I’m a PhD student in Statistics at Carnegie Mellon University where I am lucky to be advised by Aaditya Ramdas. My research is supported by an Amazon Graduate Research Fellowship.

Previously, I was an intern at Microsoft Research, Adobe Research, and SickKids. Before that, I studied math at the University of Waterloo in Canada.

I am broadly interested in statistical methods that make realistic nonparametric assumptions about data, often in adaptive, sequential settings. More specifically, I focus on

• Anytime-valid sequential inference,
• Nonparametric methods,
• Causal inference,
• Reinforcement learning, and
• Differential privacy.

---

Selected papers

(Full list on google scholar.)

• Anytime-valid off-policy inference for contextual bandits
  I. Waudby-Smith, L. Wu, A. Ramdas, N. Karampatziakis, and P. Mineiro
  arxiv

• Time-uniform central limit theory and asymptotic confidence sequences
  I. Waudby-Smith, D. Arbour, R. Sinha, E.H. Kennedy, and A. Ramdas
  arxiv

• A nonparametric extension of randomized response for private confidence sets
  I. Waudby-Smith, Z.S. Wu, and A. Ramdas
  ICML, 2023 (oral) · arxiv

• Estimating means of bounded random variables by betting
  I. Waudby-Smith and A. Ramdas
  JRSSB, 2023 (discussion paper) · arxiv

• RiLACS: Risk-limiting audits via confidence sequences
  I. Waudby-Smith, P.B. Stark, and A. Ramdas
  E-Vote-ID, 2021 (Best paper award) · arxiv · proceedings

• Confidence sequences for sampling without replacement
  I. Waudby-Smith and A. Ramdas
  NeurIPS, 2020 (spotlight) · arxiv · proceedings

---

Software

Below are some software packages that I maintain.

• confseq: Confidence sequences and uniform boundaries
  This Python package (maintained by Steve Howard and me) implements several nonasymptotic confidence sequences, including those from Howard et al. (2021), Howard & Ramdas (2022), Waudby-Smith & Ramdas (2020) and Waudby-Smith & Ramdas (2023).
  [github · pypi]

• drconfseq: Doubly robust confidence sequences for sequential causal inference
  This R package implements the asymptotic confidence sequences of Waudby-Smith et al. and specifically confidence sequences for doubly robust treatment effect estimation in observational studies and randomized experiments.
  [github]

• RiLACS: Risk-limiting election audits via confidence sequences
  This Python package implements methods introduced in Waudby-Smith, Stark, and Ramdas (2021) for using confidence sequences and betting martingale-based tests to audit elections, and more specifically, any election that can be reduced to mean testing/estimation via SHANGRLA (Stark, 2020).
  [github · pypi]

• NPRR: Nonparametric randomized response
  This Python package implements our nonparametric generalization of Warner’s randomized response as well as methods for computing differentially private confidence intervals and sequences for means of bounded random variables as in Waudby-Smith, Wu, and Ramdas (2023).
  [github · pypi]

Methods that I have developed have also appeared in some other open-source projects (by people that write much better code than me) such as Microsoft’s VowpalWabbit (see this commit) or GrowthBook’s sequential testing implementation (see this commit or the docs) but I have not contributed to them directly.

---

Connect

• Email: ianws@cmu.edu
• Github: wannabesmith
• Twitter: @iwaudbysmith

---