Wednesday, May 6, 20269:30 am - 1:00 pm
The Columbia University Data Science Institute’s Foundations of Data Science Center is hosting a workshop designed to foster collaboration and knowledge sharing. Through talks and posters, Columbia researchers will showcase their work in the diverse realms of data science methods and applications.
Location: Columbia Engineering Innovation HubAddress: 2276 12th Ave, New York, NY 10027 – Manhattanville
· · ─ · ─ · ·
9:30 AM – 10:00 AM: Breakfast & Check-In (30 min)
10:00 AM – 10:40 AM: Presentation: Genevera Allen, Professor of Statistics (40 min)
10:40 AM – 10:50 AM: Break (10 min)
10:50 AM – 11:30 AM: Presentation: Kaizheng Wang, Assistant Professor, IEOR (40 min)
11:30 AM – 11:40 AM: Break (10 min)
11:40 AM – 12:00 PM: Poster Spotlights
12:00 PM – 1:00 PM: Lunch and Poster Session
Genevera AllenProfessor of Statistics, Columbia University
Talk Title: Inference for Clustering: Conformal Sets for Cluster Labels
Abstract: While clustering is ubiquitously used across science and industry, uncertainty in cluster assignments is rarely quantified with rigorous guarantees. We propose a novel conformal inference framework for clustering that returns confidence sets for cluster labels. The key challenge is that labels are unobserved and estimated from data, so naively using deterministic cluster labels can violate exchangeability and induce severe under-coverage. To address this, we propose split conformal clustering with stochastic labels, which samples labels from soft cluster labels, fits a soft classifier to predict these stochastic labels, and calibrates conformal scores to construct confidence sets for cluster labels at any query point. We establish a finite-sample lower bound on marginal coverage that reveals how under-coverage is controlled by two properties of the clustering algorithm: consistency of estimated soft labels and replace-one stability. Under mild conditions, we prove asymptotic coverage and verify these conditions for correctly specified parametric mixture models. Simulations for mixture models show that our method attains target coverage with informative set sizes, validating our theoretical results. Applications to clustering cell types in single-cell RNA-seq data demonstrate the practical utility and interpretability of our approach to quantifying cluster label uncertainty.
Kaizheng WangAssistant Professor of Industrial Engineering and Operations Research, Columbia Engineering
Talk Title: AI Personas for Human-Centered Inference and Decision-Making
Abstract: AI personas provide a flexible framework for modeling human behavior, particularly in cold-start settings with limited data on new users or scenarios. In this talk, I will present recent and ongoing work on AI-powered human digital twins from this perspective. I will begin with two decision-oriented applications: adaptive querying with persona priors, which enables efficient information acquisition under limited question budgets, and an LLM-based demand simulator for pricing, which combines rich product information with persona-level purchase probabilities for counterfactual policy evaluation. I will then discuss a calibration framework for improving simulator fidelity to human data, and conclude with a statistical framework for uncertainty quantification in LLM-based survey simulation, including valid inference under unknown distribution shift between synthetic and real data. Together, these works show how AI personas can support principled human-centered inference and decision-making.
P01: A Pipeline for Enabling Path-Specific Causal Fairness in Observational Health Data
P02: Imperfect Influence, Preserved Rankings: A Theory of TRAK for Data Attribution
P03: Model-Free Assessment of Simulator Fidelity via Quantile Curves
P04: TabImpute: Universal Zero-Shot Imputation for Tabular Data
P05: On the Gradient Domination of LQG Problem
P06: Pseudo-Labeling for Unsupervised Domain Adaptation with Kernel GLMs
P07: SpeechScreen: Generalizability of Cognitive Impairment Detection Across Languages
P08: Explainable Asset Allocation and Portfolio Construction
P09: Buckle Up! Seatbelt: An Open-Source Python Library for Responsible AI Auditing of Generative AI Models
P10: Unsupervised Generative Framework for Event-Driven Financial Time Series
P11: Generalizing Risk Parity via Optimal Risk Budgeting with Target Returns: Exact and Tight Approximation Algorithms
P12: Evaluating LLM-Persona Generated Distributions for Decision-Making
P13: ICYM²I: The illusion of Multimodal Informativeness Under Missingness
P14: SYN-DIGITS: A Synthetic Control Framework for Calibrated Digital Twin Simulation
P15: GOPO: Policy Optimization Using Ranked Rewards
P16: Learning Entity-Conditioned Topic Representations in Embedding Space