Jan Vitek

A few weeks ago I published a preprint proposing a mathematical model for canon generation based on the inverse 1–3–5 system and the Pythagorean table. I have now made the computational implementation of the model publicly available. The repository includes scripts that enumerate cyclic seeds and verify the structural limits of canon generation. Under these constraints, the system shows a clear boundary: canons with 8 voices are possible, while 9 voices become impossible. Both the theoretical paper and the software archive are available online: Preprint https://lnkd.in/dz4TdnrG Software archive (Zenodo) https://lnkd.in/d-cZRsyh The figure shows several canonic transformations generated from a single cyclic seed. Development repository https://lnkd.in/dau4BSwF #MusicTheory #ComputationalMusicology #OpenScience #AlgorithmicComposition

2

JuliaC introduces a new way to build executables and shared libraries directly from Julia code. In this session, Jeff Bezanson and Kristoffer Carlsson explain the new compiler interface and show how JuliaC enables compact binaries and cross-language deployment. https://lnkd.in/g7D76Wwt #JuliaLang #JuliaProgramming #JuliaC #ScientificComputing #TechnicalComputing #HighPerformanceComputing #EmbeddedSystems #SystemsProgramming

7

"evoxels: Differentiable Physics for Voxel-Based Microstructure Simulation" is now published in the Journal of Open Source Software🥳🥳 After a couple months of waiting and a genuinely thoughtful, constructive review process on GitHub, it’s wonderful to see it out in the world. Huge thanks to David Zwicker, Nils Meyer and Philip Cardiff. evoxels is a differentiable physics framework for voxel-based microstructure simulations, designed to connect imaging, simulation, inverse modeling, and machine learning in a unified open-source workflow. Big shout out to my amazing collaborators Alexander Cohen, Benjamin Dörich and Sam Cooper! I’m also really excited that there’s more to come soon, with new timestepping methods, multiphase-field simulations, and other improvements already in the pipeline. Updates soon 👀 And for all nerds out there who love language beyond coding, here's a little treat inspired by the Jungle Book by Rudyard Kipling: Law of the Micros Now this is the Law of the Micros, determined by fine structure scale; Those mastering it shall prosper, solvers that don’t become stale. Ions move between electrodes, differentiable physics go forward and back For the strength of the mesh is the voxel, and the strength of the voxel its pack. Model just as complex as it must be; use data, but only what's FAIR; For powerful compute is frugal - be mindful, resourceful, aware. Brute-force may chase after fortune, but friend, as your insight has grown, Remember: the solver's your hunter. Extract hidden truths of your own. Keep pace with the law of the physics - the transport, reactions, and flow. And honour the scale of the micro, for structure decides how things go. #MaterialsScience #Simulation #MachineLearning #OpenSource #Python

247

6 Comments

IntRS 2024 workshop at #RecSys2025 started with a presentation by Zuzanna Michalewicz on "(De)Composing the Algorithm: Explaining Music Recommender Systems to Artists for Understanding, Transparency, and Empowerment", co-authored by Eelco Herder, Karlijn Dinnissen, and Hanna Hauptmann. The paper shows an excellent case of applying user-centered design approaches (including co-design) for designing an interface for the *artists* in a music recommender systems. Also a great example of a dashboard design for these users. If you had no chance to listen, check this paper here https://lnkd.in/eksN5yae

23

🧐 Are you passionate about #circuitdesign, #memristors, and #ReRAM technology? Discover how massively #parallel #computation could be redefined! 👇 Imagine #networks of #memristors acting as HW accelerators for #shortestpath and complex #maze-solving tasks! Doesn´t this sound like the future of computation ? 🤯 💡 During my research, I stumbled upon an exciting discovery: specific memristive device compositions modeling #graph edges enable the detection of "closed #loops" in shortest path solutions. Curious? 🤔 To capture this journey, I wrote an article drawing a metaphor between the #mythological "thread of Ariadne", used to navigate the Minotaur's labyrinth, and modern circuits with #memristive grids, where device configurations act as #attractors guiding computation. (See figure below 👇 ) (Full link in comments) https://t.co/CEqjpSzQIT That was the most entertaining part of my research during my #PhD at Democritus University of Thrace - (D.U.Th.) 🤩 How do you see memristors and ReRAM technology shaping the future of #computation? Explore the details in the article and tag a colleague who’s into memristive devices or #unconventional computing! 😉 #complexity #embedded #NVM #RRAM #IMC #Edge-computing Department of Electrical and Computer Engineering, Democritus University of Thrace Departamento de Electrónica USM Advanced Center for Electrical and Electronic Engineering

35

1 Comment

📑 New Publication Alert: DRAG – Dual-Channel RAG for Hybrid-Modal Document Understanding! Thrilled to share an innovative study just published in Electronics MDPI—introducing DRAG, a game-changing Retrieval-Augmented Generation (RAG) framework that solves critical limitations in hybrid-modal document (text + tables + images) understanding! 📚 Article Title: DRAG: Dual-Channel Retrieval-Augmented Generation for Hybrid-Modal Document Understanding By Zhe Xin, Shuyuan Xia & Xin Guo* (School of Artificial Intelligence and Data Science, Hebei University of Technology, Tianjin, China) 🔗 Full Access: https://lnkd.in/gN-N_9US Why This Matters for Document Understanding Large Language Models (LLMs) excel at text processing but struggle with real-world hybrid-modal documents (e.g., biomedical reports, technical manuals, elderly care records) that combine text, tables, and images. Traditional RAG systems face two key flaws: 🔎 Text-based RAG relies on OCR and misses critical visual-structural information. 🔎 Vision-based RAG loses context in text-dense pages, leading to incomplete information retrieval. DRAG bridges this gap by unifying both channels, enabling reliable understanding of complex hybrid-modal content—critical for vertical domains like healthcare, elderly care, and technical documentation. 📌 Key Innovations of DRAG 1. Dual-Channel Parallel Architecture: Separately extracts and preserves two core information types: 📎 Visual-structural information (table layouts, image semantics, spatial relationships). 📎 Deep semantic information (text context, key concepts, logical connections).This ensures no critical detail is overlooked, whether from text-heavy sections or visual elements. 2. Dynamic Weighted Fusion Mechanism: Intelligently integrates retrieval results from both channels, assigning weights based on content relevance and modal characteristics. This precision screening ensures the LLM receives the most useful information segments for generation. Performance & Application Highlights 🔎 Delivers competitive results across general hybrid-modal benchmarks, outperforming single-channel RAG systems. 🔎 Shines in vertical domains: Excels on biomedical datasets and specialized scenarios like elderly care and rehabilitation—where documents are dense with hybrid-modal information. 🔎 Mitigates LLM hallucinations and improves interpretability by grounding generation in accurate, multi-modal retrieval results. About the Special Issue This article is part of the “AI-Driven Intelligent Systems in Energy, Healthcare, and Beyond”—focused on AI innovations for real-world vertical applications. Guest Editors: Prof. Dr. @Lin Meng, Dr. Xiangbo Kong and Dr. Hengyi Li 🔗 Full Access: https://lnkd.in/g9kvtsar #RetrievalAugmentedGeneration #HybridModalAI #LLMApplications #DocumentUnderstanding #BiomedicalAI #ElderlyCareTechnology #ResearchHighlight #MDPIElectronics #AIforDocuments #MultiModalAI

6

Building on LLMs? The hardest problem isn't getting "good" output. It's getting reproducible output every single time. LLMs are stochastic by nature. But production systems need guarantees. The solution is Stochastic Determinism: the engineering goal of constraining a probabilistic system until its variance is intentional, reproducible and acceptable. While building LlamaBrain (an open-source LLM governance framework), my architecture looked flawless on paper. I did some stress-testing on it with a real-world app and hit 96.7% consistency. Sounds high, right? It isn't. A 3.3% failure rate means your constraints break roughly 1 in 30 calls. At scale, that’s a systematic failure mode. Getting from 96.7% to 100% took ten days and required building a Black Box Audit Recorder to log, replay, and structurally compare every interaction. The lesson? Determinism isn't something you just design into an AI system. It's a property you must instrument, measure, and prove. Anything less is just a toy framework. #LLM #SoftwareEngineering #AI #OpenSource #LlamaBrain

5

2 Comments

📣 Published in Electronics! #DeepLearning Recommendations of #E-Education Based on #Clustering and Sequence Authors: Furkat Safarov, Alpamis Kutlimuratov, Akmalbek Bobomirzaevich Abdusalomov, Rashid Nasimov and Young-Im Cho 🔗 Read it here: https://lnkd.in/dmnWnKyA #ElectronicsJournal #MDPI #DeepLearning #RecommenderSystems #EdTech #EEducation #AI #MachineLearning #Research #AdaptiveLearning

1