Page 61 - Flytxt

AUTOML:

BRINGING IN REAL ARTIFICIAL

INTELLIGENCE CAPABILITIES

Traditional data science practices may not be the most efficient
ones when it comes to studying dynamic and infinite data sets

like customer behavior changes. Automatic Machine Learning
(AutoML) attempts to automate the end-to-end process of

applying machine learning to real-world problems. It is expected
to handle even complex data science tasks with minimal human

intervention and could turn out to be the ultimate bringer of
true Artificial Intelligence capabilities in years to come.

- Amit Meher, Senior Manager - Data Sciences R&D, Flytxt

Traditional data science practices focus the heterogeneous nature of data types,
on solving a point problem after taking data distribution, skewness, missing
into consideration a specific data set values, outliers, etc. associated with

and domain at a given point of time. them. Consequently, data scientists end
However, this may not be an effective up building customized models on an

strategy in terms of scalability and
efficiency, as the same model may not
provide optimal results when applied to From Flytxt’s perspective,

a different data set or domain. AutoML could significantly
enhance AI capability of an
A concrete example of this inefficiency
can be seen in the process of predicting organization. With AutoML,

churners in the telecommunication data scientists could be
domain. The churn model developed relieved from doing repetitive
for a specific Communication Service tasks required to build Machine

Provider (CSP) may not yield good results Learning pipelines and can now
when applied to a data set pertaining focus on solving complex data
to a different CSP. This could be due to science problems and devising

the difference in the subscribers’ churn new algorithms. Development
behavior across CSPs and may require and maintenance of packaged

a different class of learning algorithms analytics models will become
and hyper parameter settings to yield easier and it will no longer
optimal accuracy. Also, this model, require extensive human

customized for a specific OpCo can’t be intervention.
applied across other CSPs because of

INSIGHTZ - VOLUME 03, 2018 61