning. It raises fascinating questions about whether the same kinds of mechanisms are at play - and whether progress in AI on tasks like this will echo advances in other forms of complex cognition. I’ll drop the benchmark link in the comments. Any thoughts or insights from my neuroscience friends? h/t Alek Safar

🚨 𝗡𝗲𝘄 𝗥𝗲𝘀𝗲𝗮𝗿𝗰𝗵: 𝗧𝗲𝗺𝗽𝗼𝗿𝗮𝗹 𝗗𝘆𝗻𝗮𝗺𝗶𝗰𝘀 𝗼𝗳 𝗛𝘂𝗺𝗮𝗻 𝗘𝗰𝗵𝗼𝗹𝗼𝗰𝗮𝘁𝗶𝗼𝗻 🦇 🗣️ 👩🦯 🧠 How does the brain transform successive echoes into a coherent spatial representation? In our latest preprint with Santani Teng at #SKERI, we combined 𝗘𝗘𝗚 and 𝗽𝘀𝘆𝗰𝗵𝗼𝗽𝗵𝘆𝘀𝗶𝗰𝘀 to uncover the 𝘁𝗲𝗺𝗽𝗼𝗿𝗮𝗹 𝗱𝘆𝗻𝗮𝗺𝗶𝗰𝘀 𝗼𝗳 𝘀𝗽𝗮𝘁𝗶𝗮𝗹 𝗶𝗻𝗳𝗼𝗿𝗺𝗮𝘁𝗶𝗼𝗻 𝗮𝗰𝗰𝘂𝗺𝘂𝗹𝗮𝘁𝗶𝗼𝗻 in click-based echolocation. 𝗞𝗲𝘆 𝗙𝗶𝗻𝗱𝗶𝗻𝗴𝘀 - Blind expert echolocators significantly outperformed sighted controls in localizing spatialized echoes. - Localization thresholds improved with more clicks — evidence of 𝗰𝘂𝗺𝘂𝗹𝗮𝘁𝗶𝘃𝗲 𝗶𝗻𝘁𝗲𝗴𝗿𝗮𝘁𝗶𝗼𝗻 of spatial cues. - EEG decoding revealed that the 𝗳𝗶𝗿𝘀𝘁 𝗰𝗹𝗶𝗰𝗸 already contained spatial information, with neural activity evolving systematically across successive clicks. - Neural responses mirrored behavioral improvements, providing the first detailed account of 𝗽𝗲𝗿𝗰𝗲𝗽𝘁𝘂𝗮𝗹 𝗶𝗻𝗳𝗼𝗿𝗺𝗮𝘁𝗶𝗼𝗻 𝗮𝗰𝗰𝘂𝗺𝘂𝗹𝗮𝘁𝗶𝗼𝗻 𝗼𝘃𝗲𝗿 𝘁𝗶𝗺𝗲. These results reveal the 𝗻𝗲𝘂𝗿𝗮𝗹 𝗮𝗻𝗱 𝗯𝗲𝗵𝗮𝘃𝗶𝗼𝗿𝗮𝗹 𝗳𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻𝘀 of echolocation and showcase the brain’s remarkable adaptability without vision. https://lnkd.in/ggQkgb47 #Neuroscience #AuditoryResearch #Echolocation #EEG #Neuroplasticity #BlindnessResearch