New course on Edge Machine Learning Systems Engineering at Universidade Federal de Itajubá

Building a Curiosity Engine: From a Simple Mistake to a New AI Tool As a physics teacher, I thought I knew what I was looking at when I saw an old picture of a copper device. "Calorimeter," I thought. But the stirrer looked too complex, like an circular set of paddles. I was curious: why? That simple question sparked a project. I realized we often teach students labels, not the stories behind them. I wanted a tool to bridge that gap. So, I built the Academic Image Analyzer, an AI-powered "Curiosity Engine." The goal? To turn any scientific image into a story of discovery. After several iterations of prompt engineering and fine-tuning the AI's "creativity" (its temperature), the tool was ready. The results were amazing. 1. That "calorimeter" was actually Joule's paddle wheel apparatus, and the complex paddles were the key to his experiment on the mechanical equivalent of heat. 2. A weird-looking flask became Pasteur's swan-neck flask, which disproved spontaneous generation. 3. A glass tube that looked like a broken lightbulb was revealed to be a Crookes Tube, providing early evidence of the electron. This journey taught me that the best tools are often born from our own questions. It's not just about identifying an object; it's about uncovering the incredible story hidden in its strangest details. I'm excited to keep developing this to help teachers and students everywhere. What do you think? Read the complete story @ https://lnkd.in/gsdSKaS8 . The link for the app can be found in the blog itself #EdTech #AIinEducation #Physics #STEM #Innovation #TeachingTools #IBDP #Gemini

"this makes me excited for my career" That's what a student at Calpoly said after doing a short course on Materials Informatics using the Citrine Platform. In industry AI is being embraced. Once this year's cohort graduate they will need to have data and model interpretation skills to use the tools found in materials and chemicals R&D. Citrine has been working with educators at Cal Poly and USC to understand what would be needed to incorporate AI seamlessly into the curriculum. Attend the webinar to find out more. https://lnkd.in/ey54YwH4

Knowledge grows when shared. I’ve just made public four GitHub repositories from my MSc journey at K. N. Toosi University of Technology, featuring over 20 academic projects in control systems and artificial intelligence. Each project includes complete simulation code, detailed documentation, and visualizations. I hope these resources support fellow students and learners on their educational path and contribute meaningfully to the open-source community. - Machine Learning Course Topics: Supervised, Unsupervised, and Reinforcement Learning Instructor: Dr. Mahdi Aliyari-Shoorehdeli 🔗 https://lnkd.in/d5_4sY2n - Optimal Control Course Topics: Optimization, LQR, LQT, and LQG Design Instructor: Prof. Hamid Khaloozadeh 🔗 https://lnkd.in/dqistRpU - System Identification Course Topics: Nonparametric, Parametric, and Recursive methods Instructor: Prof. Hamid Khaloozadeh 🔗 https://lnkd.in/dr8VwMTc - Model Predictive Control Course Topics: Linear, Tube, Hybrid, and Explicit MPC Design Instructor: Dr. Amirhossein nikoofard 🔗 https://lnkd.in/dywuZVeJ #Control_systems #AI #Machine_leanring #Optimal_Control #MPC #System_Identification #KNTU

Learning is constant, the evolving self a reflection of understanding our coexistence. MIT brings cognitive understanding by analyzing visual representations. I want to bring this further, with purpose, to create more diverse synaptic pathways of knowledge and synaptic mapping with greater diversities of experience. abstraction can then be contextually evolving as our context evolves, we are not a static representation of time https://lnkd.in/gkD_bXXx

Alice Barbora Tumpach SPEAKER at LOGML (London Geometry and Machine Learning) LOGML aims to bring together mathematicians and computer scientists to collaborate on a variety of problems at the intersection of geometry and machine learning. https://lnkd.in/eajzmP-W Alice Barbara Tumpach is Research Director at Wolfgang Pauli Institute (WPI), Vienna, and deputy head of the Mathematical Artificial Intelligence and Machine Learning Research Division of WPI, leading the Geometry-Informed Machine Learning Group. She is an expert on infinite-dimensional Differential Geometry, a research topic at the intersection of Geometry, Functional Analysis and Algebra, and a fundamental cornerstone for finite dimensional Geometry, Optimization, Probability and Mathematical Physics. Together with her collaborators, she develops news theoretical approaches to deal with group actions and their invariants, as well as new solutions for machine learning tasks in the presence of an infinite-dimensional group of symmetries, typically a diffeomorphism group or a gauge group. She is P.I. of several reseach projects including FWF-projects entitled “Geometric green Learning on groups and quotient spaces” , and “Banach Poisson-Lie Groups and integrable systems”. She is a reviewer for several international research agencies and for international mathematics and computer science journals. During her studies, she trained at the Ecole Normale Supérieure d’Ulm and the Ecole Polytechnique, France, where she obtained her doctorate. 

I’m thrilled to share that my research paper, "Dynamic and Regularized Visual Prompt Tuning for Few-Shot Image Classification," was presented at the Fifth International Conference on Emerging Research in Electronics, Computer Science, and Technology (ICERECT-2025)! ✨ The conference, held at P.E.S. College of Engineering, Mandya, and technically sponsored by the IEEE Bangalore Section, was an incredible platform to engage with the latest trends in tech and research. Our work introduces DRVPT, a novel method designed to improve Vision Transformers (ViTs) in few-shot learning scenarios—where models must learn to recognize new objects with only a few examples. By using a lightweight MetaNet to generate dynamic, task-specific prompts, and incorporating prompt-level regularization, DRVPT achieves: Superior accuracy Greater stability Computational efficiency—all without the need for full model fine-tuning. We saw outstanding performance on benchmark datasets like Oxford 102 Flowers and CIFAR-100! This journey wouldn't have been possible without the unwavering support and mentorship of my professors Shruti Mathur and Dr. Nilesh Marathe from the Department of Computer Science and Engineering (Data Science) at DJSCE. Their guidance and encouragement were key to this achievement. #Research #MachineLearning #AI #DeepLearning #FewShotLearning #VisionTransformer #PromptTuning #ICERECT2025 #IEEE #AcademicResearch #DRVPT #DJSCE

Another Day Another Conference - This time KATARINA project conference. The energy of collaboration truly made time fly! As one of active members of experimental projects, it's always a pleasure to connect with fellow teachers and discuss the evolving landscape of computer science education. We had two insightful talks today. First by Silvia Joachim explored bringing machine learning and AI architecture to unplugged computer science, using tangible "mushrooms" to represent data within machine learning structures. This approach really helps students visualize how models produce answers. The second talk carried out by @MonicaMcGill delved into research on the future pathways for students (CSEdResearch .org), considering how jobs and their descriptions change with AI's influence. It's clear we need to think creatively about students' future journeys and how they are connected to the CS. Our Network Lab presentation was a hit! 😊  I explained how hard it was bringing the topic of networking to life and that it was fantastic to see students engaged with Network Lab, sending packets to IP addresses, discovering their content, and solving mysteries to learn about networking. We also had a productive discussion about the gymnasium curriculum for 14 to 18-year-old students. We explored what programming concepts and algorithms are most appropriate for all students, not just those already excited about computer science. It was an intense but rewarding day, filled with valuable discussions and connections with everyone involved in these vital Slovenian experimental projects. These initiatives, supported by government subsidies, are making computer science accessible to a wider public, and it's inspiring to be a part of it. Thank for having us. #ComputerScienceEducation #EdTech #AIinEducation #STEMEducation #TeacherLife #SlovenianEducation #NetworkLab #CurriculumDevelopment #CobieAI

Super excited to introduce my new paper: ✨ AnyUp: Universal Feature Upsampling 🔎 Upsample any feature - really any feature - with the same upsampler, no need for cumbersome retraining. SOTA feature upsampling results while being feature-agnostic at inference time. 🌐 wimmerth.github.io/anyup/ Foundation models like DINO or CLIP are used in almost all modern computer vision applications. However, their features are of low resolution and many applications need pixel-wise features instead. AnyUp can upsample any features of any dimensionality to any resolution. AnyUp is a lightweight model that uses a feature-agnostic layer to obtain a canonical representation that is independent of the input dimensionality. Together with window attention-based upsampling, a new training pipeline and consistency regularization, we achieve SOTA results. When performing linear probing for semantic segmentation or normal and depth estimation, AnyUp consistently outperforms prior upsamplers. Importantly, the upsampled features also stay faithful to the input feature space, as we show in experiments with pre-trained DINOv2 probes. Generalization: AnyUp is the first learned upsampler that can be applied out-of-the-box to other features of potentially different dimensionality. In our experiments, we show that it matches encoder-specific upsamplers and that trends between different model sizes are preserved. Try it out now! Code and model weights are public. 💻 Code: github.com/wimmerth/anyup Great collaboration with Prune Truong, Marie-Julie Rakotosaona, Michael Oechsle, Federico Tombari, Bernt Schiele, and Jan Eric Lenssen. CC: Max Planck Institute for Informatics, Max Planck ETH Center for Learning Systems

Do you need to upsample vision features (e.g. DINOv3) to higher resolutions? 👉 If yes, check out Thomas Wimmer's new work AnyUp, it works on all vision features without retraining and can upsample to any resolution! Code is available! Great collaboration with our Google partners in the Via Center. Project Page: https://lnkd.in/etHiXZng Thomas Wimmer Prune Truong Marie-Julie Rakotosaona Michael Oechsle Federico Tombari Bernt Schiele Jan Eric Lenssen Max Planck Institute for Informatics Computer Vision and Machine Learning Department (D2) at the Max Planck Institute for Informatics

Can synthetic queries replace human ones❓ In this short video 🎥 , Sara Al Lawati and I share the key findings from our CIKM 2025 paper on LLM vs. human queries in information retrieval. Our study demonstrates that while synthetic queries can be beneficial for various tasks, they also introduce biases and cannot fully replicate human queries when simulating real user behavior. 👉 Watch the tl;dr video (yes, we talk straight to the camera 😅) and get the main takeaways in just a couple of minutes! #InformationRetrieval #CIKM2025 #GenerativeAI #Search ARC Centre of Excellence for Automated Decision-Making and Society RMIT School of Computing Technologies | RMIT STEM College