Laurenz Sommerlad

🔎 𝗫𝗔𝗜 𝗖𝗼𝗻𝗳𝗲𝗿𝗲𝗻𝗰𝗲 𝟮𝟬𝟮𝟱 - 𝗦𝗽𝗲𝗰𝗶𝗮𝗹 𝗧𝗿𝗮𝗰𝗸 𝗦𝗽𝗼𝘁𝗹𝗶𝗴𝗵𝘁: Debugging and improving models with xai 🖥️ ✏️ 𝗔𝘂𝘁𝗵𝗼𝗿𝘀: @christopher j anders Sebastian Lapuschkin Frederik Pahde Maximilian Dreyer Lorenz Linhardt 🔗 𝗟𝗲𝗮𝗿𝗻 𝗠𝗼𝗿𝗲 𝗛𝗲𝗿𝗲: https://lnkd.in/dz2UZf9T This track aims to debug, modify, and evaluate machine learning models to enhance their performance and reliability using eXplainable AI (XAI) techniques. The topics include, but are not limited to, (1) identifying and mitigating the effect of learned spurious correlations (also known as Clever Hans effect or shortcut learning) and distribution shifts in the data, (2) diagnosing failure modes and edge cases leading to undesired model behaviour, and (3) improving models through targeted corrections. For each of these points, XAI methods serve as a critical foundation, as they enable to understand model decision strategies (and their flaws), and to identify critical parts of the model or data set that are most amenable to improvement. Additionally, this track covers the development and application of rigorous evaluation methods for verifying that the enhancements achieve the intended outcomes. Join us to discuss best practices, emerging tools, innovative techniques, and case studies showcasing how thoughtful model refinement can lead to fair, robust, and responsible AI solutions. #XAI #debug #machinelearning #spurious #correlations #cleverhans #understanding #model #decisions #artificialintelligence

56

🚀 Exciting News! 🚀 We are thrilled to announce that Kumo AI's position paper "Relational Deep Learning: Graph Representation Learning on Relational Databases," has been accepted at #ICML 2024! 🎉 Join us in Vienna from July 21-27, where our very own Matthias Fey will be presenting this exciting new research area. 🔗 https://icml.cc/ In collaboration with Stanford University, we introduce Relational Deep Learning (#RDL), a new research area for applying deep learning on relational databases by viewing them as temporal, heterogeneous graphs. This method eliminates the need for manual feature engineering, leading to more accurate models built faster. 🚀 We’ve also developed #RelBench, a benchmark suite to facilitate research in this new area, using diverse datasets from Stack Exchange discussions to Amazon book reviews. 📚 Stay tuned for more updates on our presentation format, and check out our paper here: https://lnkd.in/dk24vbSg #ICML2024 #MachineLearning #AI #GraphNeuralNetworks #RelationalDeepLearning #DataScience #KumoAI #StanfordCollaboration

38

🚀 𝗢𝘂𝘁 𝗻𝗼𝘄: 𝗧𝘄𝗼 𝗵𝗲𝗶𝘀𝗲 𝗮𝗿𝘁𝗶𝗰𝗹𝗲𝘀 𝗼𝗻 𝗟𝗟𝗠 Alexander Schatten, senior researcher at SBA Research, and Marco Jeppe Reiser have published an interesting article: 📰 𝗪𝗵𝗮𝘁 𝗽𝗿𝗼𝗴𝗿𝗮𝗺𝗺𝗶𝗻𝗴 𝗮𝘁 𝗽𝗿𝗼𝗺𝗽𝘁 𝗹𝗲𝘃𝗲𝗹 𝗹𝗲𝗮𝗱𝘀 𝘁𝗼 Where can AI systems really help programmers, and is the approach taken here truly novel? This article examines their role. Programming involves interpreting real-world problems and processes, abstracting them, and then translating them into a deterministic language understood by technical systems. The goal is to automate tasks and processes and interact with people or other systems as needed. Besides understanding the real-world problem, programming also requires technical consideration of advantages and disadvantages, costs, legal frameworks, dependencies, and more. Actual coding is only a small part of the whole process. Alexander Schatten's second article on heise online explores the role of LLMs in software development. 📰 𝗟𝗟𝗠𝘀 𝗶𝗻 𝘀𝗼𝗳𝘁𝘄𝗮𝗿𝗲 𝗱𝗲𝘃𝗲𝗹𝗼𝗽𝗺𝗲𝗻𝘁: 𝘄𝗵𝗮𝘁 𝘁𝗵𝗲𝘆 𝗯𝗿𝗶𝗻𝗴 𝗮𝗻𝗱 𝘄𝗵𝗲𝗿𝗲 𝘁𝗵𝗲𝘆 𝗵𝗮𝘃𝗲 𝗳𝗮𝗶𝗹𝗲𝗱 Software systems are a combination of hardware, software, and the people who develop, maintain, and use these systems. This interaction of the three factors creates benefits and the possibility of further system development. Early on, software exhibited characteristics not observed in other technical systems. The first software crisis in the 1960s revealed that hardware and software capabilities were growing faster than those of human engineers, a crisis from which we have not yet fully emerged. One reason is the ever-increasing complexity of software itself and the complexity created by software and its increasing networking with other systems, affecting software further. Get the full articles here:   https://lnkd.in/gA2hHCBb   https://lnkd.in/giAV6f8C    Both articles will soon be published in iX magazine. 📷: Marcel Lehner 

19

#SimTech researchers Jonas Kneifl and Jörg Fehr from the Universität Stuttgart, along with Steven Brunton and J. Nathan Kutz from the University of Washington, have developed a new framework for modeling three-dimensional dynamical systems. 🌐🤖 Using Graph Convolutional Neural Networks (#GCNNs), they create multi-level surrogate models, making simulation processes more efficient and accessible. 🎮⚙️ This method accurately captures complex dynamics, such as a frontal impact scenario of a racing kart, and caters to different hardware requirements. 🏎️💥 A significant advancement in computer-aided engineering, enabling high-accuracy simulations with reduced computational effort. 🚀🔧 Read more: https://shorturl.at/Gvkw2

118

3 Kommentare

Here is this month’s list of where you can find us for more impactful sessions! Check us out and stay tuned for more details + how to register. 📍 VoxxedDays CERN in Geneva, Switzerland: “TrashTalk - Exploring the JVM memory management” January 14th, 2025 📍 JUG Oberpfalz, Germany: “TrashTalk - Exploring the JVM memory management” January 21st, 2025 📍 PrimeForce Webinar, Zurich, Switzerland: “Wargames - Java vulnerabilities and why you should care” January 28th, 2025 #AzulEvents #JVM #Java #JavaDeveloper #JavaVulnerabilities #Javasupport

5

The Rails 8 feature set perfectly complements SQLite's power in creating resilient, high-performance production apps, but still the question lingers: Can I really go all-in on #SQLite? Join Stephen Margheim at #RailsWorld as he illustrates how to leverage Rails and SQLite's full potential in your next venture, and when SQLite does and doesn't make sense for your application. https://lnkd.in/eaBstScy

52

1 Kommentar

We’re back with another list! Here are the top 6 resources to learn about large language models (LLMs) and generative AI for FREE 🎉 1. Large Language Models by ETH Zurich 📚 Description: This course covers the principles and foundations of large language models, focusing on their architecture, training, and applications. Highlights: Model architecture Training techniques Real-world applications 🔗https://buff.ly/4fgvSew 2. Understanding Large Language Models by Princeton 📚 Description: Dive into the complexities of large language models with clear explanations and practical examples to understand how they work and their potential. Highlights: In-depth explanations Practical examples Advanced concepts 🔗 https://buff.ly/3W33VOK 3. Transformers Course by Huggingface 📚 Description: Huggingface offers an insightful course on Transformers, the backbone of many modern LLMs. Learn about their design, functionality, and practical implementation. Highlights: Transformer architecture Practical implementations Real-world examples 🔗https://buff.ly/3y9xgiA 4. NLP Course by Huggingface 📚 Description: A comprehensive course focusing on Natural Language Processing using state-of-the-art models. Ideal for those looking to understand and implement NLP techniques. Highlights: State-of-the-art NLP models Implementation guides Practical NLP projects 🔗https://buff.ly/3y9xgiA 5. CS324 - Large Language Models by Stanford 📚 Description: Stanford's CS324 course offers an in-depth exploration of large language models, including hands-on projects and detailed lectures by experts. Highlights: Detailed lectures Hands-on projects Expert insights 🔗https://buff.ly/4cRAAh9 6. Generative AI with Large Language Models by Coursera 📚 Description: Learn how to use large language models for generative AI applications. This course covers both theoretical concepts and practical implementations. Highlights: Theory and practice Hands-on exercises Generative AI applications 🔗 https://buff.ly/3Eyg2uC #AI #LLM #GenerativeAI #FreeLearning #TechEducation

4

Medical AI Research alert! MedGraphRAG: team of researchers has developed a novel graph-based Retrieval-Augmented Generation (RAG) framework for enhancing Large Language Models (LLMs) in the medical domain. Here's why MedGraphRAG is interesting: Key Features: - Addresses challenges in applying LLMs to specialized medical fields - Uses a 3-tier hierarchical graph construction for knowledge integration - Implements hybrid static-semantic document chunking for better context understanding - Employs a U-retrieve strategy for efficient information retrieval Innovations: - Entity extraction and linking across multiple knowledge tiers - Graph construction with relationship scoring - Evidence-based responses with source citations - Explanation of medical terms for non-experts Impressive Results: - Outperforms many fine-tuned models on medical benchmarks (PubMedQA, MedMCQA, USMLE) - GPT-4 + MedGraphRAG achieves 91.3% on MedQA, setting a new state-of-the-art - Substantial improvements for smaller models, making advanced medical AI more accessible Potential Impact: - Enhanced clinical decision support systems - Improved medical education tools - Accelerated research in complex medical domains - Increased trust in medical AI through transparency and explainability Future Directions: - Testing with more diverse datasets - Exploring real-time clinical applications This research represents a significant step forward in making AI more reliable, transparent, and effective in healthcare settings. As we continue to advance in this field, it's crucial to balance innovation with patient safety and ethical considerations. What are your thoughts on the potential of MedGraphRAG in healthcare? How do you see AI reshaping medical practice in the coming years? Check the full thread https://lnkd.in/gAjybAkA #MedicalResearch #medical #clinical #clinicaltrials #healthcare #health #Radiology #pathology #llm #chatgpt #GPT4o #claude #google #genai #ai #NLProc #academia #nature #huggingface #meta #Harvard #Stanford #pfizer #AstraZeneca #gilead

25

Preprint analysis focusing on the use of AI as a tool in #OpenScience. Thinking about these AI-related challenges and opportunities at your organization? Delta Think can help you understand, define, and implement your next steps. Please reach out today to get started. #ScholComm #AI

2