Date: Wednesday, November 5, 2025
Time: 12:00 – 13:00 PM CT
Speaker: Associate Professor Dr. Jing Yang, University of Virginia
Abstract: Large language models (LLMs) have demonstrated remarkable in-context learning (ICL) capabilities, enabling them to perform new tasks from only a few labeled examples provided within the prompt. However, the predictive quality of ICL is inherently limited by the scarcity of such labeled demonstrations. At the same time, large quantities of unlabeled data, often highly relevant to the ICL task, remain untapped. This raises a fundamental question: how can unlabeled data be leveraged to provably enhance ICL performance? In this talk, we introduce an augmented ICL framework that enriches the prompt with both a small labeled subset and a block of unlabeled inputs. Focusing on multi-class linear classification, we show that when equipped with chain-of-thought (CoT) prompting, a multi-layer transformer can emulate the expectation–maximization (EM) algorithm, thereby exploiting both labeled and unlabeled information within the prompt. This yields formal guarantees of improved ICL accuracy. Furthermore, we demonstrate that such transformers can be efficiently trained via teacher forcing, achieving convergence to the target solution at a sub-linear rate. We further instantiate this framework in the domain of wireless communications. Across diverse modulation schemes and signal-to-noise regimes, our method consistently achieves lower symbol error rates than classical semi-supervised and prior ICL baselines. These results highlight the potential of augmented ICL as a general principle for integrating unlabeled data into transformer-based inference.
Bio: Jing Yang is an Associate Professor in the Department of Electrical and Computer Engineering at the University of Virginia, with a courtesy appointment in the Department of Computer Science. Previously, she was an Assistant and then tenured Associate Professor at the Pennsylvania State University. She received her B.S. from the University of Science and Technology of China (USTC), and her M.S. and PhD from the University of Maryland, College Park, all in Electrical Engineering. She is a recipient of the National Science Foundation CAREER Award and the WICE Early Achievement Award, and was recognized as one of the 2020 N2Women: Stars in Computer Networking and Communications. She served as a Symposium/Track/Workshop Co-chair for Asilomar 2023, ICC 2021, INFOCOM 2021-AoI Workshop, WCSP 2019, CTW 2015, PIMRC 2014, and an Editor for IEEE Trans. on Wireless Communications and IEEE Trans. on Green Communications and Networking. She is currently an Area Editor for IEEE Trans. on Green Communications and Networking and an Associate Editor for IEEE Trans. on Cognitive Communications and Networking. Her recent research interests include transformers and large language models, reinforcement learning, privacy-preserving machine learning, and their applications in wireless communications and networking.