Institution profile

Ludwig-Maximilians-Universität München

Academic institutioneurope · de
Official website
Research library937linked papers
Opportunities0open roles
Selected work

Representative Papers

Conformal Prediction for Causal Effects of Continuous Treatments

Jul 03, 2024arXiv.org

This work addresses the open problem of quantifying uncertainty in causal effect estimation under continuous treatments, overcoming key limitations of existing conformal prediction methods—which are restricted to binary treatments and require known propensity scores. We propose the first model-agnostic, finite-sample valid conformal prediction framework for continuous interventions. Our approach explicitly accounts for the additional uncertainty induced by propensity score estimation, provides theoretically guaranteed prediction intervals, and includes an efficient algorithm for interval construction. Experiments on synthetic and real-world datasets demonstrate that our method strictly achieves the nominal statistical coverage level and significantly outperforms baseline approaches. To our knowledge, this is the first rigorously validated tool for constructing reliable confidence intervals for potential outcomes under continuous interventions—enabling trustworthy decision-making in safety-critical domains such as personalized medicine.

9 citations1 influentialRead paper

Flexible Covariate Adjustments in Regression Discontinuity Designs

Jul 16, 2021

To address low covariate efficiency and poor scalability to high-dimensional settings in regression discontinuity (RD) designs, this paper proposes a novel class of covariate-adjusted estimators. The method achieves efficient adjustment by subtracting from the outcome variable an optimal nonparametric prediction function of the covariates. Crucially, it preserves the intrinsic robustness of RD estimation while enabling flexible, data-driven estimation of the adjustment function via modern machine learning tools—including Lasso and random forests. Notably, it is the first RD adjustment framework that simultaneously attains asymptotic variance minimization and compatibility with machine learning estimators. Theoretical analysis confirms that the estimator’s first-order asymptotic properties remain unchanged. Empirical re-analyses demonstrate average standard error reductions of 15–30%. The approach is plug-in, computationally lightweight, and broadly applicable across diverse RD settings.

8 citations2 influentialRead paper

Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models

Jan 20, 2026

This work proposes a practical, three-stage “Locate–Guide–Improve” framework that transforms mechanistic interpretability from a post-hoc diagnostic tool into an engineering-driven optimization methodology for large language models. By systematically integrating techniques for identifying critical neurons and pathways with targeted interventions—such as activation manipulation and module editing—the framework establishes a standardized protocol for model refinement while clearly distinguishing between localization and guidance mechanisms. Empirical results demonstrate significant improvements in model alignment, task performance, and reasoning efficiency, thereby advancing mechanistic interpretability toward real-world applicability.

4 citationsRead paper

The DNA of Calabi-Yau Hypersurfaces

May 14, 2024arXiv.org

Optimizing axion phenomenology—specifically axion decay constants and axion–photon couplings—in string theory compactifications. Method: We introduce a Calabi–Yau threefold construction framework based on triangulations of 4D reflexive polytopes. To address exponential redundancy, we design a homotopy-class–aware parametrization of triangulations; combined with genetic algorithms and Bayesian hyperparameter optimization, we perform the first large-scale, efficient search across the full Kreuzer–Skarke list—including the polytope with maximal $h^{1,1} = 491$. Contribution/Results: Our approach significantly outperforms MCMC and simulated annealing in convergence speed and solution quality, yielding the largest axion–photon coupling strength reported to date. It demonstrates the feasibility of global optimization over the complete KS list and establishes the first systematic geometric optimization paradigm for string phenomenology.

4 citationsRead paper

Two-way Fixed Effects and Differences-in-Differences Estimators in Heterogeneous Adoption Designs

May 07, 2024

This paper addresses causal inference in two-period panel data under the “no pure control group” setting: all units receive a strictly positive, heterogeneous continuous treatment in period two, rendering conventional difference-in-differences (DID) inapplicable due to the absence of untreated (zero-dose) units. Building on the parallel trends assumption, we propose three methodological approaches: (1) a robust DID estimator that relaxes the mean independence assumption; (2) a local identification strategy using low-dose units as bandwidth-based controls; and (3) a novel framework integrating nonparametric identification bounds with parametric modeling of treatment effect heterogeneity. Relative to Pierce & Schott (2016) and Enikolopov et al. (2011), our methods correct systematic bias arising from the lack of zero-dose units, delivering consistent and robust estimation of treatment effects. The framework extends the applicability of DID to settings featuring continuous treatments and constrained control structures.

3 citationsRead paper
Recent publications

Latest Papers