DataFunSummit2023: Deep Learning‑Driven Multi‑Experiment Causal Inference and Distributed Causal Tools

The DataFunSummit2023 online summit focuses on practical causal inference for massive experiment scenarios, inviting experts from Tencent and Kuaishou to share the latest research and industrial practices.

Speaker Cheng Da‑xi (Kuaishou Data Scientist) introduces his background in economics and data science, highlighting his work on A/B testing platforms.

Speaker Zhang Renyu (Associate Professor, The Chinese University of Hong Kong; Kuaishou Economist & Tech Lead) presents the talk “Deep Learning‑Based Multi‑Experiment Effect Causal Inference,” outlining how to estimate overall effects of multiple concurrent experiments, identify the best experiment combination using debiased deep learning (DeDL), and validate the method on real‑world A/B tests.

Audience benefits include understanding how deep learning empowers causal inference, the value of double machine learning in production, and how frontier causal methods are deployed in large online platforms.

Speaker Zhang Jingjing (Tencent WeChat Experiment Platform Data Scientist) discusses the open‑source distributed causal inference package Fast‑Causal‑Inference, covering common causal algorithms (variance‑reduced t‑test, DID, IV, matching, DML), its architecture, and real‑world usage examples.

Attendees will learn typical application scenarios, computational principles, distributed implementation techniques, and practical usage of the toolkit.

Speaker Wen Zhonghui (Tencent Advertising Data Scientist) presents “Research on Long‑Term Effect Evaluation in A/B Experiments,” explaining why short‑term metrics miss long‑term impacts, current industry approaches for long‑term effect estimation, and the speaker’s own research on this problem.

Key takeaways include causes of short‑ and long‑term effects, industrial methods for long‑term evaluation, and recent research findings.

Overall, the summit offers deep insights into how modern AI and data‑science techniques enable robust causal inference and long‑term impact assessment in large‑scale online platforms.