AIMNet2 – A Robust Neural Network Potential | Roman Zubatyuk | AMS Webinars 2024 скачать в хорошем качестве

AIMNet2 – A Robust Neural Network Potential | Roman Zubatyuk | AMS Webinars 2024

Timestamps 00:50 Supervised Machine Learning on Quantum Chemica Data 01:51 Introduction to ANI Potentials 05:17 AIMNet2 Architecture & Perfomance 09:38 Outline of the Talk 10:37 Trainingset for AIMNet2 12:53 AIMNet2 Benchmarks 15:21 AIMNet2 Application to unusual bonding (hypervalent, element organic molecules, ...) 16:49 AIMNet2 Conformational Search Benchmark 19:00 AIMNet2-NSE - Extension to Spin-Unrestricted Case 20:00 AIMNet2-NSE - Radical Reactions Benchmark 21:26 AIMNet2-NSE - Applications to Decomposition of Silylated Benzopinacol and Ring Opening Polymerization 23:24 The Need for Targeting Reactions in ML Potentials 24:37 AIMNet2-RXN - Broad transferability across Reaction Mechanisms for CHNO Chemistry 25:07 AIMNet2-RXN - Rapid Reaction Characterization 26:53 Pd-catalyzed C-C Cross Coupling Reactions (Suzuki) 28:19 AIMNet2-Pd - Benchmarking Energies & Reaction Profile 30:31 AIMNet2-Pd - Beyond Phosphine Ligands 31:06 Crystal structure refinement of Protein Structures 34:01 AIMNet2-QR - Geometry Optimization of High-Quality Protein Structures 36:17 Conclusion and Outlook 38:08 𝗗𝗘𝗠𝗢: 𝗔𝗜𝗠𝗡𝗲𝘁𝟮 𝗶𝗻 𝗔𝗠𝗦 𝗯𝘆 𝗠𝗮𝘁𝘁𝗶 𝗛𝗲𝗹𝗹𝘀𝘁𝗿𝗼𝗺 (𝗦𝗖𝗠) Free and fully functional AMS trial: https://www.scm.com/free-trial/?utm_s... Abstract: Roman Zubatiuk from the group of Olexandr Isayev introduces 𝗔𝗜𝗠𝗡𝗲𝘁𝟮, an innovative neural network potential that combines the power of machine learning with physics-based long-range electrostatic and dispersion terms. Designed for complex organic and element-organic molecules, AIMNet2 provides a scalable, high-throughput alternative to traditional quantum mechanical methods with minimal loss in accuracy.