CV

Download full CV here

Education

• M.S. in Computer Science, University of California, Los Angeles, advised by Prof. Cho-Jui Hsieh
• B.S. in Computer Science, National Yang Ming Chiao Tung University (previous National Chiao Tung University)

Work Experience

Himax Imaging

AI engineer Intern

• Investigated research and employed model compression techniques such as structural pruning and quantization to a unified face detection, head pose estimation, and face emotion recognition model on Himax AI chips running on Windows notebook, speeding up 20% of FPS and 3x times fewer FLOPs.

Appier

Machine Learning Scientist

• Improved User Lookalike models to produce the distinguished user score for each unique user id based on existing client site activities and deployed models online by using CI/CD pipelines, enhancing 20% improvement on AUROC compared to baseline.
• Extracted user behavior patterns from 1000K+ conversion funnel data and analyzed the Click-Through Rate of different campaigns using PySpark, SQL and Pandas, resulting in 120% revenue growth within 3 months.
• Decreased the uncertainty for outlier and lose bidding data in bidding models, saving up 30% of the trouble shooting time.

Umbo CV

AI engineer Intern

• Investigated research and employed model compression techniques such as structural pruning and quantization to person re-identification model on real-time streaming cameras, speeding up 20% of FPS and 3x times fewer FLOPs.

Cinnamon AI

AI Bootcamp Summer Intern

• Implemented a Seq2Seq-based model to recommend tourist attractions based on personal interest and arrange suitable trip routes deployed on Gradio to make a fast user interface.

Research Experience

• University of California, Los Angeles, Research Assistant
  • Supervisor: Prof. Cho-Jui Hsieh
  • Proposed an auto-prompting approach that combines LLMs and external symbolic solver to solve Algebra Word Problem by utilizing different prompting strategies, achieving 10% improvement in answer accuracy on both English and Chinese datasets.
  • Curated a new and larger algebraic dataset with prompt optimization which contains multiple variables questions to evaluate our proposed reasoning approaches on solving more challengeable Algebra Word Problem.
• Stanford University with National Yang Ming Chiao Tung University, Research Assistant
  • Supervisor: Prof. Yung-Ju, Chang & Stanford Screenomics Lab
  • Investigated the use of Vision Language Models (VLMs), particularly GPT-4V, in the analysis of smartphone user activities from screenshots, offering an alternative to traditional app usage data analysis limitations, providing highly convinced reliability score compared to human coders.
  • Conducted a two-week user study via users’ self-reported experience as evidence to support “killing-time” periods are opportune for user engagement with notifications that demand user attention and engagement.
• Institute of Information Science Academia Sinica, Research Assistant
  • Supervisor: Prof. Keh-Yi, Su
  • Utilized deep learning methods and conducted experiments to Knowledge-Guided Algebra Word Problem Solver, achieving better equation accuracy and problem accuracy on English algebraic datasets.
  • Designed and built a two-stage neural model, which adopts the concepts of the solving strategies by humans, generating multiple expression trees explicitly and representing the reasonable solving process behind the model’s solution equation.
• National Yang Ming Chiao Tung University, Research Assistant
  • Supervisor: Prof. Yung-Ju, Chang & Prof. Wei-Chen Chiu
  • Leveraged deep learning fusion model to investigate users’ kill time behavior based on 1000K+ mobile phone-sensor and screenshot data, which is collected by our developed Android App.
  • Employed a two-stage clustering approach to separate users into four groups according to the patterns of their phone-usage behaviors, and then built a fusion model for each group, yielding overall strong performance on AUROC.
• National Yang Ming Chiao Tung University, Research Student
  • Supervisor: Prof. Ching-Chun Huang
  • Applied model compression using structural pruning and knowledge distillation on YOLOv4. The developed models not only fit for embedded systems (Ex: Jetson TX2) but also achieve higher FPS and mAP at the same time on the multi-spectral infrared dataset.
  • Winner of the award: “2021 ACM ICMR Embedded Deep Learning Object Detection Model Compression Competition for Traffic in Asian Countries” – Final Round.

Skills

• Programming: C/C++, Python (Package: PyTorch, Tensorflow, PySpark, HuggingFace), SQL, Shell Script, HTML, CSS, MATLAB, Verilog
• DevOps & Tools: GCP, Docker, Kubernetes, Git, Jenkins, CI/CD, Airflow, System and Network Administration, Grafana, LATEX

Publications

• Yung-Ju Chang, Yu-Chun Chen, Kuei-Chun Kao, Yu-Jen Lee, Mu-Jung Cho, Yikun Chi, Thomas N. Robinson, Byron Reeves, Nilam Ram “Assessing the Reliability of Vision Language Models for Inferring Phone Activity from Smartphone Screenshots: A Preliminary Case Study with VLM”, (preprint, * indicates equal contribution)
• Kuei-Chun Kao, Ruochen Wang, Cho-Jui Hsieh “Solving for X and Beyond: Can Large Language Models Solve Complex Math Problems with More-Than-Two Unknowns?”, (preprint)
• Yu-Chun Chen, Kuei-Chun Kao, Yu-Jen Lee*, Yung-Ju Chang “Does Receiving Less Personally Relevant but Attention-demanding Notifications while ‘Killing Time’ Increase Engagement? An Exploratory Study”, (preprint, * indicates equal contribution)
• Kuei-Chun Kao, Chao-Chun Liang, Keh-Yih Su “Knowledge-Guided Algebra Word Problem Solving” (preprint)
• Yu-Chun Chen, Yu-Jen Lee, Kuei-Chun Kao, Jie Tsai, En-Chi Liang, Wei-Chen Chiu, Faye Shih, Yung-Ju Chang “Are You Killing Time? Predicting Smartphone Users’ Time-killing Moments via Fusion of Smartphone Sensor Data and Screenshots”, CHI’23
• Yu-Chun Chen, Kuei-Chun Kao, Yu-Jen Lee, Faye Shih, Wei-Chen Chiu, Yung-Ju Chang “Killing-Time Detection from Smartphone Screenshots”, UbiComp’21 (* indicates equal contribution)

Teaching

• Introduce to Natural Language Preprocessing (2022 Spring)
  • Lecturer: An-Zi Yen

Service and leadership

• CHI 2024 reviewer
• Ubicomp 2022 reviewer
• CS Student Union (academic affair)

Honors

• Scholarship for Academic Excellence performance 2 times (1% of computer science department per semester)
• NYCU GPE programming exam Ranked 1% (out of 200 students of NYCU)
• Best People’s Choice Award - Poster in Taiwan Association of Computer Human Interaction (TAICHI’21)
• NTU-IBM Q System 2020 Q-Camp, Best Presentation Award, Sep. 2020