Machine Learning
7 articles
Logistic regression beats LLM readouts on survey prediction
On a real survey prediction task, a plain logistic regression beat a language-model activation pipeline, AUC 0.769 vs 0.747; steering changed nothing useful.
Prediction-powered inference corrects AI survey imputation
Treating AI-imputed survey responses as data understated prevalence threefold; a regression adjustment lets predictions sharpen estimates without harm.
Steering vectors estimate an average regression gradient
Activation steering approximates an average regression gradient, cosine 0.63 in our data, and classic estimator choices change how hard the vector steers.
How to Build a Classifier When 94% Accuracy Means Nothing
A scikit-learn workflow for imbalanced classification, where 94% accuracy can mean nothing, and the evaluation metrics that actually matter for rare classes.
How to Interpret a Classifier with SHAP Values
A Python workflow using SHAP values to explain what drives a fraud classifier's predictions, and what SHAP importance does and does not actually measure.
Can a Classifier Find What Investigators Miss? - Too Early To Say
A prospective classifier captures 1 in 4 future fraud exclusions from billing intensity alone. The signal is fraud-specific.
400 Labels to 94% Accuracy: Validating Grocery Store Data
Google Places returned thousands of 'grocery stores.' Many weren't. Here's how a classifier separates real grocery stores from gas stations and liquor stores.