You are viewing content from a past/completed QCon

Presentation: On a Deep Journey Towards Five Nines

Track: Sequential Data: Natural Language, Time Series, and Sound

Location: Cyril Magnin I

Duration: 12:00pm - 12:40pm

Day of week: Wednesday

Share this on:


At PayPal, achieving four nines of availability is the norm. In the pursuit of exponentially complex additional nines, the company has recently embarked on applying deep learning to forecasting datacenter metrics. Seq2Seq networks are ripe for application to this difficult problem, but little has been shared to the open community.


Aashish Sheshadri shines a light on how PayPal applies Seq2Seq networks to forecasting CPU and memory metrics at scale. Forecasting enables alerting flows to get a head start reducing MTTD, augment autoremidiation, and consequentially aid MTTR. In doing so Aashish describes the ecosystem and tooling that enables developers at PayPal to experiment, build and train ML models while creating reusable, reproducible and sharable work in the Jupiter and Kubernetes ecosystem. 

Speaker: Aashish Sheshadri

Staff ML Research Engineer @PayPal

Aashish Sheshadri is a research engineer at PayPal, where he currently ideates and applies deep learning to new avenues and actively contributes to the Jupyter ecosystem and the SEIF Project. He holds an MS in computer science from the University of Texas at Austin, where his research focused on active learning with human-in-the-loop systems. 

Find Aashish Sheshadri at

2019 Tracks

  • ML in Action

    Applied track demonstrating how to train, score, and handle common machine learning use cases, including heavy concentration in the space of security and fraud

  • Deep Learning in Practice

    Deep learning use cases around edge computing, deep learning for search, explainability, fairness, and perception.

  • Handling Sequential Data Like an Expert / ML Applied to Operations

    Discussing the complexities of time (half track) and Machine Learning in the data center (half track). Exploring topics from hyper loglog to predictive auto-scaling in each of two half-day tracks.

    Half-day tracks