Zohar Bronfman’s Post

Super important point from Pecan AI CEO, Zohar Bronfman on the issue of data leakage and how it must be avoided to create good predictive models. #ceotalks #predictiveanalytics #machinelearning #modeling

Outliers are data points that significantly deviate from the majority of the data. They can negatively impact machine learning models by biasing the results and reducing the model's accuracy and interpretability.

Perfect metrics alone don’t guarantee adoption. Welcome to the Semantic Layer Paradox. 🧩 As organizations race to implement AI assistants, semantic layers are becoming essential for accurate and reliable data querying. Yet even with flawless implementations by data teams, business users often resort to rebuilding metrics themselves. Why? Not because they doubt the data—but because they need to understand the logic behind it. Our CEO, Tristan Mayer explores why the future of data trust isn't just about having the right definitions for AI to use - it's about making those definitions transparent and understandable to everyone who needs them. 👉 Read more here: https://lnkd.in/ecmV6jHd

This is a really groundbreaking technology and I can understand how the #Automotive industry is just getting started to trust and use it, but it's already quite advanced! Looking forward to find more adopters! If you're in #Germany or #EU and are interested, please let me know! #ADAS #AD #AutomotiveIndustry #Technology #AI

Is synthetic data useful for ML planning/prediction beyond ML-based perception? 🤔 Hear our engineer’s thoughts on the role of synthetic data during this transition. Watch the full panel discussion to learn more about navigating challenges and accelerating development with synthetic data: https://lnkd.in/gJeuSA9c #syntheticdata #sensorsimulation #machinelearning

Keep your pulse on ML model development but your heart in good quality data. Cubist interpretation of above by ML

Implementation of the following machine learning concepts.Outlier Detection and removal, Hypothesis Testing,Data preprocessing, Regression,Classification and clustering

High quality, clean and big-sized data is more important than hyperparameter tuning of the model. Before jumping directly into model selection and training, you have to stop and take your time to do the following: — what question do you want to answer. i.e, what is the problem you want your model to solve? Can it be answered by machine learning? If yes, proceed. — what data should you feed your model? Is this data expressive of the question you want to answer? How can you collect this data? Is it ready (downloadable) or you have to collect it yourself for example by web scraping? Is it big enough to train a model? — After data collection comes the most important step: data cleaning, preprocessing and exploring. Before fitting the model on your data, you have to ensure it is clean because if not, do not expect high performing model. As they say ‘Garabage in, garbage out’, meaning that if you feed your model bad data, then expect bad results. To conclude, model training is only a very small percentage of the code in ML.

Unveiling insights from data one algorithm at a time. Today's focus: predictive modeling for customer churn. Harnessing the power of machine learning to drive business decisions. #DataScience #MachineLearning #PredictiveAnalytics

Money could buy happiness: Catastrophic "model collapse" (due to self-consuming loops) can be avoided at the extra cost of feedback on data quality (i.e via data pruning). https://lnkd.in/e2rECXyY