During my final round interview at Tapad, the VP of Data Science noticed my academic background in neuroscience, and we started to discuss how neuroscience research could contribute to online advertising. Later in the interview, he asked “Have you heard of reinforcement learning (RL)?” And I said no. He explained this reward-guided learning paradigm, which immediately reminded me of my undergraduate research on aversive learning in fruit flies. After the interview, he emailed me the link to RL Courses on YouTube taught by David Silver, the leading scientist of AlphaGo.
Being quite intrigued with this neuroscience/computer science fusion, before I officially started my internship, I went through the video lectures, read the canonical textbook “Reinforcement learning: an introduction”, and tried to implement classical RL algorithms. After I told the VP of Data Science and my future mentor at Tapad what I had learned about RL, they said, “If you can give a brief overview for the data science team after you start your internship, you’ll be off to a great start.”
After I joined Tapad, I gave my first Reinforcement Learning presentation to the data science team. The slides were full of technical details, equations, demo codes, and latest research progress. After the presentation, I asked an engineer colleague for feedback, and he suggested that I provide more motivation to the audience before I jumped into the presentation of details. He said, “It seems to me that RL is just another type of heuristic, and it would be more attractive if you make the use cases in AdTech more clear. Otherwise, the audience may get lost and wonder ‘yeah this is a nice heuristic, but why would I care?’ ”
Why would the audience care? It is the job of the presenter to elaborate why a topic is interesting and relevant to the audience, and why they should care. Before my presentation, I was only chatting with my mentor and the VP of Data Science, both of whom already had some background of RL and its potential application in AdTech. The rest of the data science team may not be very familiar with this topic. I cannot just assume everyone can see RL’s connection to AdTech the same way I see it. After all, I knew nothing about RL when I interviewed in April, and I have spent months learning RL before I can explain it to others. I was not aware of the mental gap between what I assumed the audience knew and what they actually knew, and as a result, the audience were not on the same page with me. So I took the colleague’s suggestion and made a few changes to my slides.
After this tech-oriented presentation to the data science team, the VP of Data Science thought it would make an informative and interesting talk to the whole company, including data science, engineering, sales, business, marketing, and anyone who may be interested in this topic. To attract more audience and not to daunt non-engineers, we titled the event “Friendly Intro to Reinforcement Learning” and wrote a blurb as following:
– Have you ever wondered how Artificial Intelligence (AI) was able to beat the best human Go player?
– Have you ever wondered how we can make AI think like a human and act like a human?
– Have you ever wondered how we can achieve real-time adaptive optimization in personalized marketing?
Come join us in the “Friendly Intro to Reinforcement Learning” and get to know one of the most exciting fields in AI. No prior knowledge of AI required.
This Wednesday afternoon, more than 50 people showed up! Because this was a very diverse audience, I re-organzied my slides, removed most of the technical details, and focused more on concepts and applications. I also tried to build the connection between RL and AdTech as clear as possible early in the presentation to make sure the audience would be crystal clear why they cared about it. Of course, I used a lot of videos and audios to keep the audience awake (lol), and encouraged the audience to interrupt me and ask questions.
It was a one-hour long talk. I was glad and surprised to see no one left the talk halfway and most audience were quite engaged. Later in the week, I had more follow up discussion with people from different departments about RL and AI in general, and it was amazing to see how people viewed AI differently and what they thought of it.
Right after the talk, the VP of Data Science came to the podium and shared his valuable thoughts on my talk. I paraphrased his feedback in the following 3 points:
- Do not assume audience knows everything. Although I started to get aware of the mental gap from my first presentation in the data science team, I still had many blind spots of what the general audience actually knew. I used terms such as “stochastic” and “unsupervised learning”, which were quite common in data science, but might mean nothing to non-experts. If the audience get confused about some jargon, they may immediately get frustrated and disengaged. “Explain like I’m five” is a good start, and delve deep from there.
- Business presentation is different from academic presentation. I have presented several times at scientific conferences, and the scientific communities CARE a lot about references, footnotes, and experiment details. I adopted the same mindset in my RL talk, quoted original research articles, and spent quite some time emphasizing the academic discovery and background. However, the general audience do not really care much about the paper. They CARE a lot about the basic concepts, application, and relevance to their daily work. Rather than talking about the articles themselves, I should spend more time elaborating the conclusion and significance.
- Take the audience on a journey. This is the final goal. We may not know in advance who the audience are. For example, in a client meeting, we may not know whether we are meeting with the C-level management, the engineering team, or the sales team. The best we can do is start from the basic and top-level overview, and be adaptive as the journey goes on. Audience will be more engaged if they feel they are part of the story.
For more information about the RL talk, please visit Tapad’s Blog.