Statistical Foundations of Latent Structure and Representation Learning

November 4–5, 2026

Department of Statistics, Columbia University New York City

Registration and poster submissions are closed.

About

This two-day workshop brings together researchers in statistics, machine learning, and data science to explore the statistical foundations of latent structure and representation learning, including identifiability, statistical efficiency, uncertainty quantification, computation, causality, and interpretability.

Program

Program book and talk abstracts (PDF)  ·  Poster program (PDF)

Morning talks are in SSW C06 and afternoon talks are in SSW 903. Jump to November 4 or November 5.

Wednesday, November 4

8:30 a.m.Breakfast
8:55–9:00 a.m.Opening remarks: Yuqi Gu
9:00–9:35 a.m.Bin Yu · UC Berkeley
Not All Interpretations Are Created Equal: Principled Evaluation of AI Interpretability
9:35–10:10 a.m.Carey Priebe · Johns Hopkins University
Performance guarantees for LLMs
10:10–10:45 a.m.Yuxin Chen · University of Pennsylvania
Transformers Meet In-Context Learning: A Universal Approximation Theory
10:45–11:10 a.m.Coffee break
11:10–11:45 a.m.Jianqing Fan · Princeton University
Factor Informed Double Deep Learning For Average Treatment Effect Estimation
11:45 a.m.–12:20 p.m.Florentina Bunea · Cornell University
Softmax-Structured Gaussian Mixtures for Distribution Ensembles
12:20–1:40 p.m.Lunch
1:40–2:15 p.m.Lorenzo Mauri · Duke University
Scalable Bayesian Learning of Latent Structure: Integrating Data and Biological Knowledge
2:15–2:50 p.m.Hongtu Zhu · UNC Chapel Hill
FAST-Brain: A Flow-Aligned Spatio-Temporal Surrogate Brain Model
2:50–3:15 p.m.Coffee break
3:15–3:50 p.m.Kun Zhang · Carnegie Mellon University
Causal representation learning and causal generative AI
3:50–4:25 p.m.Yihong Gu · Harvard University
Unveiling Invariant and Transferable Latent Factors Across Heterogeneous Environments via ATLAS
4:25–4:45 p.m.Transition to poster session
4:45–6:15 p.m.Early-career researcher poster session · SSW 214

Thursday, November 5

8:30 a.m.Breakfast
9:00–9:35 a.m.Annie Qu · UC Santa Barbara
Meta Fusion: A Unified Framework For Multimodal Fusion with Adaptive Mutual Learning
9:35–10:10 a.m.Tracy Ke · Harvard University
Topic-aware Ranking with Text-estimated Context
10:10–10:40 a.m.Coffee break
10:40–11:15 a.m.Jiashun Jin · Carnegie Mellon University
Estimating weak spikes with Newton and Prony
11:15–11:50 a.m.George Michailidis · UCLA
Learning Latent Representations and Structured Dynamics in High-Dimensional Time Series
11:50 a.m.–1:40 p.m.Lunch
1:40–2:15 p.m.Hui Zou · Johns Hopkins University
Foe or Ally: The Role of Dimensionality in Prediction with Measurement Errors
2:15–2:50 p.m.Joseph Feldman · Washington University in St. Louis
Scalable Independent Factor Analyzers using Spectral Initialization for Evaluating Large Language Models
2:50–3:25 p.m.Gongjun Xu · University of Michigan
Beyond Vintage Rotation: Bias-Free Sparse Representation Learning with Oracle Inference
3:25 p.m.Closing remarks, followed by coffee

Location

School of Social Work building, Columbia University.

1255 Amsterdam Avenue, New York, NY 10027. All workshop sessions will take place in this building.

SessionDate and timeRoom
Morning talksNovember 4–5, starting at 9:00 a.m.C06
Afternoon talksNovember 4–5903
Poster sessionNovember 4, 4:45–6:15 p.m.214

Speakers

Speakers are listed alphabetically by last name.

Early-Career Researcher Poster Session

The early-career researcher poster session features work on latent structure, representation learning, and related areas, providing an opportunity to exchange ideas with workshop speakers and participants.

Poster program (PDF)

Organizer

Yuqi Gu
Department of Statistics, Columbia University

Questions about the workshop: yuqi.gu@columbia.edu