AI Fundamentals and Deep learning
Frameworks using IBM Power Systems
Satyadhyan Chickerur, Ph.D
Prof, School of Computer Science and Engineering
Agenda
• Deep Learning in AI Landscape
• Neural Network Overview
• Neural Network Architectures(FFNN,RNN, LSTM,CNN)
• C1-Automatic detection of facial expressions and emotions
• C2-LSTM Based Lip Reading Approach for Devanagiri Script
• C3-Performance Analysis of Change Detection Algorithms On Multispectral Imagery
and Classification
• C4-Combining RGB and Depth images for Indoor Scene Classification using Deep
Learning
Deep Learning in the AI landscape
Ref :IBM PowerAI: Deep Learning Unleashed on IBM Power Systems Servers / RedBooks
Neural networks overview
Single Neuron
Neural Network Architectures
A)Feed Forward Neural Network
B)Recurrent Neural Network
C)LSTM
D)CNN
A) FeedForward Neural Nets
FeedForward Neural Network
B. Recurrent Neural network
C. LSTM
Ref : https://colah.github.io/posts/2015-08-Understanding-LSTMs/
D. CNN
Ref :https://medium.com/machine-learning-researcher/convlutional-neural-network-cnn-2fc4faa7bb63
Machine Learning Vs Deep Learning
Ref : https://www.smlease.com/entries/technology/machine-learning-vs-deep-learning-what-is-the-difference-between-ml-and-dl/
Machine Learning Vs Deep Learning
Ref :IBM PowerAI: Deep Learning Unleashed on IBM Power Systems Servers / RedBooks
ML WorkFlow —> DL WorkFlow
Ref : https://www.smlease.com/entries/technology/machine-learning-vs-deep-learning-what-is-the-difference-between-ml-and-dl/
C1- Automatic detection of facial expressions and emotions
Architecture Diagram
Automatic detection of facial expressions and emotions
Keypoint Extraction
Automatic detection of facial expressions and emotions
Data set - PCL / Kinect
Automatic detection of facial expressions and emotions
3D Model - matching and Expression Recogn
Automatic detection of facial expressions and emotions
Results
C2-LSTM Based Lip Reading Approach for Devanagiri Script
Architecture Diagram
LSTM Based Lip Reading Approach for Devanagiri Script
Keypoint Extraction
LSTM Based Lip Reading Approach for Devanagiri Script
Data Set
Paragraph for the Data Set
LSTM Based Lip Reading Approach for Devanagiri Script
Scaling - Two Approaches
LSTM Based Lip Reading Approach for Devanagiri Script
Overall Model
LSTM Based Lip Reading Approach for Devanagiri Script
Results - Word Recognition
Results - Sentence Recognition
C3- Performance Analysis of Change Detection Algorithms On
Multispectral Imagery and Classification
• Change detection - Selecting the two different images from the same region with
different frames of time, comparing those images, and detecting the areas where
changes have occurred.
• The various unsupervised change detection algorithms considered for our study are (1)
Principal Component Analysis (PCA), (2) PCA with K-means clustering, (3) Multivariate
Alteration Detection (MAD), (4) Iteratively Reweighted MAD (IRMAD).
Performance Analysis of Change Detection Algorithms On
Multispectral Imagery and Classification
Overall Model
Performance Analysis of Change Detection Algorithms On
Multispectral Imagery and Classification
Pine Data Set
Performance Analysis of Change Detection Algorithms On
Multispectral Imagery and Classification
Results
C4-Combining RGB and Depth images for Indoor Scene
Classification using Deep Learning
Overall Model
Combining RGB and Depth images for Indoor Scene
Classification using Deep Learning
Combining RGB and Depth images for Indoor Scene
Classification using Deep Learning
Data Set
Combining RGB and Depth images for Indoor Scene
Classification using Deep Learning
Results
Deep learning frameworks
source: https://aws.amazon.com/ko/machine-learning/amis/
IBM PowerAI
Ref : https://siliconangle.com/2017/05/10/ibm-beefs-powerai-software-accelerate-deep-learning-models/
Thank You

Deep Learning Fundamentals and Case studies using IBM POWER Systems

  • 1.
    AI Fundamentals andDeep learning Frameworks using IBM Power Systems Satyadhyan Chickerur, Ph.D Prof, School of Computer Science and Engineering
  • 2.
    Agenda • Deep Learningin AI Landscape • Neural Network Overview • Neural Network Architectures(FFNN,RNN, LSTM,CNN) • C1-Automatic detection of facial expressions and emotions • C2-LSTM Based Lip Reading Approach for Devanagiri Script • C3-Performance Analysis of Change Detection Algorithms On Multispectral Imagery and Classification • C4-Combining RGB and Depth images for Indoor Scene Classification using Deep Learning
  • 3.
    Deep Learning inthe AI landscape Ref :IBM PowerAI: Deep Learning Unleashed on IBM Power Systems Servers / RedBooks
  • 4.
  • 5.
    Neural Network Architectures A)FeedForward Neural Network B)Recurrent Neural Network C)LSTM D)CNN
  • 6.
    A) FeedForward NeuralNets FeedForward Neural Network
  • 7.
  • 8.
    C. LSTM Ref :https://colah.github.io/posts/2015-08-Understanding-LSTMs/
  • 9.
  • 10.
    Machine Learning VsDeep Learning Ref : https://www.smlease.com/entries/technology/machine-learning-vs-deep-learning-what-is-the-difference-between-ml-and-dl/
  • 11.
    Machine Learning VsDeep Learning Ref :IBM PowerAI: Deep Learning Unleashed on IBM Power Systems Servers / RedBooks
  • 12.
    ML WorkFlow —>DL WorkFlow Ref : https://www.smlease.com/entries/technology/machine-learning-vs-deep-learning-what-is-the-difference-between-ml-and-dl/
  • 13.
    C1- Automatic detectionof facial expressions and emotions Architecture Diagram
  • 14.
    Automatic detection offacial expressions and emotions Keypoint Extraction
  • 15.
    Automatic detection offacial expressions and emotions Data set - PCL / Kinect
  • 16.
    Automatic detection offacial expressions and emotions 3D Model - matching and Expression Recogn
  • 17.
    Automatic detection offacial expressions and emotions Results
  • 18.
    C2-LSTM Based LipReading Approach for Devanagiri Script Architecture Diagram
  • 19.
    LSTM Based LipReading Approach for Devanagiri Script Keypoint Extraction
  • 20.
    LSTM Based LipReading Approach for Devanagiri Script Data Set Paragraph for the Data Set
  • 21.
    LSTM Based LipReading Approach for Devanagiri Script Scaling - Two Approaches
  • 22.
    LSTM Based LipReading Approach for Devanagiri Script Overall Model
  • 23.
    LSTM Based LipReading Approach for Devanagiri Script Results - Word Recognition Results - Sentence Recognition
  • 24.
    C3- Performance Analysisof Change Detection Algorithms On Multispectral Imagery and Classification • Change detection - Selecting the two different images from the same region with different frames of time, comparing those images, and detecting the areas where changes have occurred. • The various unsupervised change detection algorithms considered for our study are (1) Principal Component Analysis (PCA), (2) PCA with K-means clustering, (3) Multivariate Alteration Detection (MAD), (4) Iteratively Reweighted MAD (IRMAD).
  • 25.
    Performance Analysis ofChange Detection Algorithms On Multispectral Imagery and Classification Overall Model
  • 26.
    Performance Analysis ofChange Detection Algorithms On Multispectral Imagery and Classification Pine Data Set
  • 27.
    Performance Analysis ofChange Detection Algorithms On Multispectral Imagery and Classification Results
  • 28.
    C4-Combining RGB andDepth images for Indoor Scene Classification using Deep Learning Overall Model
  • 29.
    Combining RGB andDepth images for Indoor Scene Classification using Deep Learning
  • 30.
    Combining RGB andDepth images for Indoor Scene Classification using Deep Learning Data Set
  • 31.
    Combining RGB andDepth images for Indoor Scene Classification using Deep Learning Results
  • 32.
    Deep learning frameworks source:https://aws.amazon.com/ko/machine-learning/amis/
  • 33.
    IBM PowerAI Ref :https://siliconangle.com/2017/05/10/ibm-beefs-powerai-software-accelerate-deep-learning-models/
  • 34.