Caltech AI Algorithm Speeds Up Image Analysis for Scientists

Excited to share our new preprint: PhyNiKCE: A Neurosymbolic Agentic Framework for Autonomous Computational Fluid Dynamics. LLM agents can generate CFD setups that look right syntactically — but still fail physically. In this work, we propose PhyNiKCE, a neurosymbolic framework that separates neural planning from symbolic validation, so autonomous CFD agents can generate setups that are not just executable, but also physically and numerically consistent. On practical OpenFOAM cases, PhyNiKCE achieved: 1. +96% relative improvement over the baseline, 2. 59% fewer self-correction loops, and 3. 17% lower token consumption. Preprint: https://lnkd.in/gPyXnUYn The poster is generated by the GPT image 2 model. #AI #CFD #LLM #AgenticAI #NeuroSymbolicAI #OpenFOAM #RAG #TrustworthyAI #GPTimage2

💥💥💥 Fluid thinking about collective intelligence Justin Werfel Abstract Forms of collective intelligence range from natural and artificial neural networks to swarm robotics and social insect colonies. One key axis for comparing such systems is the mobility of their individual units: systems like neural networks and wireless sensor networks typically rely on fixed topology and consistent neighbour relationships, whereas mobile robots or ants may encounter each other once and never meet again. Consequently, the core mechanisms that these systems use to compute and learn differ fundamentally between static and fluid topologies. This divide has limited the exchange of ideas across domains. This Perspective examines how mobile units achieve collective learning—through plasticity within individuals, transient formations and, notably, environmental modifications—and identifies analogous mechanisms in static networks. It then explores the advantages of mobility, showing how, for certain tasks, unit mobility can allow a collective system to achieve a given level of performance using many fewer units. An analogy between robot swarms performing a consensus task and convolutional neural networks classifying images illustrates how this principle can inform the design and use of smaller static networks, yielding resource savings. Conversely, temporary immobility or predictable movement patterns can enable mobile unit networks to perform more complex computations by leveraging the benefits of static topologies. Viewing each topology through the lens of the other may inspire advances in both domains, including novel network architectures and swarm algorithms. 👉 https://lnkd.in/gdtS5Vrh #machinelearning

Did you miss the Q&A?🧠Taking Prospective Memory from Theory to Code: Last wednesday, we dropped our latest lecture on Prospective Memory—the biological blueprint for how systems "remember the future." But how does a neural mechanism for "remembering to buy milk" translate into a more efficient AI data pipeline? Re-watch the live Q&A where we took a deep dive into the technical architecture of the ENCODER and how we are using biomimetic processes to solve the most persistent bottlenecks in autonomous agency. Don't just take our word for it—test the architecture yourself. 👉 Join the Beta: serva.servamind.com #AI #Neuroscience #Servamind #CognitiveArchitecture #LiveQA

Multitec is proud to share our latest research contribution to the ML community. SMT V10 — an adaptive sparse optimizer that autonomously suppresses 93.79% of gradient updates during training while maintaining competitive accuracy, validated across multiple random seeds. The algorithm evaluates gradient distributions layer-by-layer, masking thermal noise and preserving only structurally significant updates. Designed with Edge AI and neuromorphic hardware in mind. 📄 Full paper: https://lnkd.in/gVZ8ZytQ Multitec is an active member of NVIDIA Inception, developing next-generation efficient AI algorithms. Andrei Castilho Golfeto #MachineLearning #EdgeAI #SparseLearning #NvidiaInception #EfficientAI

Emory University Physicist have taken a major step toward using AI not just to analyze data, but to uncover entirely new laws of nature. By combining a specially designed neural network with precise 3D tracking of particles in a dusty plasma—a strange “fourth state of matter” found from space to wildfires—the team revealed hidden patterns in how particles interact. Their model captured complex, one-way (non-reciprocal) forces with over 99% accuracy and even overturned long-held assumptions about how these forces behave. Source: Science Daily https://lnkd.in/gT4YgmmA

⚡ kurate: a cool new website that evaluates the scientific impact of papers hosted on arXiv. ➡️ For now, it's only available for AI, ML, robotics & quantum physics 🤞 But hopefully it'll soon include the economics & quantitative finance fields... Example for ML: https://lnkd.in/erTNUc3z

Happy to share another piece of good news within 24 hours! Our manuscript titled “Efficient Coreset Generation for Chest X-ray Imaging Using Compressed Sensing” has been accepted at the IEEE International Conference on Image Processing (ICIP) 2026, to be held in Finland from September 13–17, 2026. In this work, we propose a novel framework for generating core chest X-ray datasets grounded in compressed sensing principles. I would like to thank and congratulate Dr. Pradip Sasmal and Pradyumna Pradhan once again for their valuable efforts and contributions. #icip #AI #imageprocessing #coreset

Artificial Intelligence is advancing data-driven approaches to sensor fusion and state estimation, but its role in Positioning, Navigation and Timing (PNT) remains unresolved. Our upcoming Location SIG event, delivered in partnership with the Royal Institute of Navigation, will explore Advances in AI - including deep learning, probabilistic modelling and emerging foundation models - are being applied to core PNT challenges such as signal processing, multi-sensor fusion, and navigation in GNSS-denied or degraded environments. We’re also opening a small number of speaker slots for organisations working in the PNT & AI space to showcase products, technologies and solutions. 30th June 2026 1pm – 5pm The Bradfield Centre, Cambridge The event is free to members of Cambridge Wireless and the Royal Institute of Navigation. Interested in speaking? Please contact Clare Kettle to register your interest. Register here: https://lnkd.in/dCc44Ywx #CWLocation #GeospatialTech #NavigationInnovation #CambridgeWireless 

🎉 📣 📣 : Excited to share our most recent publication on IEEE Transactions on Radar Systems. It presents a low-cost monopulse radar receiver system that is co-designed with a novel planar, symmetric RF comparator network circuitry and a compact deep neural network (DNN) method, aiming to enhance the direction-of-arrival (DoA) estimation on remote object detection (https://lnkd.in/e6UgPEiS).

I am glad to announce my latest blog post, "A Physics-informed Coupled Oscillator Framework for Synthetic ECG Generation," [https://lnkd.in/gkCUzjKS] which marks a modest beginning in our efforts to ground AI in the biological reality of the heart. By modeling cardiac nodes as nonlinear oscillators and using an RK4 numerical engine, we have developed "idealist" waveforms that serve as a rigorous ground truth for clinical validation. This framework is now being integrated into the upcoming version of the Lepakshi-ECG Annotator, enabling a more interpretable and physically grounded approach to cardiac signal analysis. [https://lnkd.in/gkCUzjKS] #AI #MachineLearning #SignalProcessing #CardiacDynamics #HealthTech #ECG #NeuroKit2 #LepakshiAnnotator #PhysicsInformedAI #IntuitusResearch