3. 3
1
2
3
AUTOMATOR DECIDER RECOMMENDER ILLUMINATOR EVALUATOR
EXAMPLES EXAMPLES EXAMPLES EXAMPLES EXAMPLES
When AI has all
the context and
needs to quickly
reach a
conclusion..
AI should decide
and implement .
When AI has
plenty of context,
but an human
touch is needed for
execution… AI
should decide, and
humans should
implement.
When there are
multiple repetitive
decisions to be
made, but AI is
missing necessary
context.. AI should
recommend, and
humans should
decide.
When inherently
creative work will
benefit from
machine learning…
humans should
leverage
AI-generated
insights.
When there’s not
enough context,
and the stakes are
high…
humans should
generate scenarios
for AI to evaluate.
Dynamic Pricing
Engines,
Algorithmic Add
displays
Predictive
Maintenance, Call
center optimization
Promotional
Calendar creation,
Sales and
operation Planning
Product design
based on customer
usage
Large seasonal
promotions, Digital
twin simulation for
operation
AI at Scale
4. What are the benefits of AI in the enterprise?
Better
Quality
Better
Talent
Management
Business
model
innovation
and
expansion.
Improved
customer
services
Improved
Monitoring
Faster
Product
Development.
Enterprise
AI
5. Applications of AI at Work
01 05
02
03
06
07
Customer Experience Service
and Support
Targeted Marketing
Smarter supply chains
Quality Control and Quality
Assurance
Contextual Understanding
Optimization
04 08
Safe and Smart operations More Effective Learning
12. Latency sensitive
Decisions
Large Batch Predictions
Instantaneous predictions
Examples:
• Payment processing
• Fraud detection
• Loan/claim pre-approval
Real-time prediction using “fresh” and
large operational data
Examples:
• Anomaly detection
• Escalation risk prediction
• Dynamic price optimization
ENTERPRISE AI
13. ENTERPRISE AI
Accelerating and Optimizing AI lifecycle with IBM DB2
01 02
Integrating Open
Source models with
DB2
Developing and
Deploying DB2-Native
ML models
15. ENTERPRISE AI
PYTHON UDF : PYTHON MODELS VIA DB2
Export the ML pipeline
by serializing python
joblib
Db2 Server
Host OS
Db2
Instance
Python
Runtim
e
24. STEP 2. DATA EXPLORATION
Use these stored procedures to evaluate the content of the given
data
IDAX.SUMMARY1000 - Summary of up to 1000 columns
IDAX.COLUMN_PROPERTIES - Create a column properties table
IDAX.GET_COLUMN_LIST - Get a list of columns
25. STEP 3 : DATA TRANSFORMATION
Use the following stored procedures to transform the data before
passing it to a machine learning algorithm.
IDAX.IMPUTE_DATA - Impute missing data
IDAX.SPLIT_DATA - Split data into training data and test data
IDAX.STD_NORM - Standardize or normalize columns of the input table
IDAX.EFDISC - Discretization bins of equal frequency
IDAX.APPLY_DISC - Discretize data by using limits for discretization bins
26. STEP 4 : MODEL BUILDING
Use these stored procedures to build machine learning models.
Decision trees - IDAX.GROW_DECTREE A decision tree is a hierarchical, graphical structure
accurately classify a model.
Linear regression - IDAX.LINEAR_REGRESSION is the most commonly used method of predictive
analysis.
Naive Bayes IDAX.NAIVEBAYES - The Naive Bayes classification algorithm is a probabilistic
classifier.
K-means clustering IDAX.KMEANS - The K-means algorithm is the most widely used clustering
algorithm
27. STEP 5 : MODEL EVALUATION
Use these stored procedures to evaluate the performance of your model by comparing predictions to the
true values.
IDAX.CMATRIX_STATS - Calculate classification quality factors from a confusion matrix
IDAX.CONFUSION_MATRIX - Build a confusion matrix
IDAX.MAE - Calculate the mean absolute error of regression predictions
IDAX.MSE - Calculate the mean squared error of regression predictions
28. STEP 6 : MODEL INFERENCING/ DEPLOYMENT
Use these stored procedures to make predictions with your trained machine learning model.
IDAX.PREDICT_DECTREE - Apply a decision tree model
IDAX.PREDICT_KMEANS - Apply a K-means clustering model to new data
IDAX.PREDICT_LINEAR_REGRESSION - Apply a linear regression model to a target
IDAX.PREDICT_NAIVEBAYES - Apply a Naive Bayes model to new data
34. OPPORTUNITIES FOR PREDICTIVE MODELING
How long the truck can run without maintenance ?
How many drivers can quit in the next 30 days ?
Company turnover in future ?
Accident Prevention ?
Battery life on the trucks ?
Weather alerts to drivers - k-means algorithm?
Run machine learning set up on AIX-Open shift environment (Testing in progress)
35. Q & A
Q & A
srikamani@gmail.com
jssivakumar@hotmail.com