you ship your
organizational
structure
GENERALISTS
SPECIALISTS COMPLEMENTARY
Tree-shaped
SKILLS
T-SHAPED TREE-SHAPED
Slide 18
Slide 18 text
No content
Slide 19
Slide 19 text
Processing
PIPELINE
TEXT DOC
Slide 20
Slide 20 text
Processing
PIPELINE
PART-OF-
SPEECH
TAGGER
NAMED
ENTITY
RECOGNIZER
SYNTACTIC
DEPENDENCY
PARSER
TEXT DOC
Slide 21
Slide 21 text
Processing
PIPELINE
PART-OF-
SPEECH
TAGGER
NAMED
ENTITY
RECOGNIZER
SYNTACTIC
DEPENDENCY
PARSER
TEXT DOC
PERSON
Slide 22
Slide 22 text
Transfer
LEARNING
Slide 23
Slide 23 text
Transfer
LEARNING
TASK-
SPECIFIC
MODEL
TEXT
Slide 24
Slide 24 text
Transfer
LEARNING
TASK-
SPECIFIC
MODEL
TEXT
GENERAL
LANGUAGE
MODEL
Slide 25
Slide 25 text
Transformer
MODELS
accurate and reusable
subnetwork
different workflows:
working at the tensor level
Slide 26
Slide 26 text
No content
Slide 27
Slide 27 text
Problem #1
Local AI startup’s code base
“kind of hard to read”
Matt (25, Senior Engineer): “array[:, ..., :4]
– what does this even mean?”
BREAKING
Slide 28
Slide 28 text
How many
DIMENSIONS?
Slide 29
Slide 29 text
How many
DIMENSIONS?
2
Slide 30
Slide 30 text
How many
DIMENSIONS?
Slide 31
Slide 31 text
How many
DIMENSIONS?
2
Slide 32
Slide 32 text
How many
DIMENSIONS?
Slide 33
Slide 33 text
1
How many
DIMENSIONS?
Slide 34
Slide 34 text
No content
Slide 35
Slide 35 text
No content
Slide 36
Slide 36 text
No content
Slide 37
Slide 37 text
Y: Floats3d
Incompatible return value type
(got "Tuple[Floats3d, Callable[[Any], Any]]",
expected "Tuple[Floats1d, Callable[..., Any]]")
Slide 38
Slide 38 text
No content
Slide 39
Slide 39 text
No content
Slide 40
Slide 40 text
Relu: Relu
Layer outputs type (thinc.types.Floats2d) but
the next layer expects (thinc.types.Ragged) as
an input
Slide 41
Slide 41 text
No content
Slide 42
Slide 42 text
Problem #2
HYPER-
PARAMETERS
WEIGHTS
OTHER
SETTINGS
MODEL
CODE
MACHINE
LEARNING
LIBRARY
Slide 43
Slide 43 text
Problem #2
HYPER-
PARAMETERS
WEIGHTS
OTHER
SETTINGS
MODEL
CODE
MACHINE
LEARNING
LIBRARY
Slide 44
Slide 44 text
THINC.AI
→
Slide 45
Slide 45 text
1
THINC.AI
→
Slide 46
Slide 46 text
1
2
THINC.AI
→
Slide 47
Slide 47 text
1
2
THINC.AI
→
Slide 48
Slide 48 text
1
2
3
THINC.AI
→
Slide 49
Slide 49 text
No content
Slide 50
Slide 50 text
Under
THE HOOD
1
Slide 51
Slide 51 text
Under
THE HOOD
1
2
Slide 52
Slide 52 text
Coming
SOON
Slide 53
Slide 53 text
No content
Slide 54
Slide 54 text
Problem #3 WE NEED A
DATABASE OF COMPANY
ACQUISITIONS WITH
PRICES AND STOCK
TICKERS.
pytorch predict company acquisitions with prices and stock tickers
No results.
OKAY, I'M
ON IT!
Slide 55
Slide 55 text
Microsoft
acquires software
development
platform GitHub
for $7.5 billion
Slide 56
Slide 56 text
Microsoft acquires software development
platform GitHub for $7.5 billion
Slide 57
Slide 57 text
TEXT CLASSIFIER
Microsoft acquires software development
platform GitHub for $7.5 billion
Slide 58
Slide 58 text
TEXT CLASSIFIER
ENTITY RECOGNIZER
Microsoft acquires software development
platform GitHub for $7.5 billion
Slide 59
Slide 59 text
TEXT CLASSIFIER
ENTITY RECOGNIZER
ENTITY LINKER
Microsoft acquires software development
platform GitHub for $7.5 billion
Slide 60
Slide 60 text
TEXT CLASSIFIER
ENTITY RECOGNIZER
ENTITY LINKER
ATTRIBUTE LOOKUP
Microsoft acquires software development
platform GitHub for $7.5 billion
Slide 61
Slide 61 text
TEXT CLASSIFIER
ENTITY RECOGNIZER
ENTITY LINKER
ATTRIBUTE LOOKUP
CURRENCY NORMALIZER
Microsoft acquires software development
platform GitHub for $7.5 billion
Slide 62
Slide 62 text
No content
Slide 63
Slide 63 text
Problem #4
in practice
CODE
DATA
in theory
DATA
CODE
Slide 64
Slide 64 text
Pope Francis visits U.S.
Which is
CORRECT?
P E R S O N
Pope Francis visits U.S.
P E R S O N
Slide 65
Slide 65 text
I love cats .
I hate cats .
Similar
OR NOT?
Slide 66
Slide 66 text
PRODIGY.AI
→
Slide 67
Slide 67 text
PRODIGY.AI
→
Slide 68
Slide 68 text
PRODIGY.AI
→
Slide 69
Slide 69 text
PRODIGY.AI
→
Slide 70
Slide 70 text
PRODIGY.AI
→
Slide 71
Slide 71 text
Effort (training data size, time, experimenting)
Effectiveness (accuracy, quality)
typical
project
Slide 72
Slide 72 text
S W A M P O F
UN C E R TAIN T Y
Effort (training data size, time, experimenting)
Effectiveness (accuracy, quality)
typical
project
Slide 73
Slide 73 text
H I L L O F H O P E
S W A M P O F
UN C E R TAIN T Y
Effort (training data size, time, experimenting)
Effectiveness (accuracy, quality)
typical
project
Slide 74
Slide 74 text
P L A T E A U O F
F R U S T R A T I O N
H I L L O F H O P E
S W A M P O F
UN C E R TAIN T Y
Effort (training data size, time, experimenting)
Effectiveness (accuracy, quality)
typical
project
Slide 75
Slide 75 text
Effort (training data size, time, experimenting)
Effectiveness (accuracy, quality)
typical
project
Slide 76
Slide 76 text
S W A M P O F
UN C E R TAIN T Y
Effort (training data size, time, experimenting)
Effectiveness (accuracy, quality)
typical
project
Slide 77
Slide 77 text
Q U I C K S A N D O F
S U N K C O S T S
S W A M P O F
UN C E R TAIN T Y
Effort (training data size, time, experimenting)
Effectiveness (accuracy, quality)
typical
project
Slide 78
Slide 78 text
Q U I C K S A N D O F
S U N K C O S T S
S W A M P O F
UN C E R TAIN T Y
Effort (training data size, time, experimenting)
Effectiveness (accuracy, quality)
typical
project
when
TO STOP?
Slide 79
Slide 79 text
I T E R A T I V E
W E T L A N D S O F
S L I G H T L Y L E S S
UN C E R TAIN T Y
Effort (training data size, time, experimenting)
Effectiveness (accuracy, quality)
future
project
Slide 80
Slide 80 text
I T E R A T I V E
W E T L A N D S O F
S L I G H T L Y L E S S
UN C E R TAIN T Y
Effort (training data size, time, experimenting)
Effectiveness (accuracy, quality)
future
project
STOP
STOP
Slide 81
Slide 81 text
M E A D O W S
O F S U C C E S S
GO!
I T E R A T I V E
W E T L A N D S O F
S L I G H T L Y L E S S
UN C E R TAIN T Y
Effort (training data size, time, experimenting)
Effectiveness (accuracy, quality)
future
project
STOP
STOP
Slide 82
Slide 82 text
No content
Slide 83
Slide 83 text
Future
OUTLOOK
Slide 84
Slide 84 text
Future
OUTLOOK
lots of developers
generalists & specialists
WHO?
Slide 85
Slide 85 text
Future
OUTLOOK
transfer learning
component pipelines
WHAT?
lots of developers
generalists & specialists
WHO?
Slide 86
Slide 86 text
Future
OUTLOOK
transfer learning
component pipelines
WHAT?
iterative
in-house
HOW?
lots of developers
generalists & specialists
WHO?