Neural-network Field Aware Factorisation Machines for Online-behaviour Prediction
Section: Full talk Technical level: Advanced
In the AdTech mobile-app industry, bidding for each and every ad-request at a suitable price and in real-time is absolutely critical. Thus, there is always a need for more scalable and more accurate prediction models, which drive higher revenue.
Given the strengths of Deep learning to evaluate higher-order interactions between features and accordingly weight these, and the ability of Factorisation machines to reduce the cardinality of categorical features, the NFFM is a model that can work with high-cardinality categorical features and make better predictions than traditional logistic regression and tree models. Furthermore, the NFFM can also provide decent generalized predictions for unseen feature values.
Why neural nets model
Nothing required, just a basic understanding of neural nets.
There will be 2 people presenting Gunjan And Varun
Gunjan: Gunjan completed his Bachelor’s degree from Indian Insititute of Technology, Roorkee. He is currently working as Architect at InMobi where he looks over various recommendation systems, Prediction pipelines for the bidder, along with the entire backend systems. Previous to this he has worked for Google Hyderabad and Facebook Menlo Park.
Varun: Varun has done his Bachelors and Masters in Computer Science from IIT Kanpur. Currently working as a Sr Research Scientist at InMobi on predicting user response to ads, delivery metrics forecasting and applying deep learning models for deriving insights from ad creatives.
- FM: https://www.csie.ntu.edu.tw/~b97053/paper/Rendle2010FM.pdf
- FFM: http://research.criteo.com/ctr-prediction-linear-model-field-aware-factorization-machines/
- DeepFM: https://arxiv.org/pdf/1703.04247.pdf
- NFM: https://arxiv.org/pdf/1708.05027.pdf
Deep Learning Howlers: Downside of Learning only Statistical Regularities
It has been shown in a recent work ( https://arxiv.org/pdf/1711.11561.pdf), that deep convolutional learning networks do not learn higher level abstract concepts, but only statistical regularities. We investigate this claim by taking open source deep learning libraries and testing them out. more