Searcharxiv⌕ Search

arXiv subjects

Vasilis Syrgkanis

Publications and source records attributed to Vasilis Syrgkanis.

At least 19 recordsLinked to original sources
arXiv2026

Glyph: An Interpretability Engine for AI-Driven Scientific Discovery in Clinical Data

Automated scientific discovery has the potential to advance understanding of disease mechanisms and accelerate identification of early disease markers. Manual data annotation is infeasible at population scale, motivating methods that automatically extract pre-defined phenotypes in multimodal and unstructured clinical data. Such pre-defined feature sets determine and limit which associations these approaches can discover. This limitation extends to agentic AI systems that use these tools to query population-scale data. In contrast, foundation models (FMs) trained on large volumes of clinical data can perform tasks beyond their training objectives. This suggests their latent embeddings may represent novel or unexplored features that could yield insights into disease mechanisms, yet such latent features remain difficult to interpret. To address this challenge, we introduce Glyph, an automated multistage framework that extracts features from multimodal and unstructured clinical data using FMs and sparse autoencoders, interprets them as coded clinical concepts, and evaluates their value for clinical scientific discovery. We applied Glyph to abdominal CT volumes, CT radiology reports and laboratory measurements from 41,628 patients across different institutions. We identified 397 predictive concepts supported by 4,463 Bonferroni-significant associations. Moreover, a majority of discovered associations were automatically reproducible in held-out cohorts across different sites. Glyph concepts provided complementary associations to literature-derived markers, leading to gains in predictive power. We envision that interpretability can extend beyond a mechanistic understanding of neural networks to connecting implicitly learned features with clinical knowledge and establishing their significance for patient outcomes. We release Glyph at https://github.com/StanfordMIMI/Glyph.

cs.AI↗
arXiv2026

The Double Diagonal Estimator: Reducing Bias in Two-Sided Marketplace Experiments

Interference between participants in two-sided marketplace experiments can severely bias conventional A/B tests. While two-sided randomization (TSR), which randomizes both supply and demand, enables estimators that mitigate this bias, existing analyses offer limited guidance on which estimator to prefer for the balanced market conditions common in practice, as they largely focus on extreme supply- or demand-constrained regimes. We introduce the Double Diagonal (DD) estimator, a novel method for interference-robust treatment effect estimation. The DD estimator contrasts the outcomes of treated-treated and control-control pairs and uses the off-diagonal interactions (treated-control and control-treated) to correct for interference-induced bias. We provide the first theoretical comparison between the DD and the standard naive TSR (TSRN) estimator under a simplifying mean-field market model, proving two key results: (i) DD's bias is strictly smaller than the standard estimator's below a critical threshold of the outside-option parameter; in particular, DD's bias is smaller whenever the outside option is at most twice as attractive as platform options; and (ii) the two estimators exhibit biases in opposite directions, with DD underestimating and the standard estimator overestimating the true effect, respectively. Extensive numerical experiments in realistic market settings--spanning heterogeneous treatment effects, supply-demand imbalances, and varied market structures--corroborate our theoretical findings. Across a wide range of configurations, the DD estimator demonstrates the lowest bias, establishing it as a robust and often superior choice for reliable experimentation in two-sided marketplaces.

econ.EM↗
arXiv2026

PaperFormalizer: Author-in-the-Loop Autoformalization of Research Papers in Lean

Even a carefully proofread paper can leave small gaps in its proofs or its stated assumptions. We introduce PaperFormalizer, a Lean-based agentic pipeline that helps authors find and close such gaps in their own papers and turn each paper into a Lean-verified version. Starting from the LaTeX source, PaperFormalizer transcribes the paper into a dependency graph of statements, audits each statement in natural language for faithfulness, ambiguity and correctness, and then formalizes and proves the statements in Lean against Mathlib, optional domain libraries and the author's personal toolkit. Only the author knows what a statement is meant to say and how a formalization gap in a result is best fixed. The pipeline therefore stops and asks the author whenever a statement admits more than one reading, presents every repair, in natural language and TeX symbols, as a proposal that the author can accept, decline or redirect, and leaves to the author every decision that could potentially change what the paper claims. The output is an annotated and revised version of the paper, a report of the changes, and Lean code in which each declaration is linked to the statement it formalizes. The author's toolkit grows as more papers are formalized, and authors can share their verified papers so that researchers who build on them can reuse their Lean code. By making verified papers practical to produce, PaperFormalizer aims to make research that rests on mathematical proofs more robust. PaperFormalizer runs inside a coding agent (Claude Code or Codex) and is available at https://github.com/Jiyuan-Tan/PaperFormalizer. Papers with Lean code shared by users are available at https://paperformalizer.org.

cs.AI↗
arXiv2026

Model Selection for Off-Policy Evaluation via Pairwise Bellman Validation

Model selection and hyperparameter tuning pose a central challenge in off-policy evaluation (OPE) because errors of fitted Q-functions and occupancy-ratio estimates cannot be directly evaluated from held-out transitions. Bellman-residual and minimax validation methods often require auxiliary regressions or optimization over separate critic classes, adding computational and tuning costs. We introduce pairwise Bellman validation (PBV), a tuning-free minimax procedure for selecting among finite classes of fitted Q-functions and occupancy-ratio estimators using held-out transitions. PBV constructs critics from pairwise candidate differences, requiring only finitely many empirical moment calculations without auxiliary validation models or separate critic optimization. We establish finite-sample oracle inequalities relative to the best available candidate, even under candidate-class misspecification. Our key insight is that Q-function errors and log ratios between candidate and true occupancy ratios provide oracle critics whose Bellman and adjoint Bellman moments control estimation error. Pairwise candidate differences approximate these unknown critics, while penalties account for comparator approximation error. For Q-function selection, we first develop unweighted PBV, which evaluates pairwise Bellman moments under the offline distribution and admits oracle guarantees under a pairwise coercivity condition relating the offline distribution to the target-policy dynamics. We then develop target-aligned PBV (TA-PBV), which uses marginalized importance weighting to provide error control without pairwise coercivity. For occupancy-ratio selection, we develop adjoint PBV (A-PBV), which evaluates adjoint Bellman moments against log-ratio critics. On Hopper and Deep OPE, unweighted PBV performs competitively, while target-occupancy weighting can improve performance in settings with controlled overlap.

stat.ML↗
arXiv2026

Adaptive Estimation and Inference in Conditional Moment Models via the Discrepancy Principle

We study adaptive estimation and inference in linear inverse problems defined by conditional moment restrictions. These problems are typically ill-posed, so estimators require regularization, yet the optimal regularization strength depends on a source parameter that is almost never known. We develop a discrepancy principle that chooses this parameter from empirical losses and prescribed bounds on statistical error, without knowing the source parameter. The framework covers Regularized DeepIV (RDIV) and the Tikhonov Regularized Adversarial Estimator (TRAE). Under the stated realizability conditions and comparable fitting and calibration error bounds, both one-step procedures attain strong- and weak-error bounds that match known-source rates for source parameters up to one. We also extend the framework to iterative procedures under realizability of the population iterates, widening the range of source parameters covered by the rate comparison. Finally, we use these adaptive estimators to construct an asymptotically normal doubly robust estimator that adapts to the well-posedness of the primal and dual problems.

stat.ML↗
arXiv2026

Statistical Inference and Learning for Shapley Additive Explanations (SHAP)

The SHAP (short for Shapley additive explanation) framework has become an essential tool for attributing importance to variables in predictive tasks. In model-agnostic settings, SHAP uses the concept of Shapley values from cooperative game theory to fairly allocate credit to the features in a vector $X$ based on their contribution to an outcome $Y$. While the explanations offered by SHAP are local by nature, learners often need global measures of feature importance in order to improve model explainability and perform feature selection. The most common approach for converting these local explanations into global ones is to compute either the mean absolute SHAP or mean squared SHAP. However, despite their ubiquity, there do not exist approaches for performing statistical inference on these quantities. In this paper, we take a semi-parametric approach for calibrating confidence in estimates of the $p$th powers of Shapley additive explanations. We show that, by treating the SHAP curve as a nuisance function that must be estimated from data, one can reliably construct asymptotically normal estimates of the $p$th powers of SHAP. When $p \geq 2$, we show a de-biased estimator that combines U-statistics with Neyman orthogonal scores for functionals of nested regressions is asymptotically normal. When $1 \leq p < 2$ (and the hence target parameter is not twice differentiable), we construct de-biased U-statistics for a smoothed alternative. In particular, we show how to carefully tune the temperature parameter of the smoothing function in order to obtain inference for the true, unsmoothed $p$th power. We complement these results by presenting a Neyman orthogonal loss that can be used to learn the SHAP curve via empirical risk minimization and discussing excess risk guarantees for commonly used function classes.

stat.ML↗
arXiv2026

Switchback Price Experiments with Forward-Looking Demand

Switchback experiments are widely used in pricing applications. We consider a retailer running such an experiment for the price of a single product, with infinite supply. In each period, the seller chooses a price from a set of predefined prices consisting of an original price and a few discounted price levels. The goal is to estimate the demand gradient at the original price to adjust the original price and improve revenue after the experiment. In our model, in each period, a unit mass of buyers arrives on the market, with values distributed based on a time-varying process. Crucially, buyers are forward-looking with a discounted utility and will choose not to purchase now if they expect to face a discounted price in the near future. We show that such forward-looking behavior introduces bias in naive estimators of the demand gradient, due to intertemporal interference. Furthermore, although experiments with only two price points are common in practice, we prove that no estimator using data from such experiments can recover the correct demand gradient, even in the limit of an infinitely long experiment with an infinitesimal price discount. Moreover, we characterize the form of the bias of naive estimators. Finally, we show that with a simple three-price-level experiment, the seller can remove the bias due to strategic forward-looking behavior and construct an estimator for the demand gradient that asymptotically recovers the truth. Our results provide practical guidance for platforms to obtain unbiased demand gradients from price experiments in the presence of forward-looking demand.

cs.GT↗
arXiv2026

Inference on the Value of Policy Improvement in Adaptive Experiments

In this paper, we develop an asymptotically normal estimator for the value of policy improvement (VPI) under adaptively collected data. This value, which is defined as the gap between the value of the best treatment policy in a class and that of a baseline policy, gauges whether or not it is worth deploying a personalized treatment policy. Building an estimator for the VPI is naturally more difficult than estimating the value of a fixed policy. This is because, in addition to contending with the correlated data and decaying propensities associated with adaptive experiments, any estimator must also handle the non-regularity, or lack of pathwise differentiability, of the underlying estimand when the optimal policy is non-unique. Our proposed estimator works by sequentially fitting a weighted empirical welfare maximizing policy on historical data and then evaluating it using inverse propensity weighting on future samples. We neither assume that the optimal policy is unique nor make a margin assumption on the underlying heterogeneous treatment effects. A key difficulty in analyzing our estimator is showing that the regret incurred through policy learning is asymptotically smaller than the standard error of the estimator. We accomplish this by showing faster regret rates for policy learning under decaying propensities, which may be of independent interest. We not only evaluate the coverage of our estimator on a mix of synthetic and semi-synthetic data, but also employ it to estimate the VPI in an adaptive experiment evaluating the efficacy of job assistance programs in Jordan due to Caria et al. Although our estimates are generally more conservative than their hierarchical Bayesian analysis, which assumes that the optimal policy is unique, we still corroborate the benefits of personalization amongst Syrian refugees.

stat.ML↗
arXiv2026

Measuring Gift Card Program Incrementality via Causal Data Fusion

Businesses regularly offer gift card programs to drive customer spending and increase engagement. A central question is how much incremental revenue these programs generate, and which channels drive it most efficiently. Measuring the incremental revenue associated with a gift card program is a challenging problem in causal inference, requiring a firm to infer how much each customer would have spent if they never received a gift card. Observational data on past customer purchasing behavior reveal possession of a gift card only when a customer makes a purchase, thus leaving a customer's treatment status systematically censored. In this paper, we develop a novel data fusion approach to overcome this missing data challenge. We identify and estimate incrementality by combining a large observational dataset with a smaller experimental dataset from a different population. Our approach relies on a mild transferability condition, which posits that the conditional relative treatment effect of gift card receipt on the decision to purchase is invariant across the two populations. We develop a flexible, machine learning-based estimator for the incremental revenue and establish its asymptotic normality. We apply our estimator across both first- and third-party channels through which Airbnb distributes gift cards, finding heterogeneity in incrementality across segments of the population. In particular, we find not only that third-party channels are more incremental than first-party ones, but also that "self-gifters" (i.e., customers likely to have purchased their own gift cards) are more incremental than the broader population.

econ.EM↗
arXiv2026

CausalReasoningBenchmark: A Real-World Benchmark for Disentangled Evaluation of Causal Identification and Estimation

Many benchmarks for automated causal inference evaluate a system's performance based on a single numerical output, such as an Average Treatment Effect (ATE). This approach conflates two distinct steps in causal analysis: identification, formulating a valid research design under stated assumptions, and estimation, implementing that design numerically on finite data. We introduce CausalReasoningBenchmark, a benchmark of 174 queries across 139 real-world datasets, curated from 73 peer-reviewed research papers and three widely used causal-inference textbooks. For each query, a system must produce (i) a structured identification specification that names the strategy, the treatment, outcome, and control variables, and all design-specific elements, and (ii) a point estimate with a standard error. By scoring these two components separately, our benchmark enables granular diagnosis: it distinguishes failures in causal reasoning from errors in numerical execution. Baseline results with a state-of-the-art LLM show that, while the model correctly identifies the high-level strategy in 81% of cases, full identification-specification correctness drops to only 38%, revealing that the bottleneck lies in the nuanced details of research design rather than in computation. CausalReasoningBenchmark is publicly available on Hugging Face and is designed to foster the development of more robust automated causal-inference systems.

cs.AI↗
arXiv2026

Long Story Short: Omitted Variable Bias in Causal Machine Learning

We develop a general theory of omitted variable bias for a wide range of common causal parameters, including (but not limited to) averages of potential outcomes, average treatment effects, average causal derivatives, and policy effects from covariate shifts. Our theory applies to nonparametric models, while naturally allowing for (semi-)parametric restrictions (such as partial linearity) when such assumptions are made. We show how simple plausibility judgments on the maximum explanatory power of omitted variables are sufficient to bound the magnitude of the bias, thus facilitating sensitivity analysis in otherwise complex, nonlinear models. Finally, we provide flexible and efficient statistical inference methods for the bounds, which can leverage modern machine learning algorithms for estimation. These results allow empirical researchers to perform sensitivity analyses in a flexible class of machine-learned causal models using very simple, and interpretable, tools. We demonstrate the utility of our approach with two empirical examples.

econ.EM↗
arXiv2026

CausalSmith: A Formally Grounded, Self-Improving Agentic Framework for Automated Research in Causal Inference

Automating theoretical research requires generating candidate results and evaluating them reliably. Models keep getting better at the first, while the second remains hard. A common approach asks one large language model (LLM) to review what another produced, yet such reviewers are empirically unreliable: they may accept fabricated papers and catch the fabrication at close to chance rates~\citep{badscientist2025}. We present \textsc{CausalSmith}, a framework for automated theoretical research in causal inference built on the Lean proof assistant, where a proof is checked by a program rather than read by a referee. \textsc{CausalSmith} rests on \textsc{Causalean}, a foundational Lean library for causal inference holding 8,179 machine-checked definitions and theorems, developed with language-model assistance under human design and review. Around it, we build a self-improving agentic pipeline that selects research topics, proposes results, formalizes statements, constructs proofs, and presents the resulting artifacts for human inspection. Moreover, the pipeline pairs Lean verification with a statement audit that compares each formal theorem against the informal claim behind it. We evaluate the system using artifacts produced by completed autonomous research runs. The source code, formal library, and run records are available at https://github.com/Jiyuan-Tan/CausalSmith.

stat.ML↗
arXiv2026

Corrupt Plans, Clean Traces: Evading Chain-of-Thought Monitoring with Plan Injection

Chain-of-thought (CoT) monitoring is a safety strategy where the reasoning of a large language model "actor" is inspected by a "monitor" (often another language model) for signs of unsafe planning, deception, or misalignment. We find that planting harmful but benign-sounding reasoning in the actor's context can steer it to perform adversarial actions while evading monitors, an attack we term "plan injection". We initially discover this attack in the multiple-choice question-answering monitorability setting proposed by Lanham et al. (2023), using the investigator-agent elicitation framework of Li et al. (2025). We generalize the attack and show that the discovered behavior scales to harder tasks (achieving 25-33% monitor evasion rates across different monitorability benchmarks) and larger models such as DeepSeek-R1. Across the settings we study, actor models not only follow injected plans but also paraphrase them as their own reasoning, without explicit attribution to the injections. Finally, we find cases where extra monitor resources cause harm - giving the monitor access to the injected plan drops detection by as much as 50% in the Bio-Math task and in a case study on monitor reasoning budget, we find transcripts where additional thinking tokens are spent rationalizing the injected plan rather than flagging it.

cs.AI↗
arXiv2026

Synthetic Blips: Generalizing Synthetic Controls for Dynamic Treatment Effects

We propose a generalization of the synthetic control methods to the setting with dynamic treatment effects, in which each unit receives multiple treatments sequentially, according to an adaptive policy that depends on a latent, endogenously time-varying confounding state. Under a low-rank latent factor model assumption, which admits linear time-varying and time-invariant dynamic triangular systems as special cases, we develop an identification strategy for any unit-specific mean outcome under any sequence of interventions. Our method, which we term synthetic blips, is a backward induction process in which the blip effect of a treatment at each period for a target unit is recursively expressed as a linear combination of the blip effects of other units that received the designated treatment, avoiding the combinatorial donor requirements of naive synthetic control extensions. We provide easy-to-implement estimation algorithms that yield consistent estimators. Using unique Korean firm-level panel data, we estimate individualized dynamic treatment effects and optimal allocation rules in the context of financial support for exporting firms.

econ.EM↗
arXiv2026

Automatic Debiased Machine Learning for Dynamic Treatment Effects and General Nested Functionals

Many canonical models in causal inference and structural econometrics have recursive identification formulas. In causal inference, recursion arises when identification requires both pre- and post-treatment covariates. For example, short-term surrogate outcomes are measured after the treatment, and serve as necessary covariates when identifying long-term effects. Post-treatment covariates are also required for identification of dynamic difference-in-differences designs, time-varying treatment regimes, and mediation analysis. In structural econometrics, recursion arises through evolving state variables, for example in dynamic sample selection models and dynamic discrete choice models. In this paper, we propose an automatic and recursive method for inference, applicable to such formulas, allowing for flexible estimation by neural networks and random forests. As a technical contribution, we introduce recursive Riesz representers.

econ.EM↗
arXiv2026

Order-Explicit Linearization of High-Dimensional $U$-Statistics

We give an order-explicit large deviation bound for the difference between a high-dimensional $U$-statistic and its Hájek projection. In particular, we show that any $U$-statistic of order $b$ on $n$ observations, with a $d$-dimensional kernel whose coordinates have $ψ_1$-Orlicz norm at most $ϕ$, has a maximum deviation from its Hájek projection of order $O_p(ϕb n^{-1}\log^2(dn))$. The proof relies on the development of novel order-explicit moment inequalities for higher-order Hoeffding components. We show that this rate is unimprovable, up to the polynomial factor on the logarithmic term. As corollaries, we obtain new Bernstein-type concentration and Gaussian approximation results for high-dimensional $U$-statistics. We apply these results to establish the consistency of a set of resampling-based simultaneous confidence intervals built around a class of nonparametric regression estimators constructed with subsampled kernels. This class encompasses several forms of random forest regression, including Generalized Random Forests.

econ.EM↗
arXiv2026

Prescriptive Scaling Reveals the Evolution of Language Model Capabilities

Machine learning model performance improvements tend to arise from competition and application. For deployment, we consider prescriptive scaling laws: given a pre-training compute budget, what downstream accuracy is attainable with contemporary post-training practice, and how stable is that mapping as the field evolves? Using large-scale observational evaluations with 5k existing and 2k newly evaluated model checkpoints spanning 2022-2026 across six benchmarks, we estimate capability boundaries, high conditional quantiles of benchmark scores as a function of log pre-training FLOPs, via smoothed quantile regression with a monotone, saturating sigmoid parameterization. We validate temporal reliability by fitting on earlier model generations and evaluating on later releases: across four of six tasks, the out-of-distribution coverage error remains below 2%, while math reasoning exhibits a consistently advancing boundary over time. For instance, at a budget of 10^24 FLOPs, the estimated attainable accuracies are 0.83 on IFEval and 0.54 on MATH Lvl 5. We then extend our approach to analyze task-dependent saturation and to probe contamination-related shifts on math reasoning tasks. Finally, we introduce a balanced I-optimal sampling algorithm that recovers near-full-data frontiers using roughly 20% of the parameter-count-weighted evaluation budget, as low as 5% on some tasks, while maintaining comparable calibration. Together, our work releases Proteus-2k, the latest model performance evaluation dataset, and introduces a practical methodology for translating compute budgets into reliable performance expectations and for monitoring when capability boundaries shift across time.

cs.LG↗
arXiv2026

The Partial Testimony of Logs: Evaluation of Language Model Generation under Confounded Model Choice

Offline evaluation of language models from usage logs is biased when model choice is confounded: the same user-side factors that influence which model is used can also influence how its output is judged, so raw comparisons of logged scores mix self-selected populations rather than estimating a common quantity of interest. A small randomized experiment can break this bias by overriding model choice, but in practice such experiments are scarce and costly. We study a three-source design that combines a large confounded observational log (OBS) for scale, a small randomized experiment (EXP) for unconfounded scoring, and an offline simulator (SIM) that replays candidate models on cached contexts. Our main result is an identification theorem showing that the randomized experiment and the simulator are together enough to recover causal model values; the observational log enters only afterward, to reduce estimation error rather than to make the causal comparison valid. Six estimator families are evaluated in a controlled semi-synthetic validation and in two real-task cached benchmarks for summarization and coding. No family dominates every regime; relative performance depends on the amount of unbiased EXP supervision and on how closely the target reward aligns with OBS-derived structure.

cs.LG↗