Timezone: »

Environment Diversification with Multi-head Neural Network for Invariant Learning
Bo-Wei Huang · Keng-Te Liao · Chang-Sheng Kao · Shou-De Lin

Wed Nov 30 02:00 PM -- 04:00 PM (PST) @ Hall J #113

Neural networks are often trained with empirical risk minimization; however, it has been shown that a shift between training and testing distributions can cause unpredictable performance degradation. On this issue, a research direction, invariant learning, has been proposed to extract causal features insensitive to the distributional changes. This work proposes EDNIL, an invariant learning framework containing a multi-head neural network to absorb data biases. We show that this framework does not require prior knowledge about environments or strong assumptions about the pre-trained model. We also reveal that the proposed algorithm has theoretical connections to recent studies discussing properties of variant and invariant features. Finally, we demonstrate that models trained with EDNIL are empirically more robust against distributional shifts.

Author Information

Bo-Wei Huang (National Taiwan University)

I am Bo-Wei Huang, with a master’s degree in Computer Science and Information Engineering at National Taiwan University. When serving as a research student in my graduate school, I have been engaging in various machine learning projects and published two academic papers on the NeurIPS conference and a journal called Machine Learning Journal, respectively. Also, I have rich internship experience in Machine Learning. I am familiar with Deep Learning, Causality, Natural Language Processing, etc. As for programming language, I am proficient in Python, C, C++, and so on. I consider myself a highly-focused and self-motivated person, and I expect to exert my strengths and enthusiasm in the industry of technology. I am looking forward to speaking with you for more detail of the job opportunity. Please visit my LinkedIn for my resume, and feel free to contact me anytime via email.

Keng-Te Liao (National Taiwan University)
Chang-Sheng Kao (Department of computer science and informational engineering, National Taiwan University)
Shou-De Lin (National Taiwan University)

More from the Same Authors