Women in Big Data Global



Machine Learning Is the New Black

Women in Big Data

By Regina Karson,

May 30, 2017


Women in Big Data and SAP hosted a great evening that featured speakers from SAP, Google and Nvidia discussing Machine Learning, Neural Networks, Deep Learning, etc., from their respective company views and from their respective personal views.

(L. to R.) Julie Bernauer, Nazanin Zaker, Margaret Laffan, Yuan Xue

Special thank you to SAP for the venue and to these ladies for their time and expertise:

  • Nazanin Zaker, Data Scientist and ML Researcher, SAP
  • Yuan (Emily) Xue, Tech Lead Manager, Google
  • Julie Bernauer, Team Lead Pursuit Engineering Solution Architects, Nvidia
  • Margaret Laffan, Director Business Development, SAP

Nazanin kicked the evening off with a review of SAP’s history being based in enterprise business services. 76% of the world’s transaction revenue touches an SAP system and now have Machine Learning via Leonardo applied to said business data.

Machine Learning is not rule based. Computers can learn from data without explicit programming. The time for Machine Learning has come, because Big Data is a fact of life (IoT, etc.) Improvements in hardware, such as GPUs and multicore microprocessors, provide engines for high scale data, and deep learning algorithms have been developed as well. Deep learning looks at NLP, image processing, computer vision, etc.

Today the digital supply chain connects the system of record (SAP S/4HANA) with the system of differentiation (business planning) and the system of intelligence (SAP Leonardo).

See www.sap.com/trends/machine-learning.html for more info.

See SAP slide five (5) links for education website that will provide an overview course.

Yuan (Emily) was the next presenter and discussed Google’s vision to democratize AI, and how tech will bring everyone together. AI will allow new insights, new products and new use cases in retail (product recommendations, supply chain), media/entertainment (computer vision technology, such as sentiment analysis and facial expression recognition on an image), finance (fraud detection), etc.

Democratize AI via:

  • algorithms that will provide cloud APIs for speech
  • computer vision
  • translate and NLP
  • computing platform via Google cloud (TensorFlow as ML engine)
  • data (tools, market place trades, Kaggle, etc.)
  • talent (advance solutions labs within a cloud team).

For example, there are APIs made available by Google that use a pre-train model (trained on big data), leveraging deep learning neural networks. Later, a user can apply these pre-trained models by an API call and get the result in real time. Examples are facial expression recognition, object detection and image search.

 Julie discussed how Nvidia supports computing for the most demanding users with both software and hardware offerings. Deep learning is in the cloud, medicine/biology, media/entertainment and security/defense.

Machine learning (ML) is a subfield of CS that gives computers the ability to learn without being explicitly programmed. Neural networks (NN) is a subset of ML, and Deep Learning is a subset of NN.

NN is a learning algorithm that is inspired by the structure and functional aspects of biological neural networks.  NN is a collection of trainable math units that collectively learn complex functions. One neuron is one function.  A function has input that provides output.

DL consists of multiple layers in NN. In the case of image recognition, training data consists of 10-100 million images.

3 kinds of networks:

  • DNN (Deep Neural Networks) – all fully connected layers
  • CNN (Convolution Neural Network) – convolutional layers
  • RNN (Recurrent Neural Network) – recurrent neural networks

Nvidia supports the compute power needed for deep learning. Self-driving cars are a use case for an application of deep learning. GPUs are designed to accelerate the operations needed for creating Deep Learning models by converting the operations needed to train the DL model into matrix multiplications.

Nvidia is supporting DL building blocks for DL frameworks such as Caffe, Tensorflow, etc. See slides for SDK, workflow, resources info.

Panel discussion with Nazanin, Julie, Emily, Margaret:

The ladies discussed diverse backgrounds that put them on the data scientist path.

Advice to get into the data science field:

  • Formal schooling is not a requirement as one can go to MOOCs such as Coursera and Udacity; check out YouTube videos; use open source for hands on experience.
  • Showcase talent via social activities such as Kaggle competitions, GITHUB, Hackathons, Meetups, Jeremy Howard courses–and don’t forget soft skills needed to articulate value, etc.

In closing, here’s some advice on being a woman in the field: The more value you show, the more credibility you will build. Make people successful around you.

Related Posts