How to Apply Deep Learning to Real-World Problems

Download this episode

Download Video


Deep Learning is a new area of Machine Learning research, which has been introduced with the objective of moving Machine Learning closer to one of its original goals: Artificial Intelligence.

Join Jennifer Marsman as she welcomes Sonja Knoll to the show as they take a deep dive into Deep Learning as well as apply some real-world scenarios for you to try out on your own.

Looking for more information? Check out the following links below:




If you're interested in learning more about the products or solutions discussed in this episode, click on any of the below links for free, in-depth information:

Websites & Blogs:

 Follow the conversation @MS_ITPro
 Become a Fan @
Generic Episode Image Subscribe to our podcast via iTunes, Stitcher, or RSS



Available formats for this video:

Actual format may change based on video formats available and browser capability.

    The Discussion

    • User profile image

      The hyperlink tag for cognitive services is not closed properly, so the link is broken as it points to, which is not valid.

      Correct link for Cognitive Services is

    • User profile image

      @ppolyzos: Thanks for the heads-up... fixed

    • User profile image

      @ppolyzos: Thanks for the heads-up, it's been fixed

    • User profile image

      Sorta awesome stuff.
      Just a shame that the return to legacy neural networks methods, is such a slow learner technique, requiring powerful CPU, clouds for competitive deep machine learning.

    • User profile image

      have you tried to apply the word completion approach to number recognition? What would be the impact on the whole deep learning approach if the input data set is processed in real-time- the pixels of a picture for example are being introduced to the processor in random way?

    • User profile image

      Hi Tim - No, I haven’t tried that. To be clear, are you thinking of images as "sequences" of pixels? (similar to sequences of words for the word completion task). If that’s the case, I suppose one could use some sequence-related algorithms, like RNN/LSTM, but with two dimensions. Typically, CNNs are used for images since they encode the proximity of neighboring pixels. To your second point, one could submit the image to the model before it’s fully loaded, and get less-than-optimal results until the image is fully loaded.

    Comments closed

    Comments have been closed since this content was published more than 30 days ago, but if you'd like to send us feedback you can Contact Us.