讲座:Build Confounders from Images with Interpretability 发布时间:2024-11-29
嘉 宾:冯为 PhD. Candidate University of Maryland, College Park
主持人:刘佳璐 助理教授 awc777万象城娱乐官网
时 间:2024年12月2日(周一)9:30-11:00
地 点:awc777万象城娱乐官网徐汇校区安泰楼A305
内容简介:
In causal inference, the increasing use of unstructured data like texts, audios, and images is prominent. Image data, particularly valuable, allows extraction of information such as brightness, colors, and socio-economic indicators that are critical as confounders in models. Traditionally, this process required extensive domain knowledge and specialized tools, making it costly and often leading to key data omissions. It also demanded significant computational resources for large datasets. However, recent advances in deep learning and computer vision have introduced an embedding-based method that compresses image data into numerically compact embeddings, reducing the need for domain expertise but posing challenges in data interpretability and specificity control. To bridge these gaps, we propose a novel framework that not only simplifies the creation of image embeddings but also enhances their interpretability. This framework ensures that only essential information predictive of both treatment effects and outcomes is captured, disregarding irrelevant data. It incorporates a unique interpretation mechanism, enabling researchers to validate the relevance of the captured information through semi-simulated datasets. By facilitating more efficient and interpretable embeddings, our approach offers significant improvements in the accuracy and reliability of causal inferences derived from large-scale image data.
演讲人简介:
Wei Feng is a PhD candidate in Information Systems at the University of Maryland-College Park. He earned his Bachelor's in Electronic Business from Sichuan University and a Master's in Information Technology and Management from the University of Texas at Dallas.Wei's research encompasses interpretable machine learning, human-AI collaboration systems, and digital education with a focus on mitigating digital inequality. His innovative approaches aim to enhance the accuracy, efficiency, and transparency of AI within complex decision-making processes.His significant works include methodologies for enhancing the interpretability of image data in causal inference and developing frameworks that improve predictive analytics. His job market paper particularly addresses advancements in causal inference with image data.Wei has shared his research at international venues such as the Symposium on Statistical Challenges in Electronic Commerce Research and the International Conference on Information Systems. He also serves as a reviewer for academic conferences and was an editorial assistant at the Journal of Financial Services Research. As an educator, Wei has taught Information Systems at the University of Maryland, demonstrating a commitment to imparting knowledge and inspiring innovation among his students.
欢迎广大师生参加!