"Membrane Pseudo-Bacterial Potential Field with GPU Acceleration for Mobile Robot Path Planning" published in AI and Quantum Computing book.

If you're developing AI-Physics models, you've likely explored GNNs for unstructured mesh data but faced memory and scaling issues with large meshes. PhysicsNeMo engineers have addressed these challenges, successfully using GNNs on meshes with over 40 million cells. They provide a how-to guide for applying these methods: https://lnkd.in/ehPCBEVj

Scalable Graph Networks - on the way to the 1B nodes goal (and using same mesh topology as numerical simulations)!

🚀 E-track Best paper nominee at DATE2025 in Lyon “RankMap: Priority-Aware Multi-DNN Manager for Heterogeneous Embedded Devices” Andreas Karatzas, Dimitrios Stamoulis, Iraklis Anagnostopoulos from Southern Illinois University Carbondale, and The University of Texas at Austin Modern edge systems often run multiple deep neural networks (DNNs) simultaneously, but efficiently managing them across CPUs, GPUs, and other heterogeneous components remains a major challenge. RankMap takes this head-on. RankMap is a priority-aware multi-DNN manager that intelligently distributes DNN workloads across heterogeneous embedded devices. By combining fine-grained DNN partitioning, a multi-task attention-based performance estimator, and Monte Carlo Tree Search (MCTS) for smart mapping, RankMap achieves impressive gains: ⚡ Up to 3.6× higher throughput than state-of-the-art methods 🚫 Zero DNN starvation under heavy workloads 🎯 57.5× improvement in meeting priority constraints RankMap dynamically balances performance and fairness, ensuring that critical applications get the resources they need—without sacrificing system efficiency. This work paves the way for smarter, more reliable AI execution at the edge, where every millisecond and every core matters. #DATE2025 #EdgeAI #EmbeddedSystems #DeepLearning #Research #HPC #AIOptimization Aida Todri-Sanial Theo Theocharides Alberto Bosio Matteo Sonza Reorda Nele Mentens

Extropic TSU, Physics' Answer to AI's Power Crunch ⚡ Core twist: Pbits tap silicon noise for probability sampling, ditching GPU math for ten thousand times efficiency. 🔧 Architecture: Bipartite grids with local links, Gibbs sampling chains for generative flows like images. 🛤️ Roadmap: X0 proves it now, XTR-0 kits shipping, Z1 scales millions of pbits by 2026. ❄️ Dual paths: Superconducting for cryo efficiency, room-temp for everyday slots and edge dreams. ⚖️ Wins and walls: Local comms save energy, but scaling and custom algos test the promise. 🌱 My view: Timely rethink for sustainable compute, watch if it disrupts the GPU throne. 📚 Tools ready: Thrml library for sims, grants for researchers diving in. Unseat the status quo, read the deep dive on this hardware shake-up. [Read the full piece → https://lnkd.in/gRKMCMxZ ] #ExtropicTSU #PbitInnovation #ProbabilisticHardware #AIEfficiency #TechRoadmap #SiliconFuture #GenerativeTech #HardwareShift #SustainabilityInTech #ComputeRevolution

Big announcement from Extropic regarding a new chip design and accompanying AI algorithm that leverages “thermodynamic computing” that can reduce AI energy consumption by up to 10,000x. It sounds fantastical, but given that GenAI is a probabilistic output, it actually makes more sense to run it using a probabilistic computing platform (as opposed to deterministic hardware such as GPU’s). Looking forward to seeing this project move into production! https://lnkd.in/eN3RduVf

The Dawn of ‘Quantum EDA’ Read more here: https://lnkd.in/eGf_5tVy Nanoacademic Technologies and Kothar Computing are partnering to bring EDA-grade design automation to quantum hardware. Together, we’re uniting QTCAD® device-level TCAD with Kothar’s Quantum Symbolic Algebra (QSA) Engine. This will result in the ability to simulate 100 physical qubit processors for spin and superconducting technologies, unlocking a new era of device design where teams can design, test and validate prior to fabrication at interactive speeds. Classical chips scaled from thousands to billions of transistors only after TCAD/EDA software matured. This leap is within sight for quantum computing. This partnership turns quantum hardware from a fragile physics experiment into an engineering discipline: reduce cost, de-risk iteration, and accelerate time to workable, foundry-ready qubit arrays. Check us out and signup to get access to Kothar’s beta launch: kotharcomputing.com/signup And check out our new partner here: nanoacademic.com

Thrilled to announce our partnership with Nanoacademic! With Nanoacademic’s physics modeling and Kothar Computing's many-body solvers, there is a clear path to iterative, 100+ qubit quantum processor designs. Excited to see what this unlocks for researchers and hardware manufacturing teams.

🚀 New on the Bioconductor Blog: GPU Support in Bioconductor 📝 Written by Andres Wokaty Bioconductor is building stronger support for GPU-accelerated package development, enabling faster and more scalable analysis workflows. Learn how package maintainers can take advantage of this new GPU infrastructure: https://lnkd.in/e43wbPYM #Bioconductor #GPUcomputing #Bioinformatics

NVIDIA CUDA-X #libraries form the backbone of quantum research. ⚡From faster decoding of quantum errors to designing larger systems of qubits, researchers are using CUDA-X to push past the limits of classical computation, bringing the power of accelerated computing to #quantumcomputing. Learn more ➡️ https://bit.ly/4mRC4wi

NVIDIA CUDA-X #libraries form the backbone of quantum research. ⚡From faster decoding of quantum errors to designing larger systems of qubits, researchers are using CUDA-X to push past the limits of classical computation, bringing the power of accelerated computing to #quantumcomputing. Learn more ➡️ https://bit.ly/46Dqznq