AI & ML
Pragmatic DML with AI-Learned Representations
arXiv stat.ML0 views
arXiv:2610.01935v3 Announce Type: replace-cross Abstract: Text, images, and other rich objects increasingly enter causal analysis through AI-learned representations used as controls. For a broad class of estimands, replacing the raw object by its representation shifts the target by the inner product of two errors: one in the outcome regression and one in the balancing weight. The bias is therefore bounded by the product of their sizes. Three consequences follow. With a fixed representation, the usual cross-fitted debiased machine learning (DML) interval is valid for a representation-specific target, and for the causal target when the approximation bias is negligible relative to sampling error. When this bias may not be negligible, a sensitivity analysis bounds the causal target and supplies confidence bounds for the endpoints. When the embedding is tuned on the sample, three approaches preserve valid inference: fixed-menu fine-tuning, joint-loss tuning, and double representation selection. A multimodal demand application illustrates the workflow.
Source:https://arxiv.org/abs/2610.01935PreviousArbitrary-Accuracy Neural Approximation with Optimal Neuron Count and Optimal Bit ComplexityNextMixing Regularization for Average-Reward $Q$-Learning and $Q$-Planning
Related posts
Which app should you drive for? Compare them by net profit per hour, not by the payout
0dev.to Career
Who is hiring AI engineers right now? 682 companies, from OpenAI to ServiceNow (November 2026)
0dev.to Career
Who is hiring in the US right now? 2,991 companies and 113,000 open jobs (November 2026)
0dev.to Career
I opened three small services on Gumroad — voiceover, proofreading, and minimal logos
0dev.to Career
Why I Test My Downloader on a Real Android Phone, Not Just Desktop Chrome
0dev.to PHP

Day 2 of my 15-Day Graphic Design Challenge: From Rough Ideas to a Moodboard
0dev.to Beginners