Yuzhe Gu

I am a second-year MS student pursuing dual degrees in Computer and Information Science and Electrical and Systems Engineering at the University of Pennsylvania, with a concentration on Artificial Intelligence and Machine Learning.

I received my dual Bachelor's degrees in Data Science from Duke Kunshan University and Duke University, where I worked closely with Enmao Diao and Peng Sun.

Email  /  GitHub  /  Google Scholar  /  LinkedIn

Research

My ongoing research interest focuses on improving algorithm efficiency and effectiveness in deep learning problems, particularly in the context of modern generative models for vision and language. Previously, I also worked on deep learning-driven audio processing, including applications in compression and speech synthesis.

ESC: Efficient Speech Coding with Cross-Scale Residual Vector Quantized Transformers

We present ESC, a parameter-efficient speech foundation codec with cross-scale vector-quantized transformer architectures, which achieves coding performance comparable to state-of-the-art models while reducing decoding latency by 6.4x.

How Did We Get Here? Summarizing Conversation Dynamics

We introduce the task of summarizing the dynamics of conversations (SCD) by constructing a dataset of human-written summaries and exploring several automated baselines. We demonstrate that SCDs assist both humans and automated systems in forecasting whether an ongoing conversation will eventually derail into toxic behavior.

Towards Quantification of Covid-19 Intervention Policies from Machine Learning-based Time Series Forecasting Approaches

We develop a policy-aware epidemic time-series predictive model, and perform causal analysis to quantify the effects of governmental interventions during COVID-19 through counterfactual estimation.

Projects

Visual Autoregressive Language Models

An implementation of autoregressive vision generative models with discrete latent variables. Reconstruction and class-conditioned synthesis results are reproduced on StanfordDogs, a dataset containing 120 dog breeds.

Robustness Evaluation on LLM Bias and Fairness Metrics

A re-evaluation on the robustness of text continuation-based bias metrics for quantifying group fairness in LLMs. Empirical results indicate that existing approaches are sensitive to language model’s inherent non-determinism from decoding setups.

Design and source code from Jon Barron's website