Wand AI’s Post

What do active, communicative extraterrestrial civilisations and smart gift recommendations have in common? Why, probabilistic equations of course. Astrophysicist Frank Drake drew up the famous Drake Equation on a chalkboard in 1961. This was at the dawn of a worldwide search for extraterrestrial intelligence (SETI), and his thinking continues to influence the use of astronomical observatories to this day. The equation is a set of seven variables which when multiplied together yield a calculation of the possibility that humanity might someday hear from an intelligent civilisation. Similarly, at Givving we have developed an equation for computing a Gift Relevance Score (GRS), quantifying the suitability of a gift for a specific person, based on set of eight variables. We have just published our first technical white paper – An Intelligent Framework for Gift Recommendations: Leveraging AI and the Drake Equation – outlining how we have taken inspiration from this famous argument to build our universal gift commerce engine. Read the whitepaper here: https://lnkd.in/gvBDbqbS #Givving #AI #Startups

📈 Deep Stochastic Optimal Control for Quadratic Hedging with Model Uncertainty Our latest research on deep stochastic optimal control is out. We focus on the popular challenge of model uncertainty in quadratic hedging of a European Call option with transaction costs under a discrete trading schedule. 🔍 Key findings: - Parameterizing hedge ratio policies at each time step using an independent neural network aligns better with the dynamics of the gradients in ADAM optimization. - This approach enhances training and improves performance. 📄 Read the full paper here: https://lnkd.in/ewGxFjiG Thanks to Ali F. and Ahmad Aghapour #QuantFinance #DeepHedging #AI #NeuralNetworks #OptimalControl #FinanceResearch #MachineLearning

Happy to share that our paper was accepted to NeurIPS 2024! Paper title: “Fair Bilevel Neural Network (FairBiNN): On Balancing Fairness and Accuracy via Stackelberg Equilibrium” Preprint will be out soon! #NeurIPS2024 #machinelearning #AI

Steve HovlandSteve Hovland (He/Him) • 1st(He/Him) • 1st AI StrategyAI Strategy 7h • 7h • Steve HovlandSteve Hovland (He/Him) • 1st(He/Him) • 1st AI Strategy The Global Brain By Howard Bloom In this extraordinary follow-up to the critically acclaimed The Lucifer Principle, Howard Bloom - one of today's preeminent thinkers - offers us a bold rewrite of the evolutionary saga. He shows how plants and animals (including humans) have evolved together as components of a worldwide learning machine. He describes the network of life on Earth as one that is, in fact, a "complex adaptive system," a global brain in which each of us plays a sometimes conscious, sometimes unknowing role. And he reveals that the World Wide Web is just the latest step in the development of this brain. These are theories as important as they are radical. https://lnkd.in/grR-vuAv

A recent paper in Transactions on Machine Learning Research introduces Neural Operator Flows (OpFlow), a breakthrough in functional regression. Unlike traditional models that rely on Gaussian processes, OpFlow brings universal functional regression (UFR) to life by learning from complex, non-Gaussian data—a game-changer for fields like seismology and environmental science. Why OpFlow matters: 1. High Precision: It maps data into Gaussian spaces, allowing exact likelihood estimations and unprecedented accuracy in quantifying accuracy 2. True Bayesian Predictions: OpFlow provides reliable predictions even with sparse or noisy data. 3. Real-world Ready: It adapts easily to complex, real-world datasets where traditional models fall short. This approach opens up new possibilities for industries needing accurate functional predictions and insights. #MachineLearning #AI #DataScience #Innovation

🎉 Thrilled to share our work "Constrained Exploration via Reflected Replica Exchange Stochastic Gradient Langevin Dynamics", co-authored with Hengrong Du, Qi Feng, Wei Deng, and Prof. Guang Lin, was accepted at [ICML] Int'l Conference on Machine Learning. 🔬 Summary: Replica exchange Langevin Monte Carlo (reSGLD), which simulates Langevin dynamics with multiple chains in parallel, is an effective sampler for non-convex learning across large-scale datasets. However, the simulation may encounter stagnation issues when the high-temperature chain explores the distribution tails too deeply. To tackle this issue, we propose reflected reSGLD (r2SGLD): an algorithm tailored for constrained non-convex exploration by utilizing reflection steps within a bounded domain. With r2SGLD, we leverage the advantages of the high-temperature chain while mitigating the risks of over-exploration. 💻 Paper: https://lnkd.in/gnJKQ2Ws 💻 Code: https://lnkd.in/gUDBZDA6 Our poster will be presented on July 23, Poster Session 1. Feel free to come by and discuss the work with us! #ICML #MachineLearning #AI #Research #statistics

Did you catch the Optimaviz Live Demo invite? 🚀 We've just sent out exclusive invites to our upcoming webinar on November 1st! Check your inbox for a front-row seat to the future of mineral processing. 🔍 Can't find it? No worries! Maybe it got lost in the digital jungle. Just drop a comment below or DM us with your email and we'll teleport you a fresh invite link faster than you can say "process optimization." ✨ New to Optimaviz? ✨ You're in for a treat! This cutting-edge platform uses AI and machine learning to supercharge your operations, boost efficiency, and maximize profits. Don't miss this chance to see Optimaviz in action! Register now and join us for a deep dive into the future of mineral processing. New Interest👉Register: https://lnkd.in/dPy-wgG5 #Optimaviz #MineralProcessing #ProcessOptimization #IDAO #Webinar #AI #MachineLearning #Innovation  

In optimization, Newton's method is renowned for its quadratic convergence, largely due to its optimal step size determined by the Hessian matrix. Unlike gradient descent, which simplifies the step size to a single coefficient, alpha, through line search, Newton's method utilizes the full Hessian matrix. This difference is crucial. In Newton's method, the step size is a matrix that captures the curvature of the objective function, enabling faster and more accurate convergence. However, when we use gradient descent, we collapse this matrix into a scalar coefficient, which degrades performance and slows down convergence. This degeneration from a matrix to a scalar is what costs us efficiency. While calculating the Hessian can be computationally intensive, leading to the use of quasi-Newton methods like BFGS, understanding the fundamental advantage of Newton's method in maintaining a matrix step size can guide better optimization practices. How do you approach the trade-off between computational feasibility and convergence speed in your optimization tasks? Let's discuss the strategies and experiences that have worked best for you. PS if the content brought you value please share with someone who can benefit from it. If not, please tell me how can I improve my content. #Optimization #NewtonMethod #MachineLearning #GradientDescent #DataScience #AI #Algorithm #Performance #Convergence #BFGS

AI research… what’s coming next? At DAVANTIS, we've recently finished up our latest research project focused on pioneering attention-based deep learning architectures for security systems. The term "artificial intelligence" (AI) was introduced in 1956 by computer scientist John McCarthy during a conference at Dartmouth College. McCarthy and his peers employed this term to delineate the field of study and exploration aimed at crafting machines capable of exhibiting intelligent behavior akin to humans'. The initial AI systems for perimeter protection relied on motion detection, heuristics, and rule-based approaches. The subsequent generation saw the integration of machine learning techniques. Presently, we find ourselves amidst the deep learning epoch, wherein deep learning technologies are rapidly evolving. At DAVANTIS, we persistently strive to push the boundaries of AI. This project has been made possible through the financial support of CDTI Innovación - Centro para el Desarrollo Tecnológico y la Innovación and FEDER funds.   #FEDER #FondosFEDER #UnaManeraDeHacerEuropa #AyudasCDTI

Hello, LinkedIn network, The following short clip shows the interesting behavior of three famous metaheuristics: Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Ant Colony Optimization (ACO), for a continuous optimization problem (finding the minimum point in a very complex, bumpy surface). The clip demonstrates how these algorithms converge to (or become crowded around) the hollow region of the surface after sufficiently probing the solution space. The global optimization power of the population-based metaheuristics makes them excellent candidates for finding promising areas of the solution space. Integrating them with a local search algorithm at a second stage would result in even more efficient optimization toolboxes. Please follow and like if you are interested in similar posts about optimization, combinatorial optimization, algorithm design, metaheuristics, Machine Learning (ML) in optimization, and Artificial Intelligence (AI). #MATLAB #ArtificialIntelligence #Metaheuristics #EvolutionaryAlgorithms #ContinuousOptimization #GeneticAlgorithm #ParticleSwarmOptimization #AntColonyOptimization #GA #PSO #ACO