VerAI Discoveries’ Post

"Attending to Topological Spaces: The Cellular Transformer" by Rubén Bautista Ballester, Mathilde Papillon, Claudio Battiloro, Nina Miolane, PhD, Tolga Birdal, Carles Casacuberta, Sergio Escalera, Mustafa Hajij et al. "Topological Deep Learning seeks to enhance the predictive performance of neural network models by harnessing topological structures in input data. Topological neural networks operate on spaces such as cell complexes and hypergraphs, that can be seen as generalizations of graphs. In this work, we introduce the Cellular Transformer (CT), a novel architecture that generalizes graph-based transformers to cell complexes. First, we propose a new formulation of the usual self- and cross-attention mechanisms, tailored to leverage incidence relations in cell complexes, e.g., edge-face and node-edge relations. Additionally, we propose a set of topological positional encodings specifically designed for cell complexes. By transforming three graph datasets into cell complex datasets, our experiments reveal that CT not only achieves state-of-the-art performance, but it does so without the need for more complex enhancements such as virtual nodes, in-domain structural encodings, or graph rewiring." Paper: https://lnkd.in/eqJ27QNz #machinelearning #topologicaldeeplearning

How can AI help us better understand the ionosphere? 🤖🌍 Remote Sensing MDPI Researchers have developed a cutting-edge machine learning model called MMAdapGA-BP-NN that can predict ionospheric total electron content (TEC) with unprecedented accuracy! 📈 Compared to existing methods like the international reference ionosphere (IRI), this combined model reduced forecast errors by over 50% in tests. 🤯 And it even outperformed standalone neural network models by 25%. 🚀 What makes this approach so powerful? It blends techniques like statistical machine learning, genetic algorithms, and neural networks to create a highly adaptable and stable prediction system. 🧠 Could this be the future of ionospheric forecasting? Check out the full video overview to see this AI-powered innovation in action! 👇 https://lnkd.in/geDtPCce This video is adapted from https://lnkd.in/gmHNi6V2 #machinelearning #ionosphere #spaceweather #remotesensing #TEC

🌍 Successful EOP I&I Scientific Meeting – Recording Now Available! 🎥 I'm excited to share that our second Earth Orientation Parameters Innovation and Insight (EOP I&I) Scientific Meeting, held on August 7, 2024, was a success! 🚀 A big thank you to Mostafa Kiani Shahvandi from ETH Zurich for his outstanding lecture on "Explaining the Causes of Polar Motion by Physics-Informed Neural Networks." His insights into polar motion using advanced machine learning techniques were truly enlightening. For those who couldn't join us live, you can now watch the recording on GGOS's YouTube channel: 🔗 Explaining the Causes of Polar Motion by Physics-Informed Neural Networks https://lnkd.in/emfcqTci Missed our first meeting? Catch up here: 🔗 Recurrent Neural Networks and Their Applications for EOP Prediction https://lnkd.in/eturz892 We’re excited to continue this series, featuring leading researchers who are advancing the application of AI in EOP prediction. Stay tuned for more! Chair of Space Geodesy, ETH Zurich GGOS - Global Geodetic Observing System Geodesy Research Group of Alicante University Bundesamt für Kartographie und Geodäsie (BKG) #AI #MachineLearning #Geodesy #EOP #PolarMotion #GGOS #PhysicsInformedAI #NeuralNetworks #ScientificResearch #Innovation

It’s amazing how clear this video from 1983 still is! 😀 Prof. John Hopfield explains neural networks in a way that's easy to follow. Even though the video is so old, We’re lucky to have such a high-quality resource from the early days of neural networks: 🔗 Watch it here: [Prof. Hopfield on Neural Networks](https://lnkd.in/g3SH3dnn) For those who did not read: here's the original paper published in PNAS:: [Hopfield's Paper in PNAS](https://lnkd.in/gVd8jrC5) #NeuralNetworks #AIHistory #DeepLearning #MachineLearning #AIResearch #TechNostalgia #PNAS #AIpioneers

The 2024 Nobel Prize in Physics was awarded to John J. Hopfield and Geoffrey E. Hinton for their groundbreaking contributions to machine learning and artificial neural networks. John Hopfield developed an associative memory model capable of storing and reconstructing images and various data patterns. Geoffrey Hinton introduced a method that allows machines to autonomously discover patterns in data, enabling tasks such as recognizing specific elements within images. #NoblePrize #MachineLearning #DeepLearning #AI #GenerativeAI #2024 #physics

𝓑𝓲𝓰 𝓬𝓸𝓷𝓰𝓻𝓪𝓽𝓾𝓵𝓪𝓽𝓲𝓸𝓷𝓼 to the AI pioneers who won the 𝟐𝟎𝟐𝟒 𝐍𝐨𝐛𝐞𝐥 𝐏𝐫𝐢𝐳𝐞 𝐢𝐧 𝐏𝐡𝐲𝐬𝐢𝐜𝐬 for their amazing contributions to this field! • 𝐉𝐨𝐡𝐧 𝐇𝐨𝐩𝐟𝐢𝐞𝐥𝐝 created a type of artificial neural network in the 1980s, called the Hopfield Network. This system helps us understand how the brain stores and remembers information, which has been useful in both AI and brain science. • 𝐆𝐞𝐨𝐟𝐟𝐫𝐞𝐲 𝐇𝐢𝐧𝐭𝐨𝐧 took it further with the Boltzmann Machine, which was one of the first ways that computers could learn and recognize patterns. This work laid the foundation for the AI we use today, like in image and speech recognition. Their work has changed the world of AI and is helping scientists explore everything from biology to how the brain works. Their achievements remind us how important AI can be when used wisely! #NobelPrize #Physics #AI #Innovation #MachineLearning

🚀 Happy to share a slice of what we’ve been working on! Our paper, "Unveiling the Unseen: Identifiable Clusters in Trained Depthwise Convolutional Kernels", appeared at ICLR this year. 🎉We'd love to dive deeper into our discoveries with you! 📌 Poster Session 1: Location: Hall B #16 Date: Tomorrow, May 7th Time: 10:45 a.m. – 12:45 p.m. CEST We explore depthwise-separable convolutional neural networks (DS-CNNs) and have uncovered identifiable patterns in their convolutional kernels that mirror foundational principles observed in mammalian vision. Our findings reveal that these patterns, which mathematically align with Gaussian and derivatives kernels known from scale-space theory, categorize over 90% of the kernels in cutting-edge models like ConvNeXtV2 and ConvNeXt into distinct clusters. This breakthrough not only advances our understanding of CNNs but also hints that they mimic biological vision more closely than thought before. 🌟 Tagging my amazing coauthor Peyman Mohseni Kiasari Check out our full paper here: https://lnkd.in/dSJAtNBd Looking forward to some engaging discussions! #ICLR2024 #AI #DeepLearning #ConvNeXt #BiologicallyInspired

🏆 Nobel Award: Redefining Physics Boundaries 🌌 In this insightful video, we explore the groundbreaking contributions of John Hopfield and Geoffrey Hinton, pioneers in artificial intelligence and deep learning. Their innovative work has bridged the gap between AI and physics, revolutionizing how we process information and understand complex systems. 🔬 Highlights: -John Hopfield: Known for the Hopfield Network, which laid the foundation for neural network research, influencing both theoretical and practical applications in AI. -Geoffrey Hinton: Often referred to as the "Godfather of Deep Learning," his research has transformed AI, making it a driving force in technological advancements today. Discover how their achievements are reshaping our understanding of science and technology! Watch the full video here: https://lnkd.in/gDgVMcSj #NobelPrize #JohnHopfield #GeoffreyHinton #DeepLearning #ArtificialIntelligence #Physics #Innovation

Phenomenal work by Chad Bustard, PhD to apply neural networks and learn from simulations of cosmic ray transport! Not only is there a paper, now accepted to MLST(see arXiv link: https://lnkd.in/eisbvEUS), but also Chad has written a wonderful blog post on this! https://lnkd.in/e4aH-3aa Please go check it out! #machinelearning #astrophysics #astronomy #datascience #convolutionalneuralnetworks #neuralnetworks

Physics meets AI: The 2024 #NobelPrize revolutionizes machine learning! John J. Hopfield and Geoffrey E. Hinton have been awarded for their 𝐠𝐫𝐨𝐮𝐧𝐝𝐛𝐫𝐞𝐚𝐤𝐢𝐧𝐠 𝐜𝐨𝐧𝐭𝐫𝐢𝐛𝐮𝐭𝐢𝐨𝐧𝐬 to artificial neural networks. Hopfield's ingenious model showed how neural structures store and reconstruct information, while Hinton's breakthrough enabled systems to 𝐢𝐧𝐝𝐞𝐩𝐞𝐧𝐝𝐞𝐧𝐭𝐥𝐲 𝐝𝐢𝐬𝐜𝐨𝐯𝐞𝐫 𝐩𝐚𝐭𝐭𝐞𝐫𝐧𝐬 in data. Their work bridged physics and AI, laying the foundation for today's powerful machine learning technologies. A testament to the 𝐩𝐨𝐰𝐞𝐫 𝐨𝐟 𝐢𝐧𝐭𝐞𝐫𝐝𝐢𝐬𝐜𝐢𝐩𝐥𝐢𝐧𝐚𝐫𝐲 𝐫𝐞𝐬𝐞𝐚𝐫𝐜𝐡! #NobelPrizePhysics #ArtificialIntelligence #MachineLearning #NeuralNetworks #ScienceBreakthrough #AI #Physics #Innovation #STEM #TechInnovation**