Peiyuan Liao is the youngest Chinese Kaggle grandmaster at only 20 years old with numerous gold medals and 1st, 2nd, and 3rd place finishes. He helped research two deep learning papers while in high school and now researches adversarial attacks on graph neural networks at Carnegie Mellon.

Learn more about Peiyuan:

https://liaopeiyuan.github.io/

https://www.kaggle.com/alexanderliao

Want to level-up your skills in machine learning and software engineering? Join the ML Engineered Newsletter: http://bit.ly/mle-newsletter

Comments? Questions? Submit them here: http://bit.ly/mle-survey

Follow Charlie on Twitter: https://twitter.com/CharlieYouAI

Take the Giving What We Can Pledge: https://www.givingwhatwecan.org/

Subscribe to ML Engineered: https://mlengineered.com/listen

Timestamps:

03:20 How Peiyuan was first exposed to CS and ML

06:45 Researching deep learning in high school

10:30 Researching graph neural networks at Carnegie Mellon (https://arxiv.org/abs/2009.13504)

20:30 How he keeps up with the field and gets research ideas

24:05 Research tools he uses

31:30 Advice for Kaggle beginners

34:30 How Peiyuan first approaches a new Kaggle competition

40:15 His team's 3rd-place solution to the 2020 Google Landmark Recognition Challenge (https://arxiv.org/abs/2010.05350)

50:30 How he approached the Global Wheat Detection challenge (https://www.kaggle.com/c/global-wheat-detection/discussion/175961)

56:40 How he decides to quit a Kaggle competition

59:25 The difference between him and the average Kaggler

01:03:20 Contributing to open source projects

01:06:00 Rapid Fire Questions

Links:

CAE-ADMM: Implicit Bitrate Optimization via ADMM-based Pruning in Compressive Autoencoders

Graph Adversarial Networks: Protecting Information against Adversarial Attacks

Peiyuan's Kaggle Profile

Open Neural Network Exchange (ONNX)

Apache TVM

Deformable Convolutional Networks

Google JAX

Robert Harper

Google Landmark Recognition 2020 Competition Third Place Solution

ArcFace: Additive Angular Margin Loss for Deep Face Recognition

Global Wheat Detection 2nd Place Solution

Lovász-Softmax loss