Asymmetrical Latent Representation for Individual Treatment Effect Modeling
Armand Lacombe, Michèle Sebag · 2025
Read Asymmetrical Latent Representation for Individual Treatment Effect Modeling by Armand Lacombe, Michèle Sebag (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-14006.txt.
Description
Armand Lacombe and Michèle Sebag split the hidden space in two to sharpen estimates of who gains from a treatment, in a 2025 paper. The task, called CATE, asks how one person's outcome shifts with a drug, a policy, or an ad. Standard designs squeeze treated and control groups into one shared latent map, balanced so neither rules. Their method, ALRITE, instead builds twin maps: one tuned to foretell control outcomes, one for treated outcomes.
Each map can bend toward its own group's quirks instead of splitting the difference. Under mild assumptions the authors prove a cap on the method's error for varied effects. On benchmark tests the twin maps beat the leading shared-map rivals. The approach borrows its split logic from domain adaptation, where separate spaces serve separate worlds. Health, social, and ad work all stand to gain, the authors note.
Catalog record
| CREATOR | Armand Lacombe, Michèle Sebag |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 83 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2501-14006