BigML Education
DEEPNETS
October 2017
BigML Education Program 2Deepnets
In This Video
• Introduction to deepnets
• Exploration of deepnets on the BigML Dashboard
• Evaluate and make predictions with deepnets
• Comparison of other supervised learning methods
BigML Education Program 3Deepnets
DNN Myths
• Next step in evolution that will either save or destroy humanity
• Such outstanding performance that other supervised learning
methods are now irrelevant
• Great for the established niche applications (e.g., vision and
speech recognition) but less interesting for general use
BigML Education Program 4Deepnets
Logistic Level Up
Outputs
Inputs
BigML Education Program 5Deepnets
Logistic Level Up
wi
Class “a”, logistic (w, b)
BigML Education Program 6Deepnets
Logistic Level Up
Outputs
Inputs
Hidden layer
BigML Education Program 7Deepnets
Logistic Level Up
Class “a”, logistic (w, b)
Hidden node 1,

logistic (w, b)
BigML Education Program 8Deepnets
Logistic Level Up
Class “a”, logistic (w, b)
Hidden node 1,

logistic (w, b)
n
hidden nodes?
BigML Education Program 9Deepnets
Logistic Level Up
Class “a”, logistic (w, b)
Hidden node 1,

logistic (w, b)
n
hidden nodes?
BigML Education Program 10Deepnets
Logistic Level Up
Class “a”, logistic (w, b)
Hidden node 1,

logistic (w, b)
n
hidden 

layers?
BigML Education Program 11Deepnets
BigML Dashboard Demo
Predicting health inspection scores for
restaurants in San Francisco:
https://www.kaggle.com/datasf/sf-restaurant-inspection-scores/data
BigML Education Program 12Deepnets
Algorithm Comparison
BigML Education Program 13Deepnets
Caveat Emptor
• Remember deep learning is just another sort
of supervised learning algorithm
“…deep learning has existed in the neural network community for over 20 years. Recent advances are
driven by some relatively minor improvements in algorithms and models and by the availability of large
data sets and much more powerful collections of computers.” — Stuart Russell
• Deep learning is less useful for:
• Small data (where that could still be thousands of instances)
• Problems where you could benefit by iterating quickly (better
features always beats better models)
• Problems that are easy, or for which top-of-the-line
performance isn’t absolutely critical
BigML Education Program 14Deepnets
Review
• Deepnets complement the other supervised learning
methods on BigML
• BigML’s automatic parameterization options find the best
network to solve your problem with 1-Click
• BigML’s PDP visualization and Evaluation Comparison Tool
make it easy to interpret your deepnet results
• Deepnets can outperform other BigML methods and
different off-the-shelf algorithms from popular ML libraries

BigML Education - Deepnets

  • 1.
  • 2.
    BigML Education Program2Deepnets In This Video • Introduction to deepnets • Exploration of deepnets on the BigML Dashboard • Evaluate and make predictions with deepnets • Comparison of other supervised learning methods
  • 3.
    BigML Education Program3Deepnets DNN Myths • Next step in evolution that will either save or destroy humanity • Such outstanding performance that other supervised learning methods are now irrelevant • Great for the established niche applications (e.g., vision and speech recognition) but less interesting for general use
  • 4.
    BigML Education Program4Deepnets Logistic Level Up Outputs Inputs
  • 5.
    BigML Education Program5Deepnets Logistic Level Up wi Class “a”, logistic (w, b)
  • 6.
    BigML Education Program6Deepnets Logistic Level Up Outputs Inputs Hidden layer
  • 7.
    BigML Education Program7Deepnets Logistic Level Up Class “a”, logistic (w, b) Hidden node 1, logistic (w, b)
  • 8.
    BigML Education Program8Deepnets Logistic Level Up Class “a”, logistic (w, b) Hidden node 1, logistic (w, b) n hidden nodes?
  • 9.
    BigML Education Program9Deepnets Logistic Level Up Class “a”, logistic (w, b) Hidden node 1, logistic (w, b) n hidden nodes?
  • 10.
    BigML Education Program10Deepnets Logistic Level Up Class “a”, logistic (w, b) Hidden node 1, logistic (w, b) n hidden layers?
  • 11.
    BigML Education Program11Deepnets BigML Dashboard Demo Predicting health inspection scores for restaurants in San Francisco: https://www.kaggle.com/datasf/sf-restaurant-inspection-scores/data
  • 12.
    BigML Education Program12Deepnets Algorithm Comparison
  • 13.
    BigML Education Program13Deepnets Caveat Emptor • Remember deep learning is just another sort of supervised learning algorithm “…deep learning has existed in the neural network community for over 20 years. Recent advances are driven by some relatively minor improvements in algorithms and models and by the availability of large data sets and much more powerful collections of computers.” — Stuart Russell • Deep learning is less useful for: • Small data (where that could still be thousands of instances) • Problems where you could benefit by iterating quickly (better features always beats better models) • Problems that are easy, or for which top-of-the-line performance isn’t absolutely critical
  • 14.
    BigML Education Program14Deepnets Review • Deepnets complement the other supervised learning methods on BigML • BigML’s automatic parameterization options find the best network to solve your problem with 1-Click • BigML’s PDP visualization and Evaluation Comparison Tool make it easy to interpret your deepnet results • Deepnets can outperform other BigML methods and different off-the-shelf algorithms from popular ML libraries