Deep Learning
with TensorFlow
Leveraging the Power of Deep Learning Today
Justin Grammens
Lab651, LLC
AppliedAI & Emerging Technologies North 501(c)(3)
What Will We Cover?
We’ll start off by defining TensorFlow, Machine Learning
and Neural Networks. We’ll then jump into the steps and
tools needed to utilize Machine Learning. Finally, we’ll show
examples and techniques to tune TensorFlow to help get
the results you want.
About Me – Justin Grammens
• Serial entrepreneur with a passion for starting businesses and
communities in the areas of emerging technology
• Cofounder of Lab651 – Building connected products for our customers
• Cofounder of IoT Fuse – Conference on IoT hosting 1000+ people
• Owner IoT Weekly News – Publication covering trends in the IoT
• Adjunct Professor – Teaching graduate level course on IoT the
University of Saint Thomas in Saint Paul, MN
• Cofounder of Emerging Tech North & Applied AI – 501(c)(3) non-profit
to advance and grow the Emerging Technologies and AI
What is TensorFlow
TensorFlow is an end-to-end open source platform for machine learning. It has a
comprehensive, flexible ecosystem of tools, libraries and community resources that
lets researchers push the state-of-the-art in ML and developers easily build and
deploy ML powered applications.
Source: TensorFlow.org
What is a Tensor
TensorFlow derives its name from performing operations on multidimensional data arrays
Source: Edureka.co
It’s a wild and
crazy world out
there….
Source: https://towardsdatascience.com
Machine Learning Steps
Datasets
• Training Dataset
• Sample of data used to fit the model
• Validation Dataset
• Sample of data used to provide an unbiased evaluation of a model fit on the training
dataset while tuning model hyperparameters.
• Test Dataset
• Sample of data used to provide an unbiased evaluation of a final model fit on the
training dataset.
Source: https://www.toptal.com/resume/alan-reiner
What is are Jupyter Notebooks?
• An open-source web application that
allows you to create and share documents
that contain live code, equations,
visualizations and narrative text.
• Supports over 40 programming
languages, including Python, R, Julia, and
Scala.
What is Google Colab?
• Product from Google Research
• Allows anybody to write and execute
arbitrary python code through the browser
• Colab is a hosted Jupyter notebook service
that requires no setup to use, while
providing free access to computing
resources including GPUs
What is Keras
• A deep learning API written in Python
• The high-level API of TensorFlow 2
• Focus is on enabling fast
experimentation
• Accessed using the tf.keras package
Example
• Train a model to do Celsius to Fahrenheit conversion
• Train a model to recognize clothing from the Fashion MNIST dataset
General Guide To Improve Performance
• Increase or decrease the hidden layers or number of neurons
• Watch out for overfitting!
• Change the loss function
• Change the optimizer ( Adam .01 values )
• Get more training data
• Augment the data you have (skew, flip, rotate images)
• Evaluate a different model
What Did We Cover?
Defined TensorFlow, Machine Learning and Neural
Networks. We jumped into the steps and tools needed to
utilize Machine Learning. Finally, we showed examples and
techniques to tune TensorFlow to help get the results you
want.
Thank You!
Justin Grammens
justin@lab651.com
Links/Resources:
https://lab651.com
https://EmtechNorth.org
https://AppliedAI.mn
https://podcast.AppliedAI.mn

Deep Learning with TensorFlow

  • 1.
    Deep Learning with TensorFlow Leveragingthe Power of Deep Learning Today Justin Grammens Lab651, LLC AppliedAI & Emerging Technologies North 501(c)(3)
  • 2.
    What Will WeCover? We’ll start off by defining TensorFlow, Machine Learning and Neural Networks. We’ll then jump into the steps and tools needed to utilize Machine Learning. Finally, we’ll show examples and techniques to tune TensorFlow to help get the results you want.
  • 3.
    About Me –Justin Grammens • Serial entrepreneur with a passion for starting businesses and communities in the areas of emerging technology • Cofounder of Lab651 – Building connected products for our customers • Cofounder of IoT Fuse – Conference on IoT hosting 1000+ people • Owner IoT Weekly News – Publication covering trends in the IoT • Adjunct Professor – Teaching graduate level course on IoT the University of Saint Thomas in Saint Paul, MN • Cofounder of Emerging Tech North & Applied AI – 501(c)(3) non-profit to advance and grow the Emerging Technologies and AI
  • 4.
    What is TensorFlow TensorFlowis an end-to-end open source platform for machine learning. It has a comprehensive, flexible ecosystem of tools, libraries and community resources that lets researchers push the state-of-the-art in ML and developers easily build and deploy ML powered applications. Source: TensorFlow.org
  • 5.
    What is aTensor TensorFlow derives its name from performing operations on multidimensional data arrays
  • 6.
  • 12.
    It’s a wildand crazy world out there…. Source: https://towardsdatascience.com
  • 13.
  • 14.
    Datasets • Training Dataset •Sample of data used to fit the model • Validation Dataset • Sample of data used to provide an unbiased evaluation of a model fit on the training dataset while tuning model hyperparameters. • Test Dataset • Sample of data used to provide an unbiased evaluation of a final model fit on the training dataset.
  • 16.
  • 17.
    What is areJupyter Notebooks? • An open-source web application that allows you to create and share documents that contain live code, equations, visualizations and narrative text. • Supports over 40 programming languages, including Python, R, Julia, and Scala.
  • 18.
    What is GoogleColab? • Product from Google Research • Allows anybody to write and execute arbitrary python code through the browser • Colab is a hosted Jupyter notebook service that requires no setup to use, while providing free access to computing resources including GPUs
  • 19.
    What is Keras •A deep learning API written in Python • The high-level API of TensorFlow 2 • Focus is on enabling fast experimentation • Accessed using the tf.keras package
  • 20.
    Example • Train amodel to do Celsius to Fahrenheit conversion • Train a model to recognize clothing from the Fashion MNIST dataset
  • 21.
    General Guide ToImprove Performance • Increase or decrease the hidden layers or number of neurons • Watch out for overfitting! • Change the loss function • Change the optimizer ( Adam .01 values ) • Get more training data • Augment the data you have (skew, flip, rotate images) • Evaluate a different model
  • 22.
    What Did WeCover? Defined TensorFlow, Machine Learning and Neural Networks. We jumped into the steps and tools needed to utilize Machine Learning. Finally, we showed examples and techniques to tune TensorFlow to help get the results you want.
  • 23.

Editor's Notes

  • #2 March Open Source North Speakers series. Be sure to register for Open Source North Conference on May 20th
  • #4 Connect with me on linkedin – podcast as well
  • #6 Tensors are a type of data structure stored as multi-dimensional arrays. Scalar is one number, vector is list of numbers, matrix is a 2-dimentional array, tensor is a multidimentional array. 0-rank, 1-rank, 2-rank, 3 dimensions or more are typically Tensors. Computation on arrays is what is going on inside neural networks
  • #8 Choose your own adventure Supervised learning labeled.
  • #11 When the neural network is initialized, weights are set for its individual elements, called neurons. Inputs are loaded, they are passed through the network of neurons, and the network provides an output for each one, given the initial weights.
  • #14 ANN – only fed forward, does not propagate backward to improve the accuracy, CNN has back propagation and good for data that doesn’t change over time, RNN better for time series or cases in which the next word an depend on words that are 3 or 4 in the past. Experiment with various options. We’ll be focused on CNN’s for our examples. Long Short-Term Memory (LSTM) networks are a type of recurrent neural network capable of learning order dependence in sequence prediction problems.
  • #15 Backpropagation is really about determining how changing the weights impact the overall cost of the neural network.
  • #16 Gathering training data can often be the most difficult part, one you have it then preparing/cleaning it can present challenge. Most importantly – you are looking for the most optimal solution that is not over filling the data.
  • #17 Over filling occurs when you have data that is too close in the test to the training
  • #18 Always keep this in mind: deep learning models tend to be good at fitting to the training data, but the real challenge is generalization, not fitting.
  • #21 Tools