Banias Baabe

Last week I was grateful to be invited to present our current working paper “Causal Inference with Cocycles” at the Munich Center for Machine Learning’s (MCML) 3rd annual workshop on causal machine learning. Causal machine learning goes beyond predicting outcomes based on patterns, and aims to identify actual causes behind those outcomes. 🧐In our work, we look at what most state of the art methods have in common, which is the need to (often) estimate entire distributions/densities from data. 🧠We show that in many cases, one can avoid this by instead just estimating a “cocycle”, which describes how a conditional distribution changes under causal actions (eg interventions). 🔥In various situations, directly targeting the cocycles simplifies the estimation problem and improves robustness to misspecification. 💪🏼This means more reliable, robust answers to causal “what if” questions in practice! Anyone interested can find out more in our paper (see the link in the comments)! Thank you to Stefan Feuerriegel for the invitation and for organising a great selection of speakers covering causal discovery and estimation problems! It’s always inspiring for the more theoretical end of my work to learn about the applied problems researchers in academia and industry are grappling with!

81

1 Kommentar

Munich AI Founders Night: Last Chance Alert!🚨 Tomorrow is the day! Join us at the AI Founders Night and be part of Munich’s thriving AI ecosystem! 🌟 📅 When? Wednesday, November 6th at 6 PM 📍 Where? TUM.ai Homebase, Unite Office, Rosenheimer Str. 116A, 7th floor 🚀 This is your final opportunity to secure your spot at an event designed to connect AI visionaries, innovators, and ecosystem partners. With a lineup of inspiring speakers and a room full of ambitious entrepreneurs, this is an evening you won’t want to miss! Speakers include: ✨ Sophie Defauw (CEO, AthleteIQ) TUM.ai alumna, started Uni at 13, became a professor at age 22, has been a guest lecturer in sport tech at UC Berkeley and won the Elon Musk 2021 Not-a-Boring competition. Furthermore she was a pro-athlete in windsurfing, top 5 in North America and made national art exhibitions. ✨ Florian Scherl (CEO, FAST AI Movies) and Philipp Gabriel (CTO) Founders and former AI E-Lab participants who met during the program’s final pitch. Florian started coding at 12, developed ML models by 15, and Philip brings ML engineering experience from top companies like Microsoft, McKinsey and BMW. ✨ Antoine Leboyer (Managing Director, TUM Venture Labs AI & Software) A Harvard MBA and seasoned tech executive, Antoine leads an innovation hub at TUM Venture Lab dedicated to supporting tech-based startups in AI and software. His mission? Empower entrepreneurs to translate research into impactful ventures. Meet members from all partner initiatives including CDTM, Manage and More by UnternehmerTUM, START Munich, TEG | The Entrepreneurial Group, Enactus, Innovis VC, MUNICH INNOVATION ECOSYSTEM, Med-Dev, OneAIM, Young Founders Network e.V. and Gründungsmagnet. Spaces are filling fast, so RSVP now to ensure you’re part of the conversation shaping the future of AI in Munich! Let’s connect, learn, and inspire each other! 💡🤖 👉 Link in comments below #tumai #ai #artificialintelligence #machinelearning #AIEntrepreneurship #MunichAI #Networking #Innovation #Entrepreneurship #startups

62

5 Kommentare

Why you should not use LLMs for any specific use case? 🧠🤖👨🏽‍💼 Many times customers from different parts of the world ask me about what others are doing with Generative AI. They want to learn about good practices and how to avoid the most common pitfalls. Sometimes my felling, and my answer too is: “Everyone, is using LLMs, everywhere”. Unfortunately this is the biggest mistake. While LLMs like GPT, Claude or Gemini have impressive capabilities, they may not always be the best fit for every AI use case. Here’s why 👇 🔸Traditional ML Solutions: If the use case can be solved with traditional machine learning, opt for those instead. Simpler and more efficient. 🛠️ 🔸 High Costs 💸: Significant computational and monetary expenses. 🔸Latency Issues ⏱️: Longer response times, not ideal for real-time applications. 🔸Scalability Challenges 📉: Complex and resource-intensive deployment and maintenance. 🔸Environmental Impact 🌍: High energy consumption and carbon footprint. 🔸Overkill for Simple Tasks 🛠️: Unnecessary complexity and resource allocation for straightforward tasks. 🔸Ethical and Bias Concerns ⚖️: Risk of propagating biases and ethical issues. 🔸Data Privacy and Security 🔒: Potential risks in handling sensitive information #ArtificialIntelligence #MachineLearning #DeepLearning #AI #DataScience #BigData #NLP (Natural Language Processing) #ComputerVision #AIResearch #AITrends #AIEthics #TechInnovation #akilankm #Automation #SmartTechnology #PredictiveAnalytics #AIApplications #FutureOfWork #DataAnalytics #TechTrends

17

Call me biased.* But the best overview of Robustness in ML I've read is the one by Timo Freiesleben. You don't have to trust my biased opinion. Check it yourself: *It's a chapter in our joint book Supervised Machine Learning for Science 😅

52

1 Kommentar

Avoiding Data Leakage in Machine Learning 🚫 💡 Tip: Data leakage is a silent model killer. When building machine learning models, especially in Scikit-Learn, one critical aspect that often gets overlooked is data leakage. This occurs when information from outside the training dataset leaks into the model during training, leading to over-optimistic performance results. How to avoid it? • Split your data first: Always split your data into training and testing sets before any form of preprocessing (like scaling, encoding, etc.). This ensures that the model doesn’t ‘learn’ from data it shouldn’t have seen. • Cross-validation caution: Use techniques like StratifiedKFold when dealing with imbalanced datasets to maintain the distribution of classes. • Feature engineering: Be cautious when creating new features. Ensure that any feature derived from the target variable does not leak information. Data leakage can lead to significant performance drops in production, so it’s crucial to take proactive steps to prevent it from the get-go! 🚀 #DataScience #MachineLearning #DataLeakage #ModelBuilding #AI #MLTips #ScikitLearn #DataPreprocessing #DataQuality

1

Retrieval Augmented Generation (RAG) is a straightforward concept that any company can implement with little effort. You can think of RAG as an NN-based document summarization, like merging the results of a simple elastic search/apache solr query with a text summary. The original concept involved retraining the Llm to have a similar latent space, however in applied concepts the retraining has been dropped and replaced with a vector database and prompt engineering that can be paired with any Llm that exists. Based on Meta's LLAMA3, LlamaIndex, Qdrant, Ollama, Angular, and FastAPI, I built the RAG demo about funding lines for universities of applied sciences in hesse. The demo is available at (not well optimized for mobile use): https://lnkd.in/eBYQwKh4 This demo is a weekend project when the little ones were asleep, so it took me less than 3 days. Imagine your company creating a bot like this with more manpower and effort. Best of all, it's completely free and costs you nothing. Your data stays in your company without any security or GDPR violations. P.S.: On my own behalf: I also work as a speaker and consultant on AI topics. Through my research in context-aware fault diagnosis in industrial production lines, I know the state of the art in AI and have implemented various AI use cases myself using different neural networks under industry constraints. If you need some advice, just get in touch with me. I'm looking forward to talking with you about self-built solutions without buying expensive cloud services.

11

2 Kommentare

🚀 Unlock the Power of Gradient Boosting Machines (GBM)! 🚀 In my latest blog, I break down: 👉 What is GBM? 👉 How does it work for regression tasks? 👉 Step-by-step explanation of the gradient boosting process. 👉 Insights to help you better understand how the model is built from scratch. Curious about how GBM turns errors into opportunities for stronger predictions?💡Dive in to learn how it works and why it's one of the most powerful ensemble techniques in machine learning. Read more here: https://lnkd.in/gnHGYu9g #MachineLearning #GradientBoosting #DataScience #PredictiveModeling #GBM #AI #DataAnalytics

2

If only there was a model that did everything for us... Who knows, maybe there is. 🤷‍♀️ When building machine learning models, deciding which model to use takes more time than building the model itself. In this paper, you can see how I used pycaret to shorten my decision making process. 🏃‍♀️ kaggle link : https://l24.im/1mbpkTK #machinelearning #pycaret #datascience

17

I recommend his YouTube channel, "StatQuest", and his book, "The StatQuest Illustrated Guide to Machine Learning!!!". These are incredibly valuable resources for learning Machine Learning and Statistics. #ai #machinelearning #llm #statistics

15

1 Kommentar

Structured prior distributions for the covariance matrix in latent factor models

19

If you are curious on the current Gen AI panorama from an architectural (also technical) high level point of view, I've been invited to give my opinion and vision at the Software Architecture Conference 2024! August 15-16 I'll provide some of my insights and points of view on ‘AI Architecture: What Is Possible Today and Future Trends’ 😊 If curious, here is the RSVP link: https://lnkd.in/dCTr-A4y #genai #ai #aimvp #mvp #mvpBuzz #Architecture #AI #AIAgents #Cloud

19

This morning I woke up to a pleasant e-mail from Google Scholar: My most popular paper got cited 100 times! Back in 2022, Maximilian Nitsche - a former student of mine, who is now a Data Science & AI Consultant at IBM - and I worked intensively on our paper 𝗔𝗻𝗼𝗺𝗮𝗹𝘆 𝗗𝗲𝘁𝗲𝗰𝘁𝗶𝗼𝗻 𝗶𝗻 𝗔𝘂𝘁𝗼𝗻𝗼𝗺𝗼𝘂𝘀 𝗗𝗿𝗶𝘃𝗶𝗻𝗴: 𝗔 𝗦𝘂𝗿𝘃𝗲𝘆. Back then, I realized that there was no good entry point to the field of Anomaly Detection in autonomous driving, and this is just what we created. In the paper, which got accepted at a CVPR workshop, we provide an overview of 50 anomaly detection methods across all sensor modalities, categorized into five different detection techniques as proposed by Jasmin Breitenstein. Since then, a lot has changed, but many of the insights are still true, especially since there is a lack of anomaly detection methods that go beyond camera data but utilize lidar or a combination of both. One of the core issues was that only benchmarks with camera data were available. Thus, we recently released 𝗔𝗻𝗼𝗩𝗼𝘅, a sensor-agnostic benchmark for multimodal anomaly detection in autonomous driving. Also, camera-based methods have peaked, with near-perfect scores on the common benchmarks Fishyscapes and SegmentMeIfYouCan. A great example of this is the recent UNO method by Anja Delić and Matej Grcić. This was achieved by powerful semantic segmentation models and outlier exposure during training. A paradigm shift is needed to continue making progress on that front. I'm excited to see how much further the field of anomaly detection will progress, now that large language and vision models are out there! Anomaly Detection Survey: https://lnkd.in/eVpxyfB4 Guidance on Detection: https://lnkd.in/eRSvQaXF AnoVox Benchmark: https://lnkd.in/eMbGiYwb UNO Outlier Detection: https://lnkd.in/eqhCF8D5 FZI Forschungszentrum Informatik KIT Wirtschaftswissenschaften #phd #anomalydection #ood #autonomousdriving #publication #deeplearning #machinelearning

44

4 Kommentare

🚀 Demystifying OpenAI’s New Model: #o1 🤖 The buzz around #OpenAI’s o1 model is impossible to ignore—but let’s take a moment to break it down, beyond the hype. Many people see these advanced AI models as magical, mysterious black boxes. But here’s the thing: AI, especially models like o1, aren’t magic—they’re systems designed with layers of logic and reasoning, much like how our own brains approach complex problems. Imagine you’re in a brainstorming session. 🧠 You start with basic ideas, then gradually add layers of depth, considering different angles, combining thoughts, and eventually reaching a refined conclusion. That’s how models like o1 work too. Just as a deep neural network has layers—each contributing something new and unique—the o1 model processes information in layers of reasoning. It starts with simple, foundational understanding and builds toward more complex insights. It’s like how a human would reason through a problem, one step at a time. Rather than being a mystical force, the o1 model is a tool that mimics this human-like reasoning, breaking problems down into understandable pieces and working through them logically. It’s closer to having a collaborative brainstorming partner than anything magical. Let’s embrace the real power of these models: making reasoning more accessible and scalable—not mysterious. #AI #OpenAI #DeepLearning #NeuralNetworks #AIReasoning #Innovation #Brainstorming #FutureTech #LogicNotMagic #LetsBreakItDownLogically

28

The MLOps Community podcast was one of the podcasts I always enjoyed listening to. So, it was particularly exciting that Demetrios Brinkmann had asked me to contribute. In the latest Coffee Session episode, you can hear me talk about how data visualization and embeddings can support you in understanding your machine learning data. Here are some insights from our discussion: - Visualization Techniques: I discuss using UMAP to simplify data into two-dimensional maps to identify clusters and anomalies in large ML datasets. - Practical Applications: We explore how these techniques are applied in real-world scenarios, particularly in collaborations with automotive companies to enhance simulations and data analysis. - Challenges with Data: We address the challenges of large, unstructured datasets and the need for better tools. - Feedback and Iteration: We talk about the importance of incorporating real user feedback into the development of models and visualizations. 🎙️ Visualize - Bringing Structure to Unstructured Data // Markus Stoll // #258 https://lnkd.in/gyHpYr-V

29

2 Kommentare