AE

Loading...

No PhD to Harness the Power of Unstructured Data

19 views

Loading...

Loading...

Transcript

The interactive transcript could not be loaded.

Loading...

Rating is available when the video has been rented.
This feature is not available right now. Please try again later.
Premiered Dec 2, 2019

90% of the data generated every day in the world is unstructured. While it is clear how structured data provides value to machine learning projects, unstructured data is a little more difficult to use. At BEN, we focus a lot of our time on getting value out of data that comes from audio, video, and text. Large tech companies hire scores of PhDs to build incredibly complex neural networks that train for weeks on ridiculous amounts of compute power in order to take advantage of unstructured data. Most companies don’t have the time nor the money to invest in this kind of deep learning, and most data scientists wouldn’t even know where to begin if they were tasked with building a neural network to rival BERT in language modeling.

Luckily you don’t need 100 PhDs to be able to get value from unstructured data. I am going to go over some easy and practical ways that deep learning can be used to extract value from unstructured data that you may have just sitting around somewhere (don’t worry there will be example code). We are going to go beyond using a CNN to classify cats and dogs and dig into some other ways that deep learning can impact a data science project. Using pytorch and other companion libraries we will look at things like transfer learning, fine-tuning models to specific domains, feature extraction, and other techniques and how they can impact the performance of models being put into production. While this will be just a glimpse of how unstructured data can be used, it is a powerful place to start!
Speaker: Nelson griffiths

Loading...

Chat Replay is disabled for this Premiere.
When autoplay is enabled, a suggested video will automatically play next.

Up next


to add this to Watch Later

Add to

Loading playlists...