Photo of Zhenzhong Lan

Artificial intelligence & robotics

Zhenzhong Lan

Virtual psychological therapist with integrated artificial intelligence technologies to provide instant and effective consultation services.

Year Honored

Westlake University

Asia Pacific

Hails From
Asia Pacific

As one of the leading diseases around the world, mental disorders are a pressing topic to global public health. Scientists and entrepreneurs around the world are racing to find effective and scalable solutions. 

With the help of AI, Dr. Zhenzhong Lan, an assistant professor at Westlake University, is trying to tackle this problem in a unique way by integrating a set of artificial intelligence technologies to build a virtual smart therapist that can provide instant, effective, low-cost, and scalable consultation service.

Prior to joining Westlake University, Lan worked at Google AI as a research scientist and at WiZR as a computer vision and deep learning researcher for a total of 3 years. He completed his Ph.D. in Language and Information Technologies at Carnegie Mellon University in 2017. His research focused on how to index videos as a structured media and thereby perform accurate and fast large-scale video analysis and retrieval.

The work Lan managed to achieve was impressive. Together with his colleagues, they integrated early fusion and late fusion into the classic spatial pyramid algorithm to boost its performance and found an efficient way to aggregate local video features into global features. They participated in the Multimedia Event Detection (MED) competition in 2014 and ranked 1st place in 6 out of 8 conditions. 

If everything went well, Lan was already going to carry on the academic journey at Google. However, right before Lan’s Ph.D. graduation, he found out that many of his friends and colleagues were suffering from serious mental issues. 

“This completely changed my course,” said Lan. 

Since then, he started to look into mental health issues and was eager to do something to alleviate the feeling of isolation of those who were suffering. He joined WiZR, a California-based tech start-up that provides video analytics platforms, to learn how to build a business product from scratch. Meanwhile, he started to actively participate in psychological consultation, where he realized that there needed to be an automated approach to carry natural psychotherapy.

After 8 months at WiZR, Lan decided to join Google to research how to build an AI-driven conversation system. His work was integrated into Google Assistant and many other products. 

In particular, he invented an NLP (Natural Language Processing) model called ALBERT in 2020. The model was created to complete reading comprehension tasks designed for middle and high-school English exams in China. It achieved an accuracy of 89.4%, a stunning 45.3% improvement over the state-of-the-art model without full-network pretraining. 

The paper that described this work was accepted by ICLR, a top-tier AI conference. It has collected more than 1,200 citations within a year of its publication. The model, ALBERT, has also been put into real-world pilot programs to enhance its ability in a mental health-related environment. 

Lan joined Westlake University in June 2020. He wanted to leverage his expertise to help the mentally disordered patients in China. 

One of his ongoing research is how to train a conversational AI to have a long-term memory. In this way, the system can better understand the context of a conversation and even “remember what it had been told” a couple of days ago. This would significantly bring close the relationship between AI and people, which is especially beneficial during a psychological consultation. 

So far, Lan’s team has built an AI-assisted semi-automated consultation platform called Xinling (meaning listen attentively), which can do free psychological consultation. The platform has served more than 3,000 clients according to him.

“In the long run, I hope to incorporate other image and video processing techniques that we have developed into this virtual therapist model and enable it to understand human gesture and emotion,” said Lan.