Yuzhe Gu
A Graduate Student interested in Machine Learning and Natural Language Processing
![yuzhegu_headshot.jpg](/assets/img/yuzhegu_headshot.jpg?2776a085df0af50932352b0d6ed9a15a)
I am a first year master student in Department of Electrical and Systems Engineering at the University of Pennsylvania with a concentration on Machine Learning and Data Science. I received my dual B.Sc. in Data Science from Duke Kunshan University and Duke University, where I worked closely with Peng Sun and Enmao Diao.
My research interest covers a wide range of topics related to representation learning and generative models:
- Discrete Representation Learning: vector quantization networks for generative modeling, data compression, and multi-modality learning
- Efficient LLMs: leveraging quantization techniques for efficient LLM inference/finetuning
- Trustworthy LLMs: re-examination of bias, fairness and robustness in LLMs
I am currently seeking for a CS/ECE Ph.D. position related to Machine Learning and Natural Language Processing starting 2025 Fall.
Email: tracygu (at) seas (dot) upenn (dot) edu Google Scholar: @scholar |
Github: @github/yzGuu830 Linkedin: @in/yuzheguu |
Publications
![]() | ESC: Efficient Speech Coding with Cross-Scale Residual Vector Quantized Transformers Yuzhe Gu, Enmao Diao Under Review paper / code We propose Efficient Speech Codec (ESC), a lightweight, parameter-efficient speech codec based on a cross-scale residual vector quantization scheme and transformers. Our model employs mirrored hierarchical window transformer blocks and performs step-wise decoding from coarse-to-fine feature representations. ESC can achieve high-fidelity speech reconstruction with significantly lower complexity than state-of-the-art convolutional codecs. |
![]() | How Did We Get Here? Summarizing Conversation Dynamics Yilun Hua, Nicholas Chernogor, Yuzhe Gu, Seoyeon Julie Jeong, Miranda Luo, Cristian Danescu-Niculescu-Mizil Proceedings of NAACL, 2024 paper / code We introduce the task of summarizing the dynamics of conversations, by constructing a dataset of human-written summaries, and exploring several automated baselines. We evaluate whether such summaries can capture the trajectory of conversations via an established downstream task: forecasting whether an ongoing conversation will eventually derail into toxic behavior. We show that they help both humans and automated systems with this forecasting task. |
![]() | Towards Quantification of Covid-19 Intervention Policies from Machine Learning-based Time Series Forecasting Approaches Yuzhe Gu, Peng Sun, Azzedine Boukerche Proceedings of IEEE International Conference on Communications (ICC), 2024 paper / code We design a policy-aware time series forecasting model to estimate COVID-19 trends by incorporating temporal information from 16 policy indicators. Through counterfactual analysis, we quantify the causal effect of indicators and propose two static metrics lag period and average effect. Our model verifies the effectiveness of all 16 policy indicators in controlling virus transmission in the US. |