Primary speaker at the "Machine learning with TensorFlow" workshop conducted by CS department at UTDallas. Essentially discussed on topics that include image processing in TensorFlow, hyper-parameter tuning for Deep Neural Networks and integration with TensorBoard.
Environment modelling and its environmental aspects
Machine learning with TensorFlow
1. invites you to attend weekend workshop on
Machine Learning with TensorFlow
Saturday & Sunday Sep 22-23
Sat Sep 23
10am - 12:30pm
Introduction & experiences
with TensorFlow
Pavan Vutukuru
& Sruti Jain
Sat Sep 23
1 - 5pm
Hands-on demo/exercises
with TensorFlow
Pavan Vutukuru
& Sruti Jain
Sun Sep 24
11am - 12noon
Analytics using ML with
TensorFlow - WebEx
Dr. Scott Streit
Sun Sep 24
1 - 2:30pm
IOT TensorFlow Deep
Learning Demo
Russ Bodnyk
$5 for UTD folks, $25 fee for all guests
Lunch included on Saturday
Register @ bit.ly/prof-dev-utd
2. Workshop – Saturday, Sep 23 - Contents
Introduction
+ Need of Computationally efficient frameworks in Deep learning and Deep Mind project.
+ The origin of TensorFlow and progress since open-source launch.
+ Tensorflow Performance & scalability.
Machine learning - Complete implementation & Comparison
+ Normal implementation (Gradient Computation) & Tensorflow implementation
execution time comparisons, the use of broadcasting and understanding the graphical
computational model of Tensorflow for basic operations like the dot product, argmax,
element-wise multiplication etc. Using operators like argmax or matmul or anything
+ demo for TensorFlow contrib and why they have these basic implementations
+ TensorBoard: Visualize TensorFlow Graphs, monitor training performance & exploring
how the models represent the data step by step.
Our experiences with the TF Framework
+ ML at eBay: how ebay is leveraging Tensorflow, Tensorflow serving and kubernetes to
increase scalability and reliability of machine learning models in production.
+ Sruti will speak about image processing in TF, ML toolkit, integration of Keras &
Tensorflow, general problems one encounter while using TF in research.
Tensorflow internal features
+ TensorFlow Serving Models: TF Serving production models can be used for applying a
trained model in another application that are used in production environments.
+ Tensor2Tensor (Newly introduced Google Library based on TF) : T2T facilitates the
creation of state-of-the art models for a wide variety of ML applications, such as
translation, parsing, image captioning and more, enabling the exploration of various
ideas much faster than previously possible.
+ Support for implementation of large scale linear models that lets you jointly train a
linear model and a deep neural network.
External features
+ TensorFlow external compilers: Speed is everything for machine learning and
Tensorflow can make use of XLA, JIT or other compilation techniques to minimize
execution time & optimize computing resources.
+ Scaling up ML models using Distributed TensorFlow up to hundreds of TPU’s & GPU’s
and briefing on architectural designs.
+ Mobile & Embedded TensorFlow: Android to launch TensorFlow Lite for mobile
machine learning.
Conclusion
+ Comparison with other Deep learning frameworks like theano, caffe, Pytorch, CNTK
(Computational Network Toolkit by Microsoft)
+ Other exciting big-time AI models built on Tensorflow in various domains speech
recognition, image recognition, various visual detection tasks, language modeling &
language translation.
3. Talk & Demo – Sunday, Sep 24
Analytics using ML with TensorFlow – WebEx presentation
Presenter: Scott Streit, Computer Science Innovations, LLC (CSI), www.compscii.com
CSI focuses on Machine Learning and Computer Security. CSI performs analytics using
Machine Learning, primarily with Tensorflow. CSI work with Recurrent Neural Networks
(RNNs), Convolutional Neural Networks (CNNs) and a variety of other model types. CSI
have merged Big Data with Machine Learning in developing production systems for
clients.
IOT TensorFlow Deep Learning Demo
Presenter: Russ Bodnyk, Coded Intelligence
IOT data is exploding in a world of increasing complexity as new devices connect every
second. Applying intelligence to data is no longer optional, it is requisite. Security,
responsiveness, and interactivity can be improved by increasing number of intelligence
processes that run on IOT devices. Russ will demonstrate it with TensorFlow Deep
Learning processes running on device.