Fragments 2017

A conference on the mobile ecosystem in India

Running Deep Learning Models on Mobile with optimised speed

Submitted by A Naveen Kumar (@4nonymou5) on Friday, 18 August 2017

videocam_off

Technical level

Intermediate

Section

Full Talk

Status

Confirmed & Scheduled

View proposal in schedule

Vote on this proposal

Login to vote

Total votes:  +2

Abstract

The ability to train the task specific deep learning models is very easy these
days, with the wide range of available libraries and documentation around it. But,
the difficulty lies in bringing it to production ready mode. Especially, if the
application concentrates on Mobile platform.
Though there are existing wrappers of certain libraries to make them work, but,
as of now, they are slow and use up almost the entire memory space of the
phone.
In this talk, I would like to explain, what can be done to make things faster and
how to make models with reduced size. The aim of this talk is to provide insights
on what would be the difficulties which lie ahead and how to build your own
libraries in both iOS and Android.

Outline

What is Deep Learning ?
5 mins, introduction and explanation

What are the difficulties faced to push them into mobile production ?
10 minutes

How to solve it in IOS ?
5 minutes

How to solve it in Android ?
5-10 minutes

Conclusion
5 minutes

Requirements

Basic understanding of AI and their usage

Speaker bio

I am a member of the data science team at
Semantics3 - building data-powered software for ecommerce-focused companies. Over the years, I have had the chance to work on various aspects of Deep Learning, one such scenario was running the models
on mobile. We made an app named Flo, which got featured by Apple on their
twitter page for using AI and their framework to make it run faster.

Links

Slides

https://docs.google.com/presentation/d/e/2PACX-1vSUga_8ZSly2laX86lEKxWTheDn6CHsZ96zb1ZMxQXCH1yj-NYhPQ7hT1ASomFcJIepy93vw4IOeTWw/pub?start=false&loop=false&delayms=3000

Comments

  • 1
    Mohammad Fasahath (@mdfasahath) a year ago

    Cool stuff! Looking forward for this.

Login with Twitter or Google to leave a comment