An updated introduction to Machine Learning and AI: basic concepts, linear regression example, neural networks and deep learning basics, intuitive approach to AI and Machine Learning, AutoML, AI demystified, Algorithms, ML tech stack, additional resources
2. About your instructor
• Nerd (Engineer in the 90s - Commodore 64,
1MHz CPU, 64Kb RAM - Assembler, C)
• turned into Business (Corporate Exec)
• turned Entrepreneur (still shareholder)
• turned into Investing (VC and PE)
• turned into Boards (Non-Exec Board Director)
• turned into Teaching (Northeastern University,
Berkeley Center for Entrepreneurship & Tech,
ISDI)
Superskill: I can spin a book/basketball/pizza in
the air on the tip of my finger
3. The Next 10 Years : Tech Trends
1 Artificial Intelligence Above All Things Her, Ex Machina
2 Cybersecurity All Things Hacked Mr Robot
3 Metaverse All Things Virtual
Ready Player I,
Avatar
4 Spatial Computing All Things Online
IronMan, Black
Mirror
5 Autonomous Driving & Flying All Things Autonomous
Minority Report, I
Robot, Blade
Runner
6 Biotechnology All Beings CRISPR Edited Gattaca
7 Longevity All Things Anti-aging Mr Nobody
8 Scarcity
All Things Exhausting Earth
resources
Mad Max
9 De-carbonization All Things Green Avatar
10 Off world All Things Space
The Martian, 2001 a
space odissey
(bonus)
11
End of the Attention Economy All Things Ad-blocking The Social Network
4. The Next 10 Years – Recommended Books:
• Thinking Machines, L. Dormehl : AI explained, down to earth, no math
• Outliers, M.Gladwell : About outstanding people and their stories
• The 4th Industrial Revolution: K.Schwab : technology changes &
disruption
• The Innovator’s Dilemma, C.Christensen : on disruption, innovation and
how to create Unicorns, a must for new entrepreneurs
• *The Selfish Gene, R.Dawkins : How we got here, Why are we here,
Where are we going, the explanation to life and everything else
• **The Hitchhiker's Guide to the Galaxy, D.Adams : Earth is demolished to
accommodate an intergalactic super-highway, absurdly funny
*Beware: this book will profoundly change how you view and understand the world, and yes, it holds
the answers to the main questions in life: why are we here, where are we going, why things are how they
are
**This book, also known as H2G2, holds also the “answer to life, the universe and everything” (being
it the nr 42), coming as a result of a 7,5 Mn years computing project commissioned to a super-computer
called Deep Thought in a distant planet
https://simple.wikipedia.org/wiki/42_(answer)
5. Why is AI important?
• New Technologies come in waves (adoption waves)
• They shape our culture, society and even us as individuals
• Each wave creates change, disruption and new wealth
• Your future as a professional will be inevitably linked to one
or more of these waves
• You or [insert option according to age: | your degree | your
startup | the corporation you’ll work for | the company you’ll
invest in |] will inevitably specialize according to these waves
6. The term AI (Artificial Intelligence) is
used across this presentation and
appears in 3rd party materials and
references used.
In the context of this session, it refers
specifically to the ability to build
machine learning driven applications
which ultimately automate and/or
optimize business processes and,
It DOES NOT refer to general or strong
Artificial Intelligence in the formal
sense, which is not likely to happen for
decades to come (emphasis from
author)
What’s AI and what’s NOT: demistifying AI
US
15. • Machines participate in
decisión process
• Data is structured &
purpose driven
• Time: model training &
application
• Process integration in
place
• Data volumen is high or
very high
20. The New Software Paradigm
Machine Learning: an intuitive approach - building an apple recommender
if color (apple) = red then
if size (apple) = big then
if vendor (apple) = trusted and
origin (apple) = California or France
then pick (apple) else
discard (apple)
color size vendor origin action
red big trusted California pick
green small trusted Italy discard
light red medium unverified Canada discard
red big trusted France pick
Traditional Rule Based
Programming
Machine Learning - Data Driven
Rule Discovery
21. Diabetes Prediction
if (plasma_glucose <= 166):
if (blood_pressure is None):
return u'true'
if (blood_pressure > 61):
if (age is None):
return u'true'
if (age > 31):
if (bmi > 30.115):
if (blood_pressure > 91):
return u'true'
if (blood_pressure <= 91):
if (bmi > 30.615):
if (age > 33):
if (bmi > 31.45):
if (pregnancies is None):
return u'true'
if (pregnancies > 7):
if (age > 41):
return u'true'
if (age <= 41):
if (pregnancies > 10):
if (bmi > 41.2):
return u'true'
if (diabetes_pedigree <= 1.11425):
if (diabetes_pedigree > 0.3065):
return u'true'
if (diabetes_pedigree <= 0.3065):
if (bmi > 35.3):
return u'true'
if (bmi <= 35.3):
if (plasma_glucose > 177):
if (plasma_glucose > 192):
return u'true'
if (plasma_glucose <= 192):
30. Where does the Learning happen?
We learn the Coefficients, aka model parameters
{ Y = Bias + Coeff(1)* X(1) + Coeff(2)* X(2)+ … + Coeff(n)* X(n) }
31. SUPERVISED UNSUPERVISED
DATA Requires “labelled” data Does not require “labelled” data
GOAL
Goal is to predict the label often called the objective (churn,
sales predictions, etc).
Goal is “structure discovery”, with
algorithms focused on type of relation
(clustering, etc.)
EVALUATION Predictions can be compared to real labels
Each algorithm has it’s own quality
measures
ALGORITHMS
Algorithms
CLUSTER ANOMALY
TOPIC
MODEL
ASSOCIATION
TREE
MODEL
ENSEMBLE NEURAL
NETWORKS
LOGISTIC
REGRESSION
TIME SERIES
CLASSIFICATION / REGRESSION
AUTOML
LINEAR
REGRESSION
35. Denis Dmitriev - Neural Networks in 3D
https://www.youtube.com/watch?v=3JQ3hYko51Y
36. demos
3D simulation of Neural Networks (revealing!)
https://www.youtube.com/watch?v=3JQ3hYko51Y&a
b_channel=DenisDmitriev
Tensorflow Neural Network playground (challenge
yourself!)
link
Google AI experiments (fun!)
https://experiments.withgoogle.com/collection/ai
37. Where are my models? ML technology stack
BIGML INFRASTRUCTURE
• Models are stored in the BigML server, in the cloud.
• Private and On premises clouds are also available.
• Resources are unmutable, any change will result
into a new resource.
• Resources are encoded in JSON and are easy to
export.
API-first, auto-scalable, auto-deployable
distributed architecture for Machine Learning
38. How can I improve my model?
AUTOMATIC OPTIMIZATION: AutoML
AUTOMATIC OPTIMIZATION and Model
selection: evaluating multiple models with
different configurations using Bayesian
parameter optimization.
https://blog.bigml.com/2018/05/16/optiml-the-nitty-gritty/
https://static.bigml.com/pdf/BigML_OptiML.pdf?ver=79e
b166
39. How can I improve my model?
AUTOMATIC OPTIMIZATION: AutoML
AUTOMATIC OPTIMIZATION and Model
selection: evaluating multiple models with
different configurations using Bayesian
parameter optimization.
https://blog.bigml.com/2018/05/16/optiml-the-nitty-gritty/
https://static.bigml.com/pdf/BigML_OptiML.pdf?ver=79e
b166
40. How can I improve my model?
MODEL PARAMETERS: AutoML - Automatic Optimization
41. ML Onboarding Strategy
A few tips for machine learning success
• Think of Machine Learning as the ultimate optimization tool,
to use almost in every company process.
• Leverage ML tools enabling domain experts & non-data
scientists to apply machine learning with minimal or no
coding.
• Begin with MLaaS: starting in the cloud is inexpensive.
• Initially consider educating management and key personnel
in AI & machine learning vs building separate teams.
• Plan for scale: many models, many predictions, many more
internal & external users - evaluate need for a ML platform
42. Closing Remarks
• Of all tech waves for the next decade, AI is probably the most pervasive
• True AI does not exist today (nor will it for many years to come), we do
have however disruptive automation capabilities enabled by ML models
• Developing AI/ML driven SW digresses significantly from traditional SW
development processes, a different architecture and SW stack is required
• Like we have seen in the traditional SW industry, Open Source presents
significant challenges once ML ops scale (call out to AI/ML platforms)
• AI Software and ML models are only as good as their capacity to be
deployed in production and retrained (model drift)
43. Resources
• CS229 Machine Learning, Stanford
https://docs.google.com/spreadsheets/d/1OEsqqhihH-
n2OPHsT8jSA8BkLdqUMWY-GiWHgkBs3Z8/edit#gid=0
• Full Stack Deep Learning course https://fullstackdeeplearning.com/course/
• Deep Learning https://www.deeplearningbook.org/
• Machine Learning Mastery, Jason Brownlee
https://machinelearningmastery.com/
• BigML: ML platform, sign up for the academic program free PRO subscription
with your .edu email https://bigml.com/education/
• Short educational videos, all about ML https://bigml.com/education/videos