On April 22, [Mingli Lecture 2022 Issue 21] was honored to invite Associate Professor Qi Anyan of the University of Texas at Dallas to bring a presentation entitled "An Asymptotically Tight Learning Algorithm for Mobile-Promotion Platforms" in 317 of the Main Building. "The academic report, many teachers and students participated in the lecture report of Mr. Qi Anyan.
Mr. Qi Anyan first introduced that in the presence of uncertainty on both the supply side and the demand side, the mobile promotion platform conducts advertising campaigns for a single advertiser. Events arrive dynamically over time, divided into seasons; each event requires the platform to deliver a target number of mobile impressions from a desired set of locations over a desired time interval. The platform accomplishes these activities by acquiring impressions from publishers who provide ad space for the app by bidding in real-time on ad exchanges. Each position is characterized by its win curve, which is the relationship between the bid and the probability of winning an impression with that bid. The platform does not initially know the winning curves for each location of interest, it dynamically learns them based on its bids to earn impressions and the results achieved. Each impression earned is assigned to one of the ongoing campaigns. The platform's goal is to minimize its total cost (amount used to gain impressions and penalties for not meeting campaign goals) within the time frame of interest. Next, our main result is a bidding and allocation policy for this problem. We demonstrate that our policy is the best (asymptotically compact) for the problem using the notion of regret under policy, i.e. the difference between the expected total cost under the policy and the optimal cost of the perspective problem (i.e., One of the platforms has full information ahead of time on the winning curve for all positions): Regret under any strategy, where I is the number of seasons, is under our strategy. We demonstrate the performance of our strategy through numerical experiments on an instance testbed with input parameters based on our observations on a real-world mobile promotion platform.
Mr. Qi Anyan's wonderful offline report attracted many teachers and students. After the report, Mr. Qi had a nearly one-hour exchange with the participants. His kind guiding attitude and profound knowledge were appreciated by the participants. Highly rated.
