Noah Ratzan’s Post

SciAgents are drastically important concnept in automation of various science research as it utilize the selection of appropriate math and science theory tools.

Can #AI not only support but actually drive the future of scientific discovery? We are excited to introduce SciAgents, an agentic AI aimed towards scientific discovery through the integration of large-scale knowledge graphs, LLMs, and adversarial interactions between multiple experts. The model is capable of autonomously advancing scientific understanding💡by exploring novel domains, identifying complex patterns, and uncovering previously unseen connections in vast scientific data, while retrieving new data via literature search 📜. Using graph reasoning, SciAgents identifies interdisciplinary relationships that might otherwise remain hidden, offering a step-by-step strategy for discovery & innovation. The video features an audio track generated using 🍓 #o1 based on the original paper and design examples, providing an explanation of the work and its implications. Key elements include: ➡Ontological Knowledge Graphs: Structuring and connecting scientific concepts to highlight relationships across fields. ➡Multi-Agent Collaboration: AI agents autonomously generate and refine hypotheses, critique research, and evaluate emerging trends. ➡Graph-Based Reasoning: Identifying novel material designs, such as mycelium-based composites or silk-pigment blends, informed by both natural and artificial patterns. SciAgents can be used as an autonomous or collaborative tool to assist human researchers. The system offers a more powerful way to process vast data, providing innovative paths to explore nature-inspired designs or unexpected material properties. 💡In the field of materials science, for instance, SciAgents has already demonstrated how principles from biology, music, and art can converge to create new biomimetic materials. Through isomorphic mapping, parallels have been drawn between Beethoven’s 9th Symphony and biological structures, pointing to a broader applicability of AI-driven insights across disciplines. This project allows us to enhance capabilities of researchers, allowing them to explore larger datasets and propose hypotheses grounded in a vast, interconnected web of knowledge. The agentic system was built using #AutoGen. #AI #ScientificResearch #GraphReasoning #AI4Science #MaterialsScience #InterdisciplinaryResearch #SciAgents #OpenAI Paper: Alireza Ghafarollahi and Markus J. Buehler, SciAgents: Automating scientific discovery through multi-agent intelligent graph reasoning, https://lnkd.in/e-88pQ6c, 2024. Data and weights 🤗: https://lnkd.in/eku6j7JV Code: https://lnkd.in/eJVZD8BE

This is great usage of AI for scientific purpose.

I am pitching this already for so long to investors in Europe and they just don’t get it… Maybe now they understand a bit better the value of ExtensityAI and #SymbolicAI #NeurosymbolicAI #SymbolicReasoning #AgenticFrameworks

🌟 Exciting Announcement for IJCNN 2025! (Call for Papers)🌟 We are thrilled to introduce our Special Session: "Artificial Intelligence and Machine Learning Innovations (AIMLI): Showcasing Funded European Projects", taking place at the IEEE International Joint Conference on Neural Networks (IJCNN 2025) in Rome, Italy (June 30 – July 5, 2025). This session will spotlight groundbreaking AI and ML advancements addressing critical societal, economic, and environmental challenges. From sustainable mobility and smart agriculture to healthcare technologies and Industry 4.0, we aim to showcase the transformative power of European-funded initiatives. 📢 Why Participate? ✅ Share your innovative methodologies, case studies, and project outcomes. ✅ Exchange ideas with leading researchers and practitioners. ✅ Highlight the societal and economic impacts of your AI/ML solutions. 📅 Important Dates Paper Submission Deadline: January 15, 2025 Authors Notification: March 31, 2025 Camera Ready & Registration: May 1, 2025 Visit our website for all the details and submission guidelines: 👉 https://lnkd.in/d6pzDYrW Let’s shape the future of AI together! 🚀 Feel free to share this post and tag colleagues who might be interested in contributing. #IJCNN2025 #AI #MachineLearning #Innovation #Research #Sustainability #DigitalTransformation #Rome2025 #CIPARLABS #EU

🚀 Revolutionizing Scientific Discovery with AI Agents by Massachusetts Institute of Technology 🤖 🔬 At the intersection of AI and materials science, SciAgents is leading a new era of innovation by automating scientific discovery through multi-agent systems and knowledge graphs. Imagine a future where groundbreaking discoveries are powered by AI systems, accelerating the research process and uncovering hidden connections that were previously invisible! Here are the key highlights of SciAgents: 🔹 Multi-Agent Intelligence: A team of AI agents collaborates dynamically, each specializing in tasks such as hypothesis generation, refinement, and critique. 💡 🔹 Ontological Knowledge Graphs: These vast networks of interconnected scientific concepts allow AI to generate new insights and patterns that human researchers might miss. 📊 🔹 Breakthroughs in Bio-Inspired Materials: By analyzing data from over 1,000 scientific papers, SciAgents has unlocked previously unseen interdisciplinary relationships that accelerate material design. 🌱 🔹 Autonomous Hypothesis Generation: SciAgents autonomously generates, critiques, and improves research hypotheses, integrating the latest scientific data in real time. 📚 The future of scientific discovery is here – with AI at the helm, the possibilities are endless. Let’s embrace this new frontier together! 🚀 #AI #ScientificDiscovery #Innovation #Research #AIinScience #BioInspired #KnowledgeGraphs #MachineLearning

"Graph-based AI model maps the future of innovation" by Stephanie Martinovich Department of Civil and Environmental Engineering at Massachusetts Institute of Technology What if we told you that the worlds of biological tissue and Beethoven’s “Symphony No. 9” are interconnected in ways we've yet to fully understand? 🎶🔬 Thanks to groundbreaking work by Markus J. Buehler at MIT, a novel AI approach is bridging this gap, uncovering shared patterns of complexity using generative AI and graph-based computational tools. This advanced AI model analyzes vast scientific literature to create knowledge maps, revealing unexpected relationships between diverse concepts. From predicting new material designs to exploring intersections of art and science, the potential applications are staggering! 🌍✨ Imagine developing biodegradable materials inspired by Kandinsky’s art, combining functionality with innovation. As we harness AI to make sense of complex data, we open the door to unprecedented discoveries across multiple fields. Are we on the brink of a new era where creativity and science synergize like never before? #AI #MachineLearning #Innovation #BioInspiredMaterials #Interdisciplinary #GenerativeAI #DataScience #ScientificDiscovery Source: https://lnkd.in/dvGMJiAP

I spent a lot of time during my doctorate (and post-doc) doing benchmarks for electronic structure theory methods. Benchmarking, be it for density functional theory calculations or for machine learning algorithms, is essential and informative. I am proud to have helped in developing 'Benchmarks' and other article types ('Dataset articles', 'Challenges') for our #MachineLearning journals (https://lnkd.in/dNq6nGbX). I hope that dedicated article types will spread the word further, with more and more folks reading these important contributions. Ultimately, I hope that the 'jump' towards the real-world applications that Sergei Kalinin is talking about will be a little easier #EditorsLife IOP Publishing

🔍 Benchmarks in ML for Science: A Starting Point, Not the Destination 🔍 In the world of machine learning, benchmarks serve as critical reference points. They help define performance standards and drive comparability, ensuring our models meet necessary thresholds. But while benchmarks are indispensable, are they the end-all? Not quite. I would argue that the true value of any development lies not in its standalone metrics but in its ability to drive downstream applications and, indirectly, influence upstream progress. Without understanding these connections, the direction for ML (or any other) development can become lost. Drive for better internal benchmarks in ML (with suitable protection against data leakage) or high spatial/energy resolution in instrumental communities is always good - but its value is often difficult assess. For instance, when we developed deep kernel learning workflows to study structure-property relationships in microscopy, we achieved an impressive 30-100x acceleration in measurement speed benchmarked via predictive uncertainty by shifting from grid-based measurements to automated experiment. But simply speeding up measurements isn’t enough—the question is, how does this acceleration affect materials discovery downstream? For example, if we realized ML-enabled microscopy workflow that has access to multiple measurement modalities (see movie by Utkarsh Pratiush, Austin Houston, and Gerd Duscher), based on which principles do we even orchestrate it? Answering this is complex and highly problem-specific. We might analyze it for specific applications, such as understanding the physics of ferroelectric materials or tracking battery degradation, but each of these is distinct. The hope is that by solving enough of these specific cases, we’ll eventually gain insight into the broader impact: how much we’re accelerating the discovery process as a whole. This approach is undoubtedly more intricate, but it aligns with the reality that we work in an open world. Our advancements are meaningful only if they contribute to productivity and impact across the entire scientific workflow. And to figure it out, we need to adopt the system-based approach for scientific discovery and ML4Science, rather then problem specific. Benchmarks are necessary, yes, but the ultimate goal is to foster real-world progress that scales beyond the lab - and for that we have to connect our benchmarks to others' goals. #MachineLearning #Benchmarks #MaterialsScience #Microscopy #AI #DiscoveryScience #ScientificProductivity

🌟 Excited to share! 🌟 My research has been published in IEEE, and I’m thrilled to present our work: Title: “ANSAN: Data Interpretability-Based Analysis of Deep Learning Models” Our study leverages lightweight MobileNet architecture and integrates explainable AI techniques like LIME for glaucoma detection. By focusing on dataset variability and mobile deployment, we demonstrate a scalable and interpretable solution for early glaucoma screening, especially for resource-constrained settings. 🔗 Read the full paper here: https://lnkd.in/dXexXAQn I extend my heartfelt gratitude to my co-author Karthie Krishna K and our mentors,senthilkumar thangavel Sir and Bindu Avadhani Ma’am, for their invaluable guidance and support throughout this journey. This research underscores the potential of AI in revolutionizing healthcare diagnostics while emphasizing the importance of explainability and trust in AI systems. #AI #DeepLearning #ExplainableAI #HealthcareInnovation #GlaucomaDetection #Research #IEEE