Scalable hierarchical multitask learning algorithms for conversion optimization in display advertising

By Amr Ahmed, Abhimanyu Das, Alex Smola. 2014

42 citations recorded by the source corpus.

Many estimation tasks come in groups and hierarchies of related problems. In this paper we propose a hierarchical model and a scalable algorithm to perform inference for multitask learning. It infers task correlation and subtask structure in a joint sparse setting. Implementation is achieved by a distributed subgradient oracle and the successive application of prox-operators pertaining to groups and subgroups of variables. We apply this algorithm to conversion optimization in display advertising. Experimental results on over 1TB data for up to 1 billion observations and 1 million attributes show that the algorithm provides significantly better prediction accuracy while simultaneously beingefficiently scalable by distributed parameter synchronization.

Subfields: Computational Mechanics, Computer Vision and Pattern Recognition, Statistics and Probability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computational Mechanics

1. Qiang Yang
2. Yu Zhang
3. Ameet Talwalkar
4. Virginia Smith
5. Sinno Jialin Pan
6. Sulin Liu
7. Qirong Ho
8. Michele Donini

In Computer Vision and Pattern Recognition

1. Qiang Yang
2. Yun Raymond Fu
3. Dacheng Tao
4. Luc Van Gool
5. Yi Yang
6. Trevor Darrell
7. Ming-Hsuan Yang
8. Yu Zhang

In Statistics and Probability

1. Qiang Yang
2. Massimiliano Pontil
3. John Shawe-Taylor
4. David Martínez-Rego
5. Xiaodong He
6. Virginia Smith
7. Ameet Talwalkar
8. Qirong Ho

Citations in the corpus

• An Overview of Multi-task Learning (2018, neutral): “Web applications based on MTL include learning to rank in web searches [121], web search ranking [122], multi-domain collaborative filtering [123], behavioral targeting [124], and conversion maximization in display advertising [125].”
• A Survey on Multi-Task Learning (2017, methodology): “In [206], the MTRL method is extended to consider the hierarchical structure as well as structural sparsity for conversion maximization in display advertising.”
• Federated Multi-Task Learning (2017, background): “While several recent works [1, 33, 51, 52] have considered the issue of distributed MTL training, the proposed methods do not allow for flexibility of communication versus computation.”
• Deep Collaborative Filtering via Marginalized Denoising Auto-encoder (2015, methodology): “Following [44], we use the receiver operating characteristic (ROC) curve and the area under ROC (AUC-ROC) as our evaluation metrics.”
• Distributed Multi-Task Relationship Learning (2017, background): “1Though MTL is originally proposed for the problem where each task only has a small size of labeled training data, it has been shown by other researchers that when some tasks have relatively large amount of data, MTL can still help improve generalization across tasks by jointly exploiting informatio”
• Distributed variance regularized Multitask Learning (2016, background): “In [Ahmed et al., 2014], the authors develop a parallel MTL strategy for hierarchies of tasks.”
• A Multi-View Deep Learning Approach for Cross Domain User Modeling in Recommendation Systems (2015, background): “For recommendation systems, the authors in [2] leveraged the user’s historical search queries to build personalized taxonomy for recommending Ads.”