Chennai Mathematical Institute, in association with Spotle.ai, presents the Artificial Intelligence Masterclass for managers and business heads. The deck gives an overview of AI, the technology landscape and its business applications.
2. Typical situations
• Extrapolate from historicaldata
• From model test scores, predict board exam performance
• Should we give this customer a loan?
• Do these symptoms indicate H1N1?
• Look for patterns in existing data
• Readymade clothes — is S, M, L enough? Add XL?
• Should we sell a combo plan with voice calls plusSMS?
Supervised
learning
Unsupervised
learning
3. Anomaly detection
Credit card fraud
• Monitorregular transactions
• Location, amount, time, items
purchased
• Flag anomalies
• Proactively block card
• Customer dissatisfaction (false
positives) vs loss due to fraud
4. Failure prediction
Major printer manufacturer
• Drum failure
• Down time
• Servicing cost — assign support team on the fly
• Track diagnostic codes to predict failure
• Advance warning, better customer experience
• Schedule support visit efficiently, save cost
5. Recommendation systems
Netflix
• $1 millionprize to improve their in-house
algorithm to recommend movies
• Won by BellKor’s Pragmatic Chaos — two
teams from Bell Core, merged
• When DVD rental was the main business
• Still relevant for online streaming?
• Audience preferences drive content
development
6. Use of ML is exploding
• Online advertising
• E-commerce recommendations
• Screening applications
• Fraud detection
• Conversationbots
• Smart buildings
• …
• Self driving cars
10. Programming for ML
• Open source languages like R and Python
• Built in libraries
• Classical statisticaltests, time-series
analysis, classification,clustering, …
• ML models
• User friendly IDEs
11. Neural network in Python + Tensorflow
model = keras.Sequential([
keras.layers.Flatten(input_shape=(28,28)),
keras.layers.Dense(128, activation=tf.nn.relu),
keras.layers.Dense(10, activation=tf.nn.softmax)
])
model.compile(optimizer=tf.train.AdamOptimizer(),
loss='sparse_categorical_crossentropy',
metrics=[‘accuracy’])
model.fit(train_images,train_labels,epochs=5)
12. Below the hood …
• Need some understanding of
what lies beneath
• Analogy:database systems
• Move from customised data
storage to RDBMS
• Standard model
• Declarative query
language
• Still need to understand table
design, query optimisation
14. The challenge
• When do we deploy AI?
• Problem structure?
• Domain?
• If AI, what approach?
• Supervised or unsupervised?
• Which model?
15. Where AI can help
• AI solutions are driven by data
• No obvious“direct” algorithm
• Labelled data
• This bus route meets timing
requirement
• These test readings indicate
disease
• This picture shows a stress
crack
16. All data is not born equal
• Every business generates data
• How can it improve the
business?
• How usable is the data?
• Clean? Labelled?
• Is the actual data available?
• Privacy issues, silos
• IT vs ML solution
• IT — columns of table
• ML — rows of table
17. Which ML approach to use?
• Supervised or unsupervised?
• Flag good vs bad
• Segment data
• Which model to use?
• Is data categorical or continuous?
• Static or time-varying, seasonal?
• No magic formula
• Familiaritywith different
approaches
• Experience
21. ML everywhere
• Data driven solutions have wide
applicability
• Need business understanding
• Numerous off-the-shelf tools
• Python, R, Tensorflow
• Deployment is easy
• Choosing ML solution not easy
• Variety of models
• Need expertise to tune
parameters