The Business of Machine Learning: A Technical Decision Maker's Guide to Communication and Strategy
Autor Josh Holmes, Michael Lanzettaen Limba Engleză Paperback – noi 2020
What
You'll
Learn
- Understand
the
vast
potential
of
machine
learning,
and
how
and
when
to
apply
different
ML
techniques
- Devise
strategies
to
improve
efficiency
and
accuracy
in
your
business
- Get
to
know
your
customers
and
their
specific
needs
through
interpreting
highly
accurate
and
complex
data
- Communicate
more
effectively
with
teams
of
architects
and
data
scientists
as
they
develop
and
deploy
complex
machine
learning
solutions
- Contrast
the
life
cycle
of
a
machine
learning
project
to
a
software
development
project
- Master terms such as “convolutional neural network,” “nonparametric regression,” and “multi-class decision jungle”
Who
This
Book
is
For
Any
technical
business
decision
maker
who
has
to
implement
a
machine
learning
strategy
or
converse
with
data
scientists.
A
basic
level
of
technical
understanding
is
helpful,
but
does
not
have
to
be
specific
to
programming
languages
or
operating
systems.
This
book
is
open
access
under
a
CC
BY
license.
Preț: 246.11 lei
Preț vechi: 307.64 lei
-20% Nou
Puncte Express: 369
Preț estimativ în valută:
47.10€ • 48.93$ • 39.12£
47.10€ • 48.93$ • 39.12£
Carte nepublicată încă
Doresc să fiu notificat când acest titlu va fi disponibil:
Se trimite...
Preluare comenzi: 021 569.72.76
Specificații
ISBN-13: 9781484235423
ISBN-10: 1484235428
Pagini: 235
Dimensiuni: 155 x 235 mm
Ediția:1st ed.
Editura: Apress
Colecția Apress
Locul publicării:Berkeley, CA, United States
ISBN-10: 1484235428
Pagini: 235
Dimensiuni: 155 x 235 mm
Ediția:1st ed.
Editura: Apress
Colecția Apress
Locul publicării:Berkeley, CA, United States
Cuprins
Chapter
1:
What
is
ML:
Why
the
hype
right
now.
(30
pages)
a) Conversation with a Machine Learning Expert
b) Where’s ML being used today?
i) Spam checking, spell check and grammar
ii) Siri, search engines, music selection (Spotify…)
c) Short history of Machine learning dating back to the 1950s
d) What is AI and what’s its relationship to ML
e) Why Machine Learning is Hot Right now
i) Storage is more accessible than ever
ii) Access to compute, especially GPUS, is higher than ever
iii) New algorithms are being created every day
iv) New tooling making things more accessiblef) How is Machine Learning done? i) What’s inside a model? ii) What’s a feature? g) How do data scientists think about feature extraction?
i) This requires domain expertise so a data scientist in the financial space, for example, wouldn't necessarily be effective in machine translation tasks or obstacle avoidance.
h) What are some of the new tools that will help folks access machine learning
Chapter 2: What is DL, how does it differ, why now? (20 pages)
a) Conversation involving Deep Learning
b) Deep learning is a branch of machine learning.
i) Machines do their own feature extractionii) Deep learning has turned intractable problems to tractable c) What’s made this possible at this point? i) Compute, especially GPUs, are more accessible than ever ii) New Math: Back propagation and Gradient Descentd) Constructing a deep network i) Training a modelii) Using activation functionsiii) batch normalization. iv) What is dropout? v) Choosing an optimization function: SGD and Beyondvi) Evaluating a modelvii) What is a SoftMax? Chapter 3: Things that kind of look like AI and solve amazing problems but really aren't... (20 pages)
a) Conversation between a programmer and an ML expert on choosing the right tool for the job
b) Why are these things not actually Machine Learning?i) Who programmed the rules? ii) Deterministic verses Probabilistic resultsc) Ways to solve problems that look like machine learningi) Expert systems – rule based systems including state machines(1) When to use an expert system(2) Pitfalls and drawbacks of an expert systemii) Convex optimization – set of techniques for deterministically finding the optimal resourcing (1) When to use convex optimization(2) Pitfalls and drawbacks to convex optimization iii) Time-series Forecasting - using past data to predict the future, taking into account seasonal effects and short/long-term trends(1) When to use time-series forecasting (2) Pitfalls and drawbacks to using time-series forecastingiv) Dynamic programming - cleverly breaking problems down, solving the easier smaller ones, and storing their solutions(1) When to use dynamic programming(2) Pitfalls and drawbacks of dynamic programming Chapter 4: What sort of problems can you /should you solve with ML? DL? (20 pages)a) Conversation with data scientist around selecting ML toolsb) What’s ML really good at? i) Discussion of problems where ML has helped
c) What’s ML not good at?
i) A walk through a few problems where ML doesn't do very well
ii) Not enough data
iii) Curse of dimensionality
d) Recognizing a machine learning problem
i) Filtering out problems that can be solved by methods in Chapter 3
ii) Phrasing your problem as an ML problem e) When is an ML problem actually a Deep Learning problem? How to know when this answer has changed (things are moving fast!)
Chapter 5: ML: Dealing with Data (featurization) (20 pages)
a) Conversation with a data scientist about featurization of data.b) How do determine what data you need to collectc) How to store this datai) Data store ii) Formats d) How a data scientist works with data
i) Data cleaning
ii) Labeling
iii) Featurization
e) Potential pitfallsi) Feature skew and Heteroskedasticityii) Label skewiii) Interdependence of featuresiv) Outlier detection v) Data sparsity vi) Missing valuesChapter 6: ML Under Supervision: Regression and Classification (20 pages)a) The primary two methods of solving problems with classic machine learning i) Classification turns features into a single decision such as a yes or no or into different buckets. E.G. Will it rain tomorrow? ii) Regression turns features into numeric values. E.G. What’s the temperature likely to be tomorrow?Chapter 7: Unsupervised and Semi-supervised ML (20 pages)a) Conversation with a data scientist about what a computer can do without any direction b) What unsupervised learning can doi) Working with unlabeled data ii) Clustering and finding patterns in dataiii) Anomaly Detection iv) How to recognize an unsupervised learning problemc) How semi-supervised learning can help augment supervised learningi) Combines labeled and unlabeled datad) Moving from unsupervised to supervised i) Creating labels from the clusters Chapter 8: Deep Learning: On Images (CNNs) (20 pages)
a) Conversation with a data scientist about image processing
b) Short history of image recognition
i) How featurization of an image has traditionally been done c) How the introduction of deep learning has accelerated the field
d) How does Deep Learning work with Images
i) Creating image features
ii) What is a convolution iii) what is a convolutional neural network(CNN)
iv) what do width, stride and padding mean to a CNN?
Chapter 9: Deep Learning: On Text and Sound (RNNs) (20 pages)
a) Conversation with a data scientist about text and sound processing
b) Processing sequential data is far different than processing an image i) Sequencing matters
ii) Looking at sequencing
(1) Letters
(2) Words
(3) Phoneme
c) Recurrent Neural Networks
i) Semantic mapping through of sequential data
ii) Vanishing and exploding gradient problems
iii) LSTMs and GRUs
iv) Attention-based methods d) How can this be used?
i) Learn to caption images
ii) Speak like a celebrity
iii) Find information from long documents that would otherwise remain hidden with more traditional ML methods.
Chapter 10: Deep Learning: Self-Xing Y's, or Deep Reinforcement Learning (20 pages) a) Conversation with a data scientist about reinforcement learning
b) Reinforcement Learning
i) Short history of reinforcement learning
(1) old technique being given dramatic new life with Deep Learning.
c) Basics of Q-Learning i) Defining intelligent Q-functions
d) High profile uses
i) defeat the best Go player in the world,
ii) drive cars
iii) fly planes on its own.
Chapter 11: The ML Process: Data Provenance, Model Versioning, Deployment, etc. (30 pages)a) Conversation with a data scientist on the life cycle of a machine learning project. b) Traditional Software Lifecycle verses Machine Learning Software Lifecyclei) Accounting for the data(1) Where the data is coming from(2) How it’s changed over the timeii) Versioning the model itself c) Tracking quality of the model over time
i) Accuracy over time
d) Deployment of the model
i) Where does the model live?
(1) Server (2) On a device
(3) Hybrid
ii) Hierarchical models
(1) Stacked degrees of precision
iii) Model compression/quantization (1) Various techniques for doing this (a) Normalization of values(b) Giving up degrees of precision e) Updating the model
i) Retraining
ii) Redeployment
Chapter 12: Advanced Topics: (20 pages)
a
b) VAEs (Variational Auto Encoders), GANs (Generative Adversarial Networks), Cyclegans?
c) Featurizers, x2vec
d) Ensembling and voting ensembles – multiple models who are ensembled (voting on the regression discussion in chapter 7)
e) Collaborative Filtering and Recommenders
a) Conversation with a Machine Learning Expert
b) Where’s ML being used today?
i) Spam checking, spell check and grammar
ii) Siri, search engines, music selection (Spotify…)
c) Short history of Machine learning dating back to the 1950s
d) What is AI and what’s its relationship to ML
e) Why Machine Learning is Hot Right now
i) Storage is more accessible than ever
ii) Access to compute, especially GPUS, is higher than ever
iii) New algorithms are being created every day
iv) New tooling making things more accessiblef) How is Machine Learning done? i) What’s inside a model? ii) What’s a feature? g) How do data scientists think about feature extraction?
i) This requires domain expertise so a data scientist in the financial space, for example, wouldn't necessarily be effective in machine translation tasks or obstacle avoidance.
h) What are some of the new tools that will help folks access machine learning
Chapter 2: What is DL, how does it differ, why now? (20 pages)
a) Conversation involving Deep Learning
b) Deep learning is a branch of machine learning.
i) Machines do their own feature extractionii) Deep learning has turned intractable problems to tractable c) What’s made this possible at this point? i) Compute, especially GPUs, are more accessible than ever ii) New Math: Back propagation and Gradient Descentd) Constructing a deep network i) Training a modelii) Using activation functionsiii) batch normalization. iv) What is dropout? v) Choosing an optimization function: SGD and Beyondvi) Evaluating a modelvii) What is a SoftMax? Chapter 3: Things that kind of look like AI and solve amazing problems but really aren't... (20 pages)
a) Conversation between a programmer and an ML expert on choosing the right tool for the job
b) Why are these things not actually Machine Learning?i) Who programmed the rules? ii) Deterministic verses Probabilistic resultsc) Ways to solve problems that look like machine learningi) Expert systems – rule based systems including state machines(1) When to use an expert system(2) Pitfalls and drawbacks of an expert systemii) Convex optimization – set of techniques for deterministically finding the optimal resourcing (1) When to use convex optimization(2) Pitfalls and drawbacks to convex optimization iii) Time-series Forecasting - using past data to predict the future, taking into account seasonal effects and short/long-term trends(1) When to use time-series forecasting (2) Pitfalls and drawbacks to using time-series forecastingiv) Dynamic programming - cleverly breaking problems down, solving the easier smaller ones, and storing their solutions(1) When to use dynamic programming(2) Pitfalls and drawbacks of dynamic programming Chapter 4: What sort of problems can you /should you solve with ML? DL? (20 pages)a) Conversation with data scientist around selecting ML toolsb) What’s ML really good at? i) Discussion of problems where ML has helped
c) What’s ML not good at?
i) A walk through a few problems where ML doesn't do very well
ii) Not enough data
iii) Curse of dimensionality
d) Recognizing a machine learning problem
i) Filtering out problems that can be solved by methods in Chapter 3
ii) Phrasing your problem as an ML problem e) When is an ML problem actually a Deep Learning problem? How to know when this answer has changed (things are moving fast!)
Chapter 5: ML: Dealing with Data (featurization) (20 pages)
a) Conversation with a data scientist about featurization of data.b) How do determine what data you need to collectc) How to store this datai) Data store ii) Formats d) How a data scientist works with data
i) Data cleaning
ii) Labeling
iii) Featurization
e) Potential pitfallsi) Feature skew and Heteroskedasticityii) Label skewiii) Interdependence of featuresiv) Outlier detection v) Data sparsity vi) Missing valuesChapter 6: ML Under Supervision: Regression and Classification (20 pages)a) The primary two methods of solving problems with classic machine learning i) Classification turns features into a single decision such as a yes or no or into different buckets. E.G. Will it rain tomorrow? ii) Regression turns features into numeric values. E.G. What’s the temperature likely to be tomorrow?Chapter 7: Unsupervised and Semi-supervised ML (20 pages)a) Conversation with a data scientist about what a computer can do without any direction b) What unsupervised learning can doi) Working with unlabeled data ii) Clustering and finding patterns in dataiii) Anomaly Detection iv) How to recognize an unsupervised learning problemc) How semi-supervised learning can help augment supervised learningi) Combines labeled and unlabeled datad) Moving from unsupervised to supervised i) Creating labels from the clusters Chapter 8: Deep Learning: On Images (CNNs) (20 pages)
a) Conversation with a data scientist about image processing
b) Short history of image recognition
i) How featurization of an image has traditionally been done c) How the introduction of deep learning has accelerated the field
d) How does Deep Learning work with Images
i) Creating image features
ii) What is a convolution iii) what is a convolutional neural network(CNN)
iv) what do width, stride and padding mean to a CNN?
Chapter 9: Deep Learning: On Text and Sound (RNNs) (20 pages)
a) Conversation with a data scientist about text and sound processing
b) Processing sequential data is far different than processing an image i) Sequencing matters
ii) Looking at sequencing
(1) Letters
(2) Words
(3) Phoneme
c) Recurrent Neural Networks
i) Semantic mapping through of sequential data
ii) Vanishing and exploding gradient problems
iii) LSTMs and GRUs
iv) Attention-based methods d) How can this be used?
i) Learn to caption images
ii) Speak like a celebrity
iii) Find information from long documents that would otherwise remain hidden with more traditional ML methods.
Chapter 10: Deep Learning: Self-Xing Y's, or Deep Reinforcement Learning (20 pages) a) Conversation with a data scientist about reinforcement learning
b) Reinforcement Learning
i) Short history of reinforcement learning
(1) old technique being given dramatic new life with Deep Learning.
c) Basics of Q-Learning i) Defining intelligent Q-functions
d) High profile uses
i) defeat the best Go player in the world,
ii) drive cars
iii) fly planes on its own.
Chapter 11: The ML Process: Data Provenance, Model Versioning, Deployment, etc. (30 pages)a) Conversation with a data scientist on the life cycle of a machine learning project. b) Traditional Software Lifecycle verses Machine Learning Software Lifecyclei) Accounting for the data(1) Where the data is coming from(2) How it’s changed over the timeii) Versioning the model itself c) Tracking quality of the model over time
i) Accuracy over time
d) Deployment of the model
i) Where does the model live?
(1) Server (2) On a device
(3) Hybrid
ii) Hierarchical models
(1) Stacked degrees of precision
iii) Model compression/quantization (1) Various techniques for doing this (a) Normalization of values(b) Giving up degrees of precision e) Updating the model
i) Retraining
ii) Redeployment
Chapter 12: Advanced Topics: (20 pages)
a
b) VAEs (Variational Auto Encoders), GANs (Generative Adversarial Networks), Cyclegans?
c) Featurizers, x2vec
d) Ensembling and voting ensembles – multiple models who are ensembled (voting on the regression discussion in chapter 7)
e) Collaborative Filtering and Recommenders
Notă biografică
Josh
Holmesis
CTO
of
the
Commercial
Software
Engineering
Americas
team
at
Microsoft.
Prior
to
joining
Microsoft,
Josh
consulted
for
a
variety
of
clients
ranging
from
large
Fortune
500
firms
to
startups.
Josh
speaks
and
presents
globally
on
the
topics
of
IoT
and
machine
learning.
A
tireless
and
passionate
advocate
for
the
tech
community,
Josh
has
founded
and/or
run
numerous
organizations,
including
the
Great
Lakes
Area
.NET
Users
Group
and
the
Ann
Arbor
Computer
Society.
He
was
also
on
the
forming
committee
for
CodeMash.
You
can
contact
Josh
through
his
blog.
Mike
Lanzettadesigns
and
implements
machine
learning
solutions
for
Fortune
500
companies
at
Microsoft.
He
has
been
doing
software
development
for
more
than
20
years,
working
at
four-person
startups
to
Amazon.
His
experience
runs
the
gamut
from
electronic
circuit
design,
travel
optimization,
and
drug
discovery
to
demand
forecasting
at
Amazon
and
machine
learning
at
Microsoft.
Mike
regularly
presents
and
chairs
at
conferences
nationally
and
internationally.
He
has
an
M.Sc.
in
CSE
from
UW
and
a
B.Sc.
in
CE
from
UCSC.
He
is
often
found
blogging
or
tweeting
on
the
topic
of
machine
learning.
Textul de pe ultima copertă
Successfully
and
proactively
take
charge
of
your
machine
learning
strategy. Machine
learning
(ML)
is
permeating
every
sector
and
aspect
of
business,
from
evaluating
the
success
of
a
massive
online
marketing
campaign,
to
predicting
insurance
payouts,
to
crime
scene
analysis.
This
book
shows
how
to
interpret
patterns
and
redundancies
from
massive
amounts
of
existing
data
to
help
your
business
cut
costs,
operate
more
efficiently
and
effectively,
and
get
to
the
next
level.
You
will
learn
how
to
analyze,
communicate,
and
launch
a
viable
program
that,
when
done
correctly,
will
positively
transform
your
business.
The
authors
engage
you
to
experience
the
business
of
machine
learning
through
actual
conversations
that
open
with
an
exchange
between
a
data
scientist
and
and
his
counterpart
in
business,
the
technical
decision
maker.
You
will
learn
where
to
go
when
the
conversation
leads
to
an
impasse
and
work
step-by-step
to
methodically
resolve
the
challenges.
After
reading
this
book,
you
will
come
away
with
the
confidence
to
tackle
a
machine
learning
strategy
customized
for
your
team
or
business
objectives.
Revel
in
the
vast
capabilities
of
machine
learning
tools
at
your
disposal
and
reach
that
"a
ha"
moment
when
you
discover
the
profound
and
enduring
impact
machine
learning
can
have
on
your
business.
What
You'll
Learn:
- Understand
the
vast
potential
of
machine
learning,
and
how
and
when
to
apply
different
ML
techniques
- Devise
strategies
to
improve
efficiency
and
accuracy
in
your
business
- Get
to
know
your
customers
and
their
specific
needs
through
interpreting
highly
accurate
and
complex
data
- Communicate
more
effectively
with
teams
of
architects
and
data
scientists
as
they
develop
and
deploy
complex
machine
learning
solutions
- Contrast
the
life
cycle
of
a
machine
learning
project
to
a
software
development
project
- Master terms such as “convolutional neural network,” “nonparametric regression,” and “multi-class decision jungle”
This
book
is
forany
technical
business
decision
maker
who
has
to
implement
a
machine
learning
strategy
or
converse
with
data
scientists.
A
basic
level
of
technical
understanding
is
helpful,
but
does
not
have
to
be
specific
to
programming
languages
or
operating
systems.
This
book
is
open
access
under
a
CC
BY
license.
Josh
Holmesis
CTO
of
the
Commercial
Software
Engineering
Americas
team
at
Microsoft.
Prior
to
joining
Microsoft,
Josh
consulted
for
a
variety
of
clients
ranging
from
large
Fortune
500
firms
to
startups.
Josh
speaks
and
presents
globally
on
the
topics
of
IoT
and
machine
learning.
A
tireless
and
passionate
advocate
for
the
tech
community,
Josh
has
founded
and/or
run
numerous
organizations,
including
the
Great
Lakes
Area
.NET
Users
Group
and
the
Ann
Arbor
Computer
Society.
Mike
Lanzettadesigns
and
implements
machine
earning
solutions
for
Fortune
500
companies
at
Microsoft.
He
has
been
doing
software
development
for
more
than
20
years,
working
at
four-person
startups
to
Amazon.
Mike
regularly
presents
and
chairs
at
conferences
nationally
and
internationally.
He
has
an
M.Sc.
in
CSE
from
UW
and
a
B.Sc.
in
CE
from
UCSC.
He
is
often
found
blogging
or
tweeting
on
the
topic
of
machine
learning.
Caracteristici
Dives
into
the
business
value
of
machine
learning
in
order
to
understand
its
game
changing
potential
for
your
business
Explains
the
terms
of
this
complex,
math-based
technology,
in
order
to
contribute
and
participate
in
the
conversations
with
data
scientists
that
will
allow
you
to
set
strategy
Written by ML experts Josh Holmes and Mike Lanzetta who work for Microsoft where they tag team ML solutions for Fortune 500 companies and speak globally on the topic of machine learning
Written by ML experts Josh Holmes and Mike Lanzetta who work for Microsoft where they tag team ML solutions for Fortune 500 companies and speak globally on the topic of machine learning