Dimensions overall score 7.0

GitHub Code Pulse

No public code linked for this paper yet.

• Self-Evolving AI Can Wang 1, Hongyu Zhao 1, Yiqun Chen 1,2 1Department of Biostatistics, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD 21205, USA 2Department of Computer Science, Johns Hopkins University, Baltimore, MD 21218, USA Causal inference is central to scientific discoveryclaim 72%
• Causal Assumptions Ignorability, overlap, SUTVA, ... ... ... ... ... ... Input: Causal Taskclaim 72%
• New Parent Program New Program • Evolutionary Trace • Best Evolved Program • Benchmarked Performance Evaluation Score = f(metric 1, metric 2, ...) Stage 1: Fail Fast Stage 2: Batch Evaluation • ATE bias • RMSE/PEHE • Coverage • ... Feedback Evaluate a 20 40 60 80ACIC 2022 RMSE Zero-shot Evolved Human 0.4 0.6 0.8 1claim 72%
• LaLonde Baseline Within-dataset progression Early Mid Late Program Embedding Space Across Evolution 1 25 50 75 Evolution iteration 0.0 0.5 1.0 Fraction of final gain Figure 1 Overview of InferenceEvolve for causal estimator discovery. a , Schematic of the workflow: a causal task specification seeds a zero-shot programclaim 72%
• ended LLM-agent framework for causal estimator discovery across four diverse benchmarks: ACIC 2022 ( Thal and Finucane, 2023), ACIC 2016 ( Dorie et al. , 2019), IHDP ( Hill, 2011), and LaLonde ( LaLonde, 1986; Dehe- jia and Wahba, 1999), which span panel data, cross-sectional, semi-syntheticclaim 72%
• vuut1 N NX i=1 (ˆτi −τi)2 = ∥ˆτ −τ ∥2√ N , (1) and the absolute ATE error |ATE error|= 1 N NX i=1 ˆτi − 1 N NX i=1 τi . (2) In words, PEHE emphasizes the more granular unit-level heterogeneityclaim 72%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card