The Power of Data and AI: Highlights from the 2020 Tencent Advertising Algorithm Live Week

The 2020 Tencent Advertising Algorithm Live Week, organized by Tencent Advertising, featured a series of expert‑led live sessions covering frontier topics such as federated learning, machine learning, and big data, and the content has been distilled into a Q&A collection for algorithm enthusiasts.

Live Guest Liu Peng – Big Data and Artificial Intelligence expert, Vice President of iFlytek.

Live Theme "Where Does the Power of Data Come From"

Live Content The talks examined data‑centric perspectives on cutting‑edge technologies, explained the principles and practical applications of deep learning, and used computational advertising as a case study to illustrate AI’s evolution and transformation.

It is widely recognized that the Internet is a world of data, but leveraging massive datasets to extract commercial value is a new challenge for enterprises; deep learning is a key solution that unlocks data power. The article explores how deep learning releases data potential and surveys current AI frontiers.

Q: Why has artificial intelligence become so popular in recent years? A: AI has a 60‑year history with three hype cycles and two winters. The first wave (expert systems) relied on hand‑crafted rules, the second (statistical methods) let data teach machines, and the current third wave, driven by deep learning and neural networks, has achieved practical, large‑scale impact.

Q: Why is deep learning so powerful? A: Deep learning learns from massive data; the more data, the better the performance. For example, iFlytek’s speech‑recognition error rate dropped from 14.3% in 2012 to 1.6% in 2018 as user data accumulated.

Q: Why can AlphaGo defeat top players while AI assistants struggle with everyday conversation? A: AlphaGo generated huge amounts of game data for training, whereas everyday dialogue requires common‑sense knowledge that is scarce in training data, limiting current AI performance.

Q: Are real‑world advertising problems similar to the 2020 Tencent Advertising Algorithm Competition? A: Competition data is clean and ready‑to‑use, while real advertising must consider data acquisition, scale, and the “data‑first, problem‑second” workflow, where abundant data drives the discovery of optimization opportunities.

Q: Can you give an example of the "data‑first, problem‑second" approach? A: In an independent e‑commerce site, product attributes, descriptions, and sales figures constitute data that can train models to predict product popularity, enabling automated selection of millions of items versus manual analysis of a few hundred.

Q: How do human judgment and data‑driven learning compare? A: Human judgment can err; algorithms can identify higher‑revenue opportunities by analyzing large‑scale interaction data that humans cannot process, such as recommending ads to users who have previously shown interest.

Q: Does the sheer volume of data make manual filtering impossible? A: Yes; for example, an e‑commerce catalog may contain billions of SKUs, which no human can evaluate, but a ranking model can efficiently surface the most promising items.

The page also includes links to additional Q&A collections on federated learning, automated machine learning, and common advertising fraud techniques, and invites readers to watch the full live replay.