Qi Ma

Jakub Grudzien Kuba, Mengting Gu, Qi Ma et al.

Large language models (LLMs) have advanced rapidly in recent years, driven by scale, abundant high-quality training data, and reinforcement learning. Yet this progress faces a fundamental bottleneck: the need for ever more data from which models can continue to learn. In this work, we propose a reinforcement learning approach that removes this dependency by enabling models to improve without additional data. Our method leverages a game-theoretic framework of self-play, where a model's capabilities are cast as performance in a competitive game and stronger policies emerge by having the model play against itself - a process we call Language Self-Play (LSP). Experiments with Llama-3.2-3B-Instruct on instruction-following benchmarks show that pretrained models can not only enhance their performance on challenging tasks through self-play alone, but can also do so more effectively than data-driven baselines.

Steven Wilkins-Reeves, Xu Chen, Qi Ma et al.

Distribution shifts are ubiquitous in real-world machine learning applications, posing a challenge to the generalization of models trained on one data distribution to another. We focus on scenarios where data distributions vary across multiple segments of the entire population and only make local assumptions about the differences between training and test (deployment) distributions within each segment. We propose a two-stage multiply robust estimation method to improve model performance on each individual segment for tabular data analysis. The method involves fitting a linear combination of the based models, learned using clusters of training data from multiple segments, followed by a refinement step for each segment. Our method is designed to be implemented with commonly used off-the-shelf machine learning models. We establish theoretical guarantees on the generalization bound of the method on the test risk. With extensive experiments on synthetic and real datasets, we demonstrate that the proposed method substantially improves over existing alternatives in prediction accuracy and robustness on both regression and classification tasks. We also assess its effectiveness on a user city prediction dataset from Meta.

Qi Ma, Sujit K. Ghosh

The presence of missing values within high-dimensional data is an ubiquitous problem for many applied sciences. A serious limitation of many available data mining and machine learning methods is their inability to handle partially missing values and so an integrated approach that combines imputation and model estimation is vital for down-stream analysis. A computationally fast algorithm, called EMFlow, is introduced that performs imputation in a latent space via an online version of Expectation-Maximization (EM) algorithm by using a normalizing flow (NF) model which maps the data space to a latent space. The proposed EMFlow algorithm is iterative, involving updating the parameters of online EM and NF alternatively. Extensive experimental results for high-dimensional multivariate and image datasets are presented to illustrate the superior performance of the EMFlow compared to a couple of recently available methods in terms of both predictive accuracy and speed of algorithmic convergence. We provide code for all our experiments.