ROBBIE: Robust Bias Evaluation of Large Generative Language Models

As generative large language models (LLMs) grow more performant and prevalent, we must develop comprehensive enough tools to measure and improve their fairness. Different prompt-based datasets can be used to measure social bias across multiple text domains and demographic axes, meaning that testing LLMs on more datasets can potentially help us characterize their biases more fully, and better ensure equal and equitable treatment of marginalized demographic groups. In this work, our focus is two-fold: (1) Benchmarking: a comparison of 6 different prompt-based bias and toxicity metrics across 12 demographic axes and 5 families of generative LLMs. Out of those 6 metrics, AdvPromptSet and HolisticBiasR are novel datasets proposed in the paper. The comparison of those benchmarks gives us insights about the bias and toxicity of the compared models. Therefore, we explore the frequency of demographic terms in common LLM pre-training corpora and how this may relate to model biases. (2) Mitigation: we conduct a comprehensive study of how well 3 bias/toxicity mitigation techniques perform across our suite of measurements. ROBBIE aims to provide insights for practitioners while deploying a model, emphasizing the need to not only measure potential harms, but also understand how they arise by characterizing the data, mitigate harms once found, and balance any trade-offs. We open-source our analysis code in hopes of encouraging broader measurements of bias in future LLMs.

Causal Inference under Data Restrictions

This dissertation focuses on modern causal inference under uncertainty and data restrictions, with applications to neoadjuvant clinical trials, distributed data networks, and robust individualized decision making. In the first project, we propose a method under the principal stratification framework to identify and estimate the average treatment effects on a binary outcome, conditional on the counterfactual status of a post-treatment intermediate response. Under mild assumptions, the treatment effect of interest can be identified. We extend the approach to address censored outcome data. The proposed method is applied to a neoadjuvant clinical trial and its performance is evaluated via simulation studies. In the second project, we propose a tree-based model averaging approach to improve the estimation accuracy of conditional average treatment effects at a target site by leveraging models derived from other potentially heterogeneous sites, without them sharing subject-level data. The performance of this approach is demonstrated by a study of the causal effects of oxygen therapy on hospital survival rates and backed up by comprehensive simulations. In the third project, we propose a robust individualized decision learning framework with sensitive variables to improve the worst-case outcomes of individuals caused by sensitive variables that are unavailable at the time of decision. Unlike most existing work that uses mean-optimal objectives, we propose a robust learning framework by finding a newly defined quantile- or infimum-optimal decision rule. From a causal perspective, we also generalize the classic notion of (average) fairness to conditional fairness for individual subjects. The reliable performance of the proposed method is demonstrated through synthetic experiments and three real-data applications.

RISE: Robust Individualized Decision Learning with Sensitive Variables

This paper introduces RISE, a robust individualized decision learning framework with sensitive variables, where sensitive variables are collectible data and important to the intervention decision, but their inclusion in decision making is prohibited due to reasons such as delayed availability or fairness concerns. A naive baseline is to ignore these sensitive variables in learning decision rules, leading to significant uncertainty and bias. To address this, we propose a decision learning framework to incorporate sensitive variables during offline training but not include them in the input of the learned decision rule during model deployment. Specifically, from a causal perspective, the proposed framework intends to improve the worst-case outcomes of individuals caused by sensitive variables that are unavailable at the time of decision. Unlike most existing literature that uses mean-optimal objectives, we propose a robust learning framework by finding a newly defined quantile- or infimum-optimal decision rule. The reliable performance of the proposed method is demonstrated through synthetic experiments and three real-world applications.

When Doubly Robust Methods Meet Machine Learning for Estimating Treatment Effects from Real-World Data: A Comparative Study

Observational cohort studies are increasingly being used for comparative effectiveness research and to assess the safety of therapeutics. Recently, various doubly robust methods have been proposed for average treatment effect estimation by combining the treatment model and the outcome model via different vehicles, such as matching, weighting, and regression. The key advantage of the doubly robust estimators is that they require either the treatment model or the outcome model to be correctly specified to obtain a consistent estimator of the average treatment effect, and therefore lead to a more accurate and often more precise inference. However, little work has been done to understand how doubly robust estimators differ due to their unique strategies of using the treatment and outcome models and how machine learning techniques can be combined with these estimators to boost their performance. Also, little has been understood about the challenges of covariates selection, overlapping of the covariate distribution, and treatment effect heterogeneity on the performance of these doubly robust estimators. Here we examine multiple popular doubly robust methods in the categories of matching, weighting, or regression, and compare their performance using different treatment and outcome modeling via extensive simulations and a real-world application. We found that incorporating machine learning with doubly robust estimators such as the targeted maximum likelihood estimator outperforms. Practical guidance on how to apply doubly robust estimators is provided.

Point3D: tracking actions as moving points with 3D CNNs

Spatio-temporal action recognition has been a challenging task that involves detecting where and when actions occur. Current state-of-the-art action detectors are mostly anchor-based, requiring sensitive anchor designs and huge computations due to calculating large numbers of anchor boxes. Motivated by nascent anchor-free approaches, we propose Point3D, a flexible and computationally efficient network with high precision for spatio-temporal action recognition. Our Point3D consists of a Point Head for action localization and a 3D Head for action classification. Firstly, Point Head is used to track center points and knot key points of humans to localize the bounding box of an action. These location features are then piped into a time-wise attention to learn long-range dependencies across frames. The 3D Head is later deployed for the final action classification. Our Point3D achieves state-of-the-art performance on the JHMDB, UCF101-24, and AVA benchmarks in terms of frame-mAP and video-mAP. Comprehensive ablation studies also demonstrate the effectiveness of each module proposed in our Point3D.

A Tree-based Federated Learning Approach for Personalized Treatment Effect Estimation from Heterogeneous Data Sources

Federated learning is an appealing framework for analyzing sensitive data from distributed health data networks due to its protection of data privacy. Under this framework, data partners at local sites collaboratively build an analytical model under the orchestration of a coordinating site, while keeping the data decentralized. However, existing federated learning methods mainly assume data across sites are homogeneous samples of the global population, hence failing to properly account for the extra variability across sites in estimation and inference. Drawing on a multi-hospital electronic health records network, we develop an efficient and interpretable tree-based ensemble of personalized treatment effect estimators to join results across hospital sites, while actively modeling for the heterogeneity in data sources through site partitioning. The efficiency of our method is demonstrated by a study of causal effects of oxygen saturation on hospital mortality and backed up by comprehensive numerical results.

Towards Improving Spatiotemporal Action Recognition in Videos

Spatiotemporal action recognition deals with locating and classifying actions in videos. Motivated by the latest state-of-the-art real-time object detector You Only Watch Once (YOWO), we aim to modify its structure to increase action detection precision and reduce computational time. Specifically, we propose four novel approaches in attempts to improve YOWO and address the imbalanced class issue in videos by modifying the loss function. We consider two moderate-sized datasets to apply our modification of YOWO - the popular Joint-annotated Human Motion Data Base (J-HMDB-21) and a private dataset of restaurant video footage provided by a Carnegie Mellon University-based startup, Agot.AI. The latter involves fast-moving actions with small objects as well as unbalanced data classes, making the task of action localization more challenging. We implement our proposed methods in the GitHub repository https://github.com/stoneMo/YOWOv2.