REXASI-PRO’s Post

The 2024 Nobel Prize in Physics was awarded to John J. Hopfield and Geoffrey E. Hinton for their pioneering work in artificial neural networks and machine learning, achievements deeply grounded in principles of physics. Both laureates applied concepts from statistical mechanics and theoretical physics to revolutionize how neural networks process and analyze complex data. John Hopfield is renowned for developing the Hopfield network, a type of recurrent neural network that models the collective behavior of neurons. Drawing from the physics of spin systems, Hopfield's network can store and reconstruct patterns by minimizing the system's energy—a method analogous to physical systems seeking equilibrium. His work demonstrated how physical principles could inform computational models. Geoffrey Hinton, often referred to as one of the "Godfathers of AI," expanded upon Hopfield's ideas by introducing the Boltzmann machine, which employs statistical physics to identify patterns and properties in data. Hinton's contributions also include the development of backpropagation, a critical technique for training deep neural networks, which remains foundational in modern artificial intelligence. Their groundbreaking innovations not only transformed AI but also underscored the interdisciplinary nature of physics, showing how physical methods like energy minimization and probabilistic models can solve complex computational problems. This recognition highlights the essential role physics plays in shaping transformative technologies such as AI, with far-reaching applications in both material science and computational theory. https://lnkd.in/dEqkA24T

🎉 Congratulations to Yamin Sepehri from the Signal Processing Laboratory 4 on the successful defense of his PhD thesis “Smart Edge for Hierarchical Vision Systems” under the supervision of Prof. Pascal Frossard and Prof. Andrea Dunbar. 🔎 Yamin's thesis focuses on the challenges of deploying deep neural network models on camera-equipped edge devices. He proposes innovative hierarchical edge-cloud solutions that address privacy concerns, reduce communication costs, and optimize runtime, all while placing minimal strain on these resource-limited devices. His research introduces novel smart edge processes to effectively tackle these issues in both the inference and training phases of deep neural network execution.

Wow! The University of Tokyo's recent research is nothing short of mind-blowing. They have created "neural organoids" from human stem cells that resemble our own brain's development. Could this lead to a future where machines not only calculate, but actually think and learn, mirroring the human intellect? This leap forward in lab-grown brain mimicking tissue *may* dramatically alter the AI landscape. Some brief insight on today's approach with AI relative to our biological approach to learning - Deep artificial neural networks (ANNs) are, at their essence, intricate mathematical frameworks. They’re built upon a complex architecture of connections and weighted interactions that go far beyond the scope of simple back-propagation algorithms. While these algorithms are key in refining ANNs by adjusting weights to reduce errors, the true nature of ANNs is found in their multi-layered networks of ‘neurons,’ each fine-tuned through a variety of activation functions and data propagation methods. With advanced techniques like dropout and batch normalization, ANNs are not just learning models; they are vast mathematical algorithms that capture and interpret complex data patterns. Despite drawing inspiration from biological processes, deep ANNs ultimately stand as a testament to the power of mathematics to model and perhaps one day, replicate the depth and adaptability of human cognition. But make no mistake, they remain firmly rooted in the realm of complex mathematics, not biology. #NeuralOrganoids #ai #aibrains Link: https://lnkd.in/gKcRS45B

📢 We are happy to share our latest connectomics work published in iScience. We explore the effect of the higher order structure on neuronal function in our biologically detailed model as well as the MICrONS EM data set from Allen Institute. Simplified models of neural networks revealed the need of a tradeoff in brain coding: more redundancy in neuronal activity enhances fault tolerance but reduces memory capacity. The severity of the tradeoff is mediated by the level of neuronal variability. The diverse architecture of biological neural networks helps regulate the tradeoff, letting subpopulations optimize for different goals. We developed a metric, based on simplicial complexes, that captures connectivity complexity allowing us to distinguish between subpopulations.   Subpopulations with low simplicial complexity drive efficient activity, while those with high complexity support network reliability, easing the robustness-efficiency tradeoff. This approach may help reconcile seemingly paradoxical findings that assume uniform connectivity. Full paper: https://lnkd.in/evwCSUpt With Daniela Egas Santander, Christoph Pokorny, András Ecker, Jānis Lazovskis, Matteo Santoro, Jason P. Smith, Kathryn Hess, Ran Levi, Michael Reimann

This year, the Nobel Prize in Physics went to Geoffrey Hinton and Princeton University’s John Hopfield for their work on the foundations of modern artificial intelligence. The Nobel committee noted that machine learning is now “revolutionizing science, engineering and daily life.” 🏆 Several pieces of research were cited as evidence of this revolution. One is a 2022 paper by Princeton researchers that revealed deep new insights into the formation of ice. 🧊 The research — conducted by Athanassios Panagiotopoulos, Pablo Debenedetti, Roberto Car, Pablo Piaggi and Jack Weis — details a machine learning model of water that simulates ice formation. This is something previously considered impossible due to sheer complexity and the thousands of atoms involved. 💡 The model derived its insights from the atoms’ electronic structures, a first-principles approach that would have been unthinkable without the computational efficiencies the researchers leveraged with artificial neural networks. https://lnkd.in/eaPERhU4

In a stunning turn of events, the 2024 The Nobel Prize have celebrated the triumph of AI in science! 🏆✨ Demis Hassabis and John Jumper from Google DeepMind were awarded the Nobel Prize in Chemistry for their revolutionary work on AlphaFold, an AI system that predicts protein structures with remarkable accuracy. This breakthrough has transformed our understanding of proteins, which are essential to life. 🧬Alongside them, David Baker was recognised for his innovative contributions to computational protein design. Together, these scientists have opened new avenues for drug discovery and disease understanding, potentially leading to better vaccines and treatments. 💊🔬In physics, the Nobel Prize went to Geoffrey Hinton and John Hopfield for their pioneering work in artificial neural networks. This recognition highlights how AI continues to reshape our world, driving discoveries that were once thought impossible. 🌍💡These awards mark a pivotal moment in science, showcasing AI's incredible potential to accelerate research and solve complex challenges facing humanity today! 🚀

Revolutionizing Computation: The 'Muscle' Molecules That Think and Act 🧠💪 Looks like we're flexing without thinking! 💪🤯 Dual Functionality 💡: Discover how 'muscle' molecules go beyond mere execution to perform complex calculations, rivaling neural networks in decision-making processes. Physics-Powered Decisions ⚛️: Unveil the study revealing that phase transitions in molecular structures can sense and respond to environmental cues, similar to the brain's function. Robust and Scalable 🏗️: Learn about DNA nanotechnology experiments demonstrating robust, scalable decision-making in multi-component systems. Implications for Evolution 🌱: Explore the potential for evolution to leverage these 'thinking' molecules for efficient, energy-saving cellular functions. This groundbreaking research opens new avenues for computational methods, where physical principles drive both recognition and response mechanisms within cells. 🧬 It's not just a step forward for biology but a leap for technology, offering a fresh perspective on how we might design future computational systems. 🚀 #InnovativeComputing #CellularCognition #MuscleMolecules #NeuralNetworks #ComputationalBiology #DNAtechnology #Biophysics #TechTrends #FutureOfTech #ScienceBreakthrough #BiologyMeetsTech #Nanotechnology #EvolutionaryBiology #ArtificialIntelligence #MachineLearning #DataScience #TechInnovation #SustainableTech #EnergyEfficiency #ResearchAndDevelopment #ScienceCommunication #EmergingTechnologies #ScientificDiscovery #STEM #NextGenComputing

#PhD #position at CRISMAT https://lnkd.in/eziqDEzF Development of Novel #Synthesis and #AI Optimization Methods for the Research of New #Magnetocaloric #Materials

The 2024 @NobelPrize in physics and chemistry have given us a glimpse of the future of science. The #Conversation writes that #AI was central to the discoveries in both categories. Many more #Nobel medals will likely to be awarded to researchers making use of AI tools. We may find scientific methods honored by the Nobel committee depart from straightforward categories such as physics, chemistry and physiology. We may also see scientific backgrounds of recipients retain a looser connection with these categories. While @HopfieldJohn is a physicist and #Baker is a biochemist, @GeoffreyHinton studied psychology before gravitating to AI, and @DemisHassabis and #Jumper are both computer scientists #investors #investing #startup #venturecapital #viaduct #viaductventures #tech https://lnkd.in/d9TD8-Kg

I'm excited to share that my first research paper has been published! 🎉 Title: "Modeling Of Body Flow In A Two-dimensional Channel Based On Physics-Informed Neural Networks" In this work, we investigated how different parameters affect PINN performance in solving computational fluid dynamics problems. We specifically focused on: - Impact of activation functions - Training configurations - Adaptive regularization techniques - Adaptive grid sampling strategies While this is just a first step in my research journey, I'm grateful for the opportunity to contribute to the field of scientific machine learning. Published in: Journal of Applied Mechanics and Technical Physics DOI: https://lnkd.in/e-kYutRt #MachineLearning #PINN #ScientificML #CFD #Research #PhysicsInformedML