Zohar Bronfman’s Post

🚨 Yes, it's possible for a machine learning model to learn *too* well. 🚨 Learn how to recognize and avoid this ML pitfall from Zohar Bronfman. P.S.: The Pecan platform helps you catch and correct this issue. 💪

Are rate adjustments driving away your best customers? When you leverage the power of machine learning to adequately rate risks, you can keep your best customers while protecting your bottom line. https://lnkd.in/e-MFxk64

Do you understand what it means to run things based on very sophisticated models that tell where, when, why, and how a tipping point might be reached? And what happens after that, based on variable scenarios and variable choices leading to variable results based on variable parameters?

When it comes to assessing a water main's risk of failure, should your community rely on traditional weighted models or cutting-edge machine learning? 🤔 In our latest blog, we break down the benefits and limitations of each, helping communities understand why predictive modeling offers the best balance of accuracy, efficiency, and actionability. From historical data patterns to advanced machine learning insights, discover which approach may be the most reliable for safeguarding your infrastructure 👉 https://hubs.li/Q02Y7MWs0

how do we get a machine (given a dataset and task) to design its own meaningful loss using first principles and clever inductive biases pertaining to the problem?

When you try to fix an issue using a model based on very wrong assumptions

Are you turning off entire segments to avoid risks? It’s a common approach, but it can cause auto insurers to miss out on potentially profitable business. This infographic shows how machine learning provides a better way, using the red corvette as an example. https://lnkd.in/erAC_XRj

A nice short article highlighting the importance of uncertainty estimation for decision making. A point prediction with known large uncertainty associated with it is so much more useful and informative for decision-making than a point estimate with high precision but unknown uncertainty. #datascience #uncertainty #risk #decisionmaking

get the model to -- step 1: brainstorm ideas using natural language step 2: turn clever natural language insights and inductive biases (in the context of a given problem) into heuristics and mathematical problem formulations step 3: turn the precise mathematical formulations into algorithms and architectures for implementation

Check out my new article on Medium where I dive deep into advanced regression models. From Linear Regression to Stacked Models, see how different techniques stack up and which model ultimately delivered the best results. Your thoughts and insights are welcome! 🙂