Sharing MSc projects in control systems and AI on GitHub

🚀 Thrilled to Share Our Latest Research Publication! Our paper, titled “Golden Eagle Optimized Hybrid RNN-GRU Model for Stock Price Prediction: A Data-Driven Deep Learning Approach,” has been officially published in the IEEE Xplore Digital Library as part of the 2025 8th International Conference on Computing and Communications Technologies (ICCCT). This research introduces a hybrid deep learning framework that integrates RNN and GRU architectures optimized with the Golden Eagle algorithm — aimed at enhancing accuracy in stock price prediction amidst market volatility and non-linear behavior. 📊🤖 A heartfelt thanks to my co-authors K. Loganathan, R. Indumathi, P. Mukilan, J. Prisca Mary, and C. Pandi for their valuable collaboration throughout this work. 🙌 📘 DOI: 10.1109/ICCPCT65132.2025.11176673 📚 Read on IEEE Xplore: https://lnkd.in/gX7PwyQn #IEEE #Research #DeepLearning #ArtificialIntelligence #MachineLearning #DataScience #StockPrediction #Publication #AcademicResearch Karpagam Academy of Higher Education

A look back on the #AISchool2025 in Leipzig: a participatory week dedicated to #MachineLearning, #DataVisualization and #HPC workflows! Over five days, participants from across the Leibniz Association MMS network: 🔹 Gained introductions to machine learning techniques, deep learning, and data visualization 🔹 Explored the use of LLMs for programming support and synthetic data generation 🔹 Investigated deeper through three Specialization Topics 📂 Learning materials and resources remain openly available here: https://lnkd.in/gJrfidb8 The winter school was hosted and organized by Leibniz Institute of Surface Engineering (IOM) with expertise, instructors (Robert Haase & Matthias Täschner) and materials provided by ScaDS.AI Dresden/Leipzig.

Teaching machines to see the invisible: In a first-of-its-kind search at a particle collider, the CMS experiment uses physics-informed machine learning to look for partially visible sprays of particles (jets), containing leptons and dark matter 🌌 👉 https://lnkd.in/eP9-RAv8

Thrilled to share that our paper “Enhancing Time-Series Forecasting Stability with a Connected Neural Dynamical System Approach (C2DS-L)” has been accepted in Physica D: Nonlinear Phenomena. Together with Prof. Boumediene Hamzi, we introduce C2DS-L, a hybrid framework that couples empirical mode decomposition with neural differential operators under stability constraints. On synthetic and real datasets, it improves predictive accuracy, long-term stability, and interpretability—bringing data-driven learning closer to dynamical-systems theory. #TimeSeries #DynamicalSystems #NeuralODE Elsevier Download link: https://lnkd.in/gpcyzcpa Code: https://lnkd.in/g9yKMDM4

🚀 New Paper Accepted at NeurIPS 2025! Excited to share our latest work: “LexiCon: A Benchmark for Planning under Temporal Constraints in Natural Language” (with Rishi Hazra, Pedro Zuidberg Dos Martires, and Luc De Raedt) 📄 Paper: https://lnkd.in/eb98Gtys 💻 Code: https://lnkd.in/e8sY6A5V 🔍 𝐇𝐨𝐰 𝐠𝐨𝐨𝐝 𝐚𝐫𝐞 𝐦𝐨𝐝𝐞𝐫𝐧 𝐋𝐋𝐌𝐬 𝐚𝐭 𝐩𝐥𝐚𝐧𝐧𝐢𝐧𝐠 𝐮𝐧𝐝𝐞𝐫 𝐜𝐨𝐧𝐬𝐭𝐫𝐚𝐢𝐧𝐭𝐬?  - LLMs have made natural language-based reasoning increasingly feasible, though many questions remain.   - We explore whether LLMs truly possess the reasoning capabilities required for planning under constraints. ⚙️ 𝐋𝐞𝐱𝐢𝐂𝐨𝐧: A simulator that generates benchmarks for planning under temporal constraints in natural language, allowing systematic evaluation of LLMs. 💡 𝐅𝐢𝐧𝐝𝐢𝐧𝐠𝐬: The planning abilities of state-of-the-art LLMs (GPT-5, o3, R1) deteriorate as the tasks produced by LexiCon become more constrained. 🧩 𝐖𝐡𝐲 𝐋𝐞𝐱𝐢𝐂𝐨𝐧 𝐦𝐚𝐭𝐭𝐞𝐫𝐬: It is 𝐞𝐱𝐭𝐞𝐧𝐬𝐢𝐛𝐥𝐞 with new planning environments and arbitrary constraint complexity, leading to 𝐟𝐮𝐭𝐮𝐫𝐞-𝐩𝐫𝐨𝐨�� 𝐛𝐞𝐧𝐜𝐡𝐦𝐚𝐫𝐤 construction for next generation LLMs. Thrilled that this marks my first paper as a Postdoctoral Researcher at the Machine Perception and Interaction (MPI) Lab, Örebro University 🇸🇪, supported by WASP – Wallenberg AI, Autonomous Systems and Software Program funded by the Knut and Alice Wallenberg Foundation. More to come soon! 🚀 #NeurIPS2025 #WASP #AIresearch #LLMs #Planning #Reasoning #NeuroSymbolicAI #MachineLearning #Research

LexiCon benchmark for planning under constraints. You can easily integrate new domains and the simulator will automatically churn out constrained tasks. Obviously, All LLMs struggle under constraints

Thrilled to see our paper, co-authored with Dr. Mohebbi and published in 2023 in Computers & Operations Research—a leading journal in Operations Research—gaining attention! 👀 Our work on Causal Inference addresses the challenge of aggregating Directed Acyclic Graphs (DAGs), going beyond the mathematics to improve real-world ensemble learning. By preserving all useful information from individual learners, our approach can enhance applications across infrastructure resilience, healthcare, environmental engineering, and beyond. #causalinference #ensemblelearning #operationsresearch #healthcare #infrastructureresilience #environmentalengineering https://lnkd.in/e2GY3BUG

Day 7—Fixed Point Iteration Method-Day 7/15 As I continue my 15-day numerical analysis journey, today’s focus is on a simple but very important root-finding technique—the fixed point iteration method. This method converts the equation f(x)=0 into the form: x = g(x) Then, starting from an initial guess, we generate the next values using the rule: Xn+1 = g(Xn) We keep repeating this process until the difference between two successive values becomes very small this final value is called the fixed point, where x=g(x) Concept in Simple Words Think of it like this: we take a number, apply the function to it, and use the output as the next input. If this process keeps getting closer to one specific value, that’s the fixed point our solution! However, for the method to work (i.e., to converge), an important condition must be satisfied: |g'(x)| < 1 If this condition fails, the process may move away from the root instead of approaching it. Real-Life Applications The fixed-point iteration method is used in many real-world fields: Engineering: solving nonlinear equations in heat transfer and fluid flow. Economics: finding equilibrium points in supply-demand or market models. Computer Science: iterative algorithms in graphics, machine learning, and optimization. It’s also a foundation for understanding how modern numerical algorithms work through repeated refinement. Advantages and Limitations Advantages: Easy to understand and implement. Great for learning how iteration leads to convergence. Limitations: Convergence depends strongly on the choice of. It may be slow or even diverge if. Key Takeaway The Fixed Point Iteration Method beautifully shows how small, consistent steps can lead to accurate results. It’s not just a numerical tool it’s a lesson in patience and precision in mathematics. #NumericalAnalysis #Mathematics #FixedPointIteration #MathStudents #EngineeringMath #LearningJourney #AppliedMathematics #STEM

We will host Ulisses Braga-Neto from Texas A&M University at our data-driven physical simulation (#DDPS) seminar, Lawrence Livermore National Laboratory, on October 23rd, 2025. It will start 11 AM in California time. Anyone can join through the Webex room: https://lnkd.in/gqWzhexj We expect a great presentation, so please take advantage of it. Title: #Scientific #Machine #Learning: From #Physics-#Informed to #Data-#Driven Abstract: Scientific machine learning (#SciML) is a rapidly-expanding field that brings together scientific computation and machine learning. SciML became widely known initially through the success of physics-informed neural networks (PINN) and well as physics-informed Gaussian processes (PIGP). Both methods are collocation approaches that enforce the underlying differential equation at a grid of points. More recently, operator learning has emerged as a powerful SciML paradigm for parametric differential equation models. Unlike the previous approaches, operator learning is typically data-driven and does not require knowledge of the underlying differential equations. This is often encapsulated into the maxim "the physics is in the data". In this talk, we will review both the physics-informed and data-driven approaches, highlighting their advantages and disadvantages. We will describe some of our own contributions along the way, including recent results for the #Hasegawa – #Wakatani model of #plasma #turbulence in #tokamak fusion devices. Bio: Ulisses Braga-Neto received his Ph.D. in Electrical and Computer Engineering from The Johns Hopkins University in 2002. He is currently a Professor in the Electrical and Computer Engineering Department at Texas A&M University. His research focuses on Machine Learning and Statistical Signal Processing. Dr. Braga-Neto is the founding Director of the Scientific Machine Learning Lab at the Texas A&M Institute of Data Science (TAMIDS). He has published two textbooks and around 170 peer-reviewed articles and book chapters. Dr. Braga-Neto received the NSF CAREER Award in 2009. 📚 DDPS seminar is organized by libROM team ( www.librom.net ). 🗓️ DDPS seminar schedule: https://lnkd.in/gnGsx6qq