Image similarity and transfer learning in real-world applications

Main Talk:

Image similarity and transfer learning in real-world applications

Main Talk Description:

Image classification is a key component of many real-world applications, from visual search to face detection. Unfortunately, many image datasets do not have labels, making them harder to be used in supervised learning tasks.

In this presentation, I will leverage deep features that were extracted during the "Image similarity and transfer learning in real-world applications" talk by Rajat Arya, as well as clustering and graph analytics to provide a framework that could be used for creating labels for classification tasks.

I will show how to use deep features for building similarity graphs, and then apply graph algorithms such as pagerank and label propagation to help create labeled images in the dataset. The presentation will be entirely delivered from an IPython Notebook and written in Python using Dato's GraphLab Create.

Speaker:

Charlie Maalouf (Dato) -- Prior to joining Dato, Charlie spend 3 years building machine learning systems in the intersection of the energy and weather sectors. Charlie is also trained as an applied econometrician at Northwestern University where his research was focused on time series analysis of different financial and economic markets

Tentative Schedule:

6:30pm-7:00pm -- socializing

7:00pm-7:15pm -- lightning talk (TBD)

7:20pm - 8:20pm -- main talk

8:20pm - 9:00pm -- socializing

Special Thanks:

Nitro for hosting!

IMPORTANT NOTE: for building security purposes, YOU WILL BE ASKED FOR YOUR FULL NAME (as shown on your ID). You will need your ID present on the day of the event to enter.