Structured and Unstructured Spaces Guy Van den Broeck DeLBP Aug - - PowerPoint PPT Presentation

PSDDs for Tractable Learning in Structured and Unstructured Spaces

Guy Van den Broeck

DeLBP Aug 18, 2017

Probabilistic Sentential Decision Diagrams

Doga Kisa, Guy Van den Broeck, Arthur Choi and Adnan Darwiche KR, 2014

Learning with Massive Logical Constraints

Doga Kisa, Guy Van den Broeck, Arthur Choi and Adnan Darwiche ICML LTPM workshop, 2014

Tractable Learning for Structured Probability Spaces

Arthur Choi, Guy Van den Broeck and Adnan Darwiche IJCAI, 2015

Tractable Learning for Complex Probability Queries

Jessa Bekker, Jesse Davis, Arthur Choi, Adnan Darwiche, Guy Van den Broeck. NIPS, 2015

Learning the Structure of PSDDs

Yitao Liang, Jessa Bekker and Guy Van den Broeck UAI, 2017

Towards Compact Interpretable Models: Learning and Shrinking PSDDs

Yitao Liang and Guy Van den Broeck IJCAI XAI workshop, 2017

References

(P)SDDs in Melbourne

• Sunday: Logical Foundations for Uncertainty and

Machine Learning Workshop

– Adnan Darwiche: “On the Role of Logic in Probabilistic Inference and Machine Learning” – YooJung Choi: “Optimal Feature Selection for Decision Robustness in Bayesian Networks”

• Sunday: Explainable AI Workshop

– Yitao Liang: “Towards Compact Interpretable Models: Learning and Shrinking PSDDs”

• Tuesday: IJCAI

– YooJung Choi (again)

Structured vs. unstructured probability spaces?

Courses:

• Logic (L)
• Knowledge Representation (K)
• Probability (P)
• Artificial Intelligence (A)

Data

Running Example

Courses:

• Logic (L)
• Knowledge Representation (K)
• Probability (P)
• Artificial Intelligence (A)

Data

• Must take at least one of

Probability or Logic.

• Probability is a prerequisite for AI.
• The prerequisites for KR is

either AI or Logic.

Constraints

Running Example

L K P A 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

unstructured

Probability Space

L K P A 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

unstructured

L K P A 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

structured

Structured Probability Space

7 out of 16 instantiations are impossible

• Must take at least one of

Probability or Logic.

• Probability is a prerequisite for AI.
• The prerequisites for KR is

either AI or Logic.

Learning with Constraints

Data Constraints

(Background Knowledge) (Physics)

Statistical Model

(Distribution)

Learn

Learning with Constraints

Learn a statistical model that assigns zero probability to instantiations that violate the constraints.

Data Constraints

(Background Knowledge) (Physics)

Statistical Model

(Distribution)

Learn

Example: Video

[Lu, W. L., Ting, J. A., Little, J. J., & Murphy, K. P. (2013). Learning to track and identify players from broadcast sports videos.]

Example: Video

[Lu, W. L., Ting, J. A., Little, J. J., & Murphy, K. P. (2013). Learning to track and identify players from broadcast sports videos.]

Example: Robotics

[Wong, L. L., Kaelbling, L. P., & Lozano-Perez, T., Collision-free state estimation. ICRA 2012]

Example: Robotics

[Wong, L. L., Kaelbling, L. P., & Lozano-Perez, T., Collision-free state estimation. ICRA 2012]

Example: Language

• Non-local dependencies:

At least one verb in each sentence

[Chang, M., Ratinov, L., & Roth, D. (2008). Constraints as prior knowledge],…, [Chang, M. W., Ratinov, L., & Roth, D. (2012). Structured learning with constrained conditional models.], [https://en.wikipedia.org/wiki/Constrained_conditional_model]

Example: Language

• Non-local dependencies:

At least one verb in each sentence

• Sentence compression

If a modifier is kept, its subject is also kept

[Chang, M., Ratinov, L., & Roth, D. (2008). Constraints as prior knowledge],…, [Chang, M. W., Ratinov, L., & Roth, D. (2012). Structured learning with constrained conditional models.], [https://en.wikipedia.org/wiki/Constrained_conditional_model]

Example: Language

• Non-local dependencies:

At least one verb in each sentence

• Sentence compression

If a modifier is kept, its subject is also kept

• Information extraction

[Chang, M., Ratinov, L., & Roth, D. (2008). Constraints as prior knowledge],…, [Chang, M. W., Ratinov, L., & Roth, D. (2012). Structured learning with constrained conditional models.], [https://en.wikipedia.org/wiki/Constrained_conditional_model]

Example: Language

• Non-local dependencies:

At least one verb in each sentence

• Sentence compression

If a modifier is kept, its subject is also kept

• Information extraction
• Semantic role labeling
• … and many more!

[Chang, M., Ratinov, L., & Roth, D. (2008). Constraints as prior knowledge],…, [Chang, M. W., Ratinov, L., & Roth, D. (2012). Structured learning with constrained conditional models.], [https://en.wikipedia.org/wiki/Constrained_conditional_model]

Example: Deep Learning

[Graves, A., Wayne, G., Reynolds, M., Harley, T., Danihelka, I., Grabska-Barwińska, A., et al.. (2016). Hybrid computing using a neural network with dynamic external memory. Nature, 538(7626), 471-476.]

Example: Deep Learning

[Graves, A., Wayne, G., Reynolds, M., Harley, T., Danihelka, I., Grabska-Barwińska, A., et al.. (2016). Hybrid computing using a neural network with dynamic external memory. Nature, 538(7626), 471-476.]

Example: Deep Learning

[Graves, A., Wayne, G., Reynolds, M., Harley, T., Danihelka, I., Grabska-Barwińska, A., et al.. (2016). Hybrid computing using a neural network with dynamic external memory. Nature, 538(7626), 471-476.]

Example: Deep Learning

[Graves, A., Wayne, G., Reynolds, M., Harley, T., Danihelka, I., Grabska-Barwińska, A., et al.. (2016). Hybrid computing using a neural network with dynamic external memory. Nature, 538(7626), 471-476.]

What are people doing now?

• Ignore constraints
• Handcraft into models
• Use specialized distributions
• Find non-structured encoding
• Try to learn constraints
• Hack your way around

What are people doing now?

• Ignore constraints
• Handcraft into models
• Use specialized distributions
• Find non-structured encoding
• Try to learn constraints
• Hack your way around

Accuracy ? Specialized skill ? Intractable inference ? Intractable learning ? Waste parameters ? Risk predicting out of space ? you are on your own 

+

Structured Probability Spaces

• Everywhere in ML!

– Configuration problems, inventory, video, text, deep learning – Planning and diagnosis (physics) – Causal models: cooking scenarios (interpreting videos) – Combinatorial objects: parse trees, rankings, directed acyclic graphs, trees, simple paths, game traces, etc.

Structured Probability Spaces

• Everywhere in ML!

– Configuration problems, inventory, video, text, deep learning – Planning and diagnosis (physics) – Causal models: cooking scenarios (interpreting videos) – Combinatorial objects: parse trees, rankings, directed acyclic graphs, trees, simple paths, game traces, etc.

• Some representations: constrained conditional

models, mixed networks, probabilistic logics.

Structured Probability Spaces

• Everywhere in ML!

– Configuration problems, inventory, video, text, deep learning – Planning and diagnosis (physics) – Causal models: cooking scenarios (interpreting videos) – Combinatorial objects: parse trees, rankings, directed acyclic graphs, trees, simple paths, game traces, etc.

• Some representations: constrained conditional

models, mixed networks, probabilistic logics.

No statistical ML boxes out there that take constraints as input! 

Structured Probability Spaces

• Everywhere in ML!

– Configuration problems, inventory, video, text, deep learning – Planning and diagnosis (physics) – Causal models: cooking scenarios (interpreting videos) – Combinatorial objects: parse trees, rankings, directed acyclic graphs, trees, simple paths, game traces, etc.

• Some representations: constrained conditional

models, mixed networks, probabilistic logics.

No statistical ML boxes out there that take constraints as input! 

Goal: Constraints as important as data! General purpose!

Specification Language: Logic

L K P A 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

unstructured

L K P A 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

structured

Structured Probability Space

7 out of 16 instantiations are impossible

• Must take at least one of

Probability or Logic.

• Probability is a prerequisite for AI.
• The prerequisites for KR is

either AI or Logic.

L K P A 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

unstructured

L K P A 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

structured

Boolean Constraints

7 out of 16 instantiations are impossible

Combinatorial Objects: Rankings

10 items: 3,628,800 rankings 20 items: 2,432,902,008,176,640,000 rankings

rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10 cucumber roll rank sushi 1 shrimp 2 sea urchin 3 salmon roe 4 fatty tuna 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10 cucumber roll

Combinatorial Objects: Rankings

rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10 cucumber roll rank sushi 1 shrimp 2 sea urchin 3 salmon roe 4 fatty tuna 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10 cucumber roll

Aij item i at position j (n items require n2 Boolean variables)

Combinatorial Objects: Rankings

rank sushi 1 fatty tuna 2 sea urchin 3 salmon roe 4 shrimp 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10 cucumber roll rank sushi 1 shrimp 2 sea urchin 3 salmon roe 4 fatty tuna 5 tuna 6 squid 7 tuna roll 8 see eel 9 egg 10 cucumber roll

Aij item i at position j (n items require n2 Boolean variables)

An item may be assigned to more than one position A position may contain more than one item

Encoding Rankings in Logic

Aij : item i at position j

pos 1 pos 2 pos 3 pos 4 item 1 A11 A12 A13 A14 item 2 A21 A22 A23 A24 item 3 A31 A32 A33 A34 item 4 A41 A42 A43 A44

Encoding Rankings in Logic

Aij : item i at position j

pos 1 pos 2 pos 3 pos 4 item 1 A11 A12 A13 A14 item 2 A21 A22 A23 A24 item 3 A31 A32 A33 A34 item 4 A41 A42 A43 A44

constraint: each item i assigned to a unique position (n constraints)

Encoding Rankings in Logic

Aij : item i at position j

pos 1 pos 2 pos 3 pos 4 item 1 A11 A12 A13 A14 item 2 A21 A22 A23 A24 item 3 A31 A32 A33 A34 item 4 A41 A42 A43 A44

constraint: each item i assigned to a unique position (n constraints) constraint: each position j assigned a unique item (n constraints)

Encoding Rankings in Logic

Aij : item i at position j

pos 1 pos 2 pos 3 pos 4 item 1 A11 A12 A13 A14 item 2 A21 A22 A23 A24 item 3 A31 A32 A33 A34 item 4 A41 A42 A43 A44

constraint: each item i assigned to a unique position (n constraints) constraint: each position j assigned a unique item (n constraints)

Structured Space for Paths

• cf. Nature paper

Structured Space for Paths

• cf. Nature paper

Good variable assignment (represents route) 184

Structured Space for Paths

• cf. Nature paper

Good variable assignment (represents route) 184 Bad variable assignment (does not represent route) 16,777,032

Structured Space for Paths

• cf. Nature paper

Good variable assignment (represents route) 184 Bad variable assignment (does not represent route) 16,777,032

Space easily encoded in logical constraints  See [Choi, Tavabi, Darwiche, AAAI 2016]

Unstructured probability space: 184+16,777,032 = 224

Structured Space for Paths

• cf. Nature paper

Good variable assignment (represents route) 184 Bad variable assignment (does not represent route) 16,777,032

Space easily encoded in logical constraints  See [Choi, Tavabi, Darwiche, AAAI 2016]

“Deep Architecture”

Logic + Probability

• L K

L  P A P  L

• L 
• P A

P

• L K

L  P

• P 

K K A A A A

Logical Circuits

• L K

L  P A P  L

• L 
• P A

P

• L K

L  P

• P 

K K A A A A

Property: Decomposability

• L K

L  P A P  L

• L 
• P A

P

• L K

L  P

• P 

K K A A A A

Property: Decomposability

• L K

L  P A P  L

• L 
• P A

P

• L K

L  P

• P 

K K A A A A

Input: L, K, P, A

Property: Determinism

• L K

L  P A P  L

• L 
• P A

P

• L K

L  P

• P 

K K A A A A

Input: L, K, P, A

Sentential Decision Diagram (SDD)

• L K

L  P A P  L

• L 
• P A

P

• L K

L  P

• P 

K K A A A A

Input: L, K, P, A

Sentential Decision Diagram (SDD)

• L K

L  P A P  L

• L 
• P A

P

• L K

L  P

• P 

K K A A A A

Input: L, K, P, A

Sentential Decision Diagram (SDD)

Tractable for Logical Inference

• Is structured space empty? (SAT)
• Count size of structured space (#SAT)
• Check equivalence of spaces

Tractable for Logical Inference

• Is structured space empty? (SAT)
• Count size of structured space (#SAT)
• Check equivalence of spaces
• Algorithms linear in circuit size 

(pass up, pass down, similar to backprop)

Tractable for Logical Inference

• Is structured space empty? (SAT)
• Count size of structured space (#SAT)
• Check equivalence of spaces
• Algorithms linear in circuit size 

(pass up, pass down, similar to backprop)

• L K

L 

1

P A P 

1

L

• L 

1

• P A

P

0.6 0.4

• L K

L 

1

P

• P 

1

K K

0.8 0.2

A A

0.25 0.75

A A

0.9 0.1 0.1 0.6 0.3

PSDD: Probabilistic SDD

• L K

L 

1

P A P 

1

L

• L 
• P A

P

0.6

• L K

L 

1

P

• P 

1

K K

0.2

A A

0.75

A A

0.9 0.1 0.1 0.6

Input: L, K, P, A

PSDD: Probabilistic SDD

1 0.3 0.4 0.8 0.25

• L K

L 

1

P A P 

1

L

• L 

1.0

• P A

P

0.6 0.4

• L K

L 

1

P

• P 

1

K K

0.8 0.2

A A

0.25 0.75

A A

0.9 0.1 0.1 0.6 0.3

Input: L, K, P, A

PSDD: Probabilistic SDD

• L K

L 

1

P A P 

1

L

• L 

1.0

• P A

P

0.6 0.4

• L K

L 

1

P

• P 

1

K K

0.8 0.2

A A

0.25 0.75

A A

0.9 0.1 0.1 0.6 0.3

Input: L, K, P, A Pr(L,K,P,A) = 0.3 x 1.0 x 0.8 x 0.4 x 0.25 = 0.024

PSDD: Probabilistic SDD

• L K

L 

1

P A P 

1

L

• L 

1

• P A

P

0.6 0.4

• L K

L 

1

P

• P 

1

A A

0.8 0.2

A A

0.25 0.75

A A

0.9 0.1 0.1 0.6 0.3

PSDD nodes induce a normalized distribution!

• L K

L 

1

P A P 

1

L

• L 

1

• P A

P

0.6 0.4

• L K

L 

1

P

• P 

1

A A

0.8 0.2

A A

0.25 0.75

A A

0.9 0.1 0.1 0.6 0.3

PSDD nodes induce a normalized distribution!

• L K

L 

1

P A P 

1

L

• L 

1

• P A

P

0.6 0.4

• L K

L 

1

P

• P 

1

A A

0.8 0.2

A A

0.25 0.75

A A

0.9 0.1 0.1 0.6 0.3

Can read probabilistic independences off the circuit structure

PSDD nodes induce a normalized distribution!

Tractable for Probabilistic Inference

• MAP inference: Find most-likely assignment

(otherwise NP-complete)

• Computing conditional probabilities Pr(x|y)

(otherwise PP-complete)

• Sample from Pr(x|y)

Tractable for Probabilistic Inference

• MAP inference: Find most-likely assignment

(otherwise NP-complete)

• Computing conditional probabilities Pr(x|y)

(otherwise PP-complete)

• Sample from Pr(x|y)
• Algorithms linear in circuit size 

(pass up, pass down, similar to backprop)

Learning PSDDs

Logic + Probability + ML

• L K

L 

1

P A P 

1

L

• L 

1

• P A

P

0.6 0.4

• L K

L 

1

P

• P 

1

K K

0.8 0.2

A A

0.25 0.75

A A

0.9 0.1 0.1 0.6 0.3

Parameters are Interpretable

Explainable AI DARPA Program

• L K

L 

1

P A P 

1

L

• L 

1

• P A

P

0.6 0.4

• L K

L 

1

P

• P 

1

K K

0.8 0.2

A A

0.25 0.75

A A

0.9 0.1 0.1 0.6 0.3

Student takes course L

Parameters are Interpretable

Explainable AI DARPA Program

• L K

L 

1

P A P 

1

L

• L 

1

• P A

P

0.6 0.4

• L K

L 

1

P

• P 

1

K K

0.8 0.2

A A

0.25 0.75

A A

0.9 0.1 0.1 0.6 0.3

Student takes course L Student takes course P

Parameters are Interpretable

Explainable AI DARPA Program

• L K

L 

1

P A P 

1

L

• L 

1

• P A

P

0.6 0.4

• L K

L 

1

P

• P 

1

K K

0.8 0.2

A A

0.25 0.75

A A

0.9 0.1 0.1 0.6 0.3

Student takes course L Student takes course P Probability of P given L

Parameters are Interpretable

Explainable AI DARPA Program

Learning Algorithms

• Parameter learning:

Closed form max likelihood from complete data One pass over data to estimate Pr(x|y)

Learning Algorithms

• Parameter learning:

Closed form max likelihood from complete data One pass over data to estimate Pr(x|y) Not a lot to say: very easy!

Learning Algorithms

• Parameter learning:

Closed form max likelihood from complete data One pass over data to estimate Pr(x|y)

• Circuit learning (naïve):

Compile constraints to SDD circuit – Use SAT solver technology Circuit does not depend on data Not a lot to say: very easy!

Learning Preference Distributions

Special-purpose distribution: Mixture-of-Mallows

– # of components from 1 to 20 – EM with 10 random seeds – implementation of Lu & Boutilier PSDD

Learning Preference Distributions

Special-purpose distribution: Mixture-of-Mallows

– # of components from 1 to 20 – EM with 10 random seeds – implementation of Lu & Boutilier PSDD

This is the naive approach, circuit does not depend on data!

Learn Circuit from Data

Even in unstructured spaces

Tractable Learning

Bayesian networks Markov networks

Tractable Learning

Bayesian networks Markov networks Do not support linear-time exact inference

Tractable Learning

SPNs Cutset Networks Historically: Polytrees, Chow-Liu trees, etc. Both are Arithmetic Circuits (ACs)

[Darwiche, JACM 2003]

PSDDs are Arithmetic Circuits

2 1 n p1 s1 p2 s2 pn sn PSDD AC +

* * * * * *

1 2 n p1 s1 p2 s2 pn sn

Tractable Learning

Strong Properties Representational Freedom

Tractable Learning

Strong Properties Representational Freedom

DNN

Tractable Learning

Strong Properties Representational Freedom

DNN

Tractable Learning

Strong Properties Representational Freedom

DNN

SPN Cutset

Tractable Learning

Perhaps the most powerful circuit proposed to date

Strong Properties Representational Freedom

DNN

SPN Cutset

PSDDs for the Logic-Phobic

PSDDs for the Logic-Phobic

Bottom-up

each node is a distribution

PSDDs for the Logic-Phobic

Bottom-up

each node is a distribution

PSDDs for the Logic-Phobic

PSDDs for the Logic-Phobic

Multiply independent distributions

PSDDs for the Logic-Phobic

PSDDs for the Logic-Phobic

Weighted mixture of lower level distributions

PSDDs for the Logic-Phobic

PSDDs for the Logic-Phobic

Variable Trees (vtrees)

PSDD Vtree Correspondence

Learning Variable Trees

• How much do vars depend on each other?
• Learn vtree by hierarchical clustering

Learning Variable Trees

• How much do vars depend on each other?
• Learn vtree by hierarchical clustering

Learning Primitives

Learning Primitives

Learning Primitives

Primitives maintain PSDD properties and structured space!

LearnPSDD

Vtree learning Construct the most naïve PSDD LearnPSDD (search for better structure)

1 2 3

LearnPSDD

Vtree learning Construct the most naïve PSDD LearnPSDD (search for better structure)

1 2 3

Simulate

• perations

Execute the best Generate candidate

• perations

Experiments on 20 datasets

Experiments on 20 datasets

Compare with O-SPN: smaller size in 14, better LL in 11, win on both in 6 Compare with L-SPN: smaller size in 14, better LL in 6, win on both in 2

Experiments on 20 datasets

Compare with O-SPN: smaller size in 14, better LL in 11, win on both in 6 Compare with L-SPN: smaller size in 14, better LL in 6, win on both in 2 Comparable in performance & Smaller in size

Ensembles of PSDDs

Ensembles of PSDDs

EM/Bagging

Ensembles of PSDDs

EM/Bagging

State-of-the-Art Performance

State-of-the-Art Performance

State of the art in 6 datasets

What happens if you ignore constraints?

What happens if you ignore constraints?

Roadmap

Compile logic into a SDD Convert to a PSDD: Parameter estimation LearnPSDD

1 2 3

What happens if you ignore constraints?

Roadmap

Compile logic into a SDD Convert to a PSDD: Parameter estimation LearnPSDD

1 2 3

Simulate

• perations

Execute the best Generate candidate

• perations

What happens if you ignore constraints?

Discrete multi-valued data

𝑩: 𝒃𝟐, 𝒃𝟑, 𝒃𝟒 𝒃𝟐 ∧ ¬𝒃𝟑∧ ¬𝒃𝟒 ∨ ¬𝒃𝟐 ∧ 𝒃𝟑∧ ¬𝒃𝟒 ∨ ¬𝒃𝟐 ∧ ¬𝒃𝟑∧ 𝒃𝟒

What happens if you ignore constraints?

Discrete multi-valued data

𝑩: 𝒃𝟐, 𝒃𝟑, 𝒃𝟒 𝒃𝟐 ∧ ¬𝒃𝟑∧ ¬𝒃𝟒 ∨ ¬𝒃𝟐 ∧ 𝒃𝟑∧ ¬𝒃𝟒 ∨ ¬𝒃𝟐 ∧ ¬𝒃𝟑∧ 𝒃𝟒

What happens if you ignore constraints?

Discrete multi-valued data

𝑩: 𝒃𝟐, 𝒃𝟑, 𝒃𝟒 𝒃𝟐 ∧ ¬𝒃𝟑∧ ¬𝒃𝟒 ∨ ¬𝒃𝟐 ∧ 𝒃𝟑∧ ¬𝒃𝟒 ∨ ¬𝒃𝟐 ∧ ¬𝒃𝟑∧ 𝒃𝟒

What happens if you ignore constraints?

Discrete multi-valued data

𝑩: 𝒃𝟐, 𝒃𝟑, 𝒃𝟒 𝒃𝟐 ∧ ¬𝒃𝟑∧ ¬𝒃𝟒 ∨ ¬𝒃𝟐 ∧ 𝒃𝟑∧ ¬𝒃𝟒 ∨ ¬𝒃𝟐 ∧ ¬𝒃𝟑∧ 𝒃𝟒 Never omit domain constraints

Complex queries

and

Learning from constraints

Incomplete Data

id X Y Z 1 x1 y2 z1 2 x2 y1 z2 3 x2 y1 z2 4 x1 y1 z1 5 x1 y2 z2

a classical complete dataset closed-form (maximum-likelihood estimates are unique)

Incomplete Data

id X Y Z 1 x1 y2 z1 2 x2 y1 z2 3 x2 y1 z2 4 x1 y1 z1 5 x1 y2 z2

a classical complete dataset

id X Y Z 1 x1 y2

?

2 x2 y1

?

3

? ?

z2 4

?

y1 z1 5 x1 y2 z2

a classical incomplete dataset closed-form (maximum-likelihood estimates are unique) EM algorithm (on PSDDs)

Incomplete Data

id X Y Z 1 x1 y2 z1 2 x2 y1 z2 3 x2 y1 z2 4 x1 y1 z1 5 x1 y2 z2

a classical complete dataset

id X Y Z 1 x1 y2

?

2 x2 y1

?

3

? ?

z2 4

?

y1 z1 5 x1 y2 z2

a classical incomplete dataset a new type of incomplete dataset

id X Y Z 1 X  Z 2 x2 and (y2 or z2) 3 x2  y1 4 X  Y  Z  1 5 x1 and y2 and z2

closed-form (maximum-likelihood estimates are unique) EM algorithm (on PSDDs) Missed in the ML literature

id 1st sushi 2nd sushi 3rd sushi  1 fatty tuna sea urchin salmon roe  2 fatty tuna tuna shrimp  3 tuna tuna roll sea eel  4 fatty tuna salmon roe tuna  5 egg squid shrimp 

a classical complete dataset (e.g., total rankings)

id 1st sushi 2nd sushi 3rd sushi  1 fatty tuna sea urchin

?

 2 fatty tuna

? ?

 3 tuna tuna roll

?

 4 fatty tuna salmon roe

?

 5 egg

? ?



a classical incomplete dataset (e.g., top-k rankings)

Structured Datasets

id 1st sushi 2nd sushi 3rd sushi  1 fatty tuna sea urchin salmon roe  2 fatty tuna tuna shrimp  3 tuna tuna roll sea eel  4 fatty tuna salmon roe tuna  5 egg squid shrimp 

a classical complete dataset (e.g., total rankings)

id 1st sushi 2nd sushi 3rd sushi  1 (fatty tuna > sea urchin) and (tuna > sea eel)  2 (fatty tuna is 1st) and (salmon roe > egg)  3 tuna > squid  4 egg is last  5 egg > squid > shrimp 

a new type of incomplete dataset (e.g., partial rankings) (represents constraints on possible total rankings)

Structured Datasets

Learning from Incomplete Data

• Movielens Dataset:

– 3,900 movies, 6,040 users, 1m ratings – take ratings from 64 most rated movies – ratings 1-5 converted to pairwise prefs.

• PSDD for partial rankings

– 4 tiers – 18,711 parameters

rank movie 1 The Godfather 2 The Usual Suspects 3 Casablanca 4 The Shawshank Redemption 5 Schindler’s List 6 One Flew Over the Cuckoo’s Nest 7 The Godfather: Part II 8 Monty Python and the Holy Grail 9 Raiders of the Lost Ark 10 Star Wars IV: A New Hope

movies by expected tier

PSDD Sizes

Structured Queries

rank movie 1 Star Wars V: The Empire Strikes Back 2 Star Wars IV: A New Hope 3 The Godfather 4 The Shawshank Redemption 5 The Usual Suspects

Structured Queries

rank movie 1 Star Wars V: The Empire Strikes Back 2 Star Wars IV: A New Hope 3 The Godfather 4 The Shawshank Redemption 5 The Usual Suspects

• no other Star Wars movie in top-5
• at least one comedy in top-5

Structured Queries

rank movie 1 Star Wars V: The Empire Strikes Back 2 Star Wars IV: A New Hope 3 The Godfather 4 The Shawshank Redemption 5 The Usual Suspects

• no other Star Wars movie in top-5
• at least one comedy in top-5

rank movie 1 Star Wars V: The Empire Strikes Back 2 American Beauty 3 The Godfather 4 The Usual Suspects 5 The Shawshank Redemption

Structured Queries

rank movie 1 Star Wars V: The Empire Strikes Back 2 Star Wars IV: A New Hope 3 The Godfather 4 The Shawshank Redemption 5 The Usual Suspects

• no other Star Wars movie in top-5
• at least one comedy in top-5

rank movie 1 Star Wars V: The Empire Strikes Back 2 American Beauty 3 The Godfather 4 The Usual Suspects 5 The Shawshank Redemption

diversified recommendations via logical constraints

Conclusions

• Structured spaces are everywhere 
• PSDDs build on logical circuits
• 1. Tractability
• 2. Semantics
• 3. Natural encoding of structured spaces
• Learning is effective

1. From constraints encoding structured space

State of the art learning preference distributions

2. From standard unstructured datasets using search

State of the art on standard tractable learning datasets

• Novel settings for inference and learning

Structured spaces / learning from constraints / complex queries

Probabilistic Sentential Decision Diagrams

Doga Kisa, Guy Van den Broeck, Arthur Choi and Adnan Darwiche KR, 2014

Learning with Massive Logical Constraints

Doga Kisa, Guy Van den Broeck, Arthur Choi and Adnan Darwiche ICML LTPM workshop, 2014

Tractable Learning for Structured Probability Spaces

Arthur Choi, Guy Van den Broeck and Adnan Darwiche IJCAI, 2015

Tractable Learning for Complex Probability Queries

Jessa Bekker, Jesse Davis, Arthur Choi, Adnan Darwiche, Guy Van den Broeck. NIPS, 2015

Learning the Structure of PSDDs

Yitao Liang, Jessa Bekker and Guy Van den Broeck UAI, 2017

Towards Compact Interpretable Models: Learning and Shrinking PSDDs

Yitao Liang and Guy Van den Broeck IJCAI XAI workshop, 2017

References

(P)SDDs in Melbourne

• Sunday: Logical Foundations for Uncertainty and

Machine Learning Workshop

– Adnan Darwiche: “On the Role of Logic in Probabilistic Inference and Machine Learning” – YooJung Choi: “Optimal Feature Selection for Decision Robustness in Bayesian Networks”

• Sunday: Explainable AI Workshop

– Yitao Liang: “Towards Compact Interpretable Models: Learning and Shrinking PSDDs”

• Tuesday: IJCAI

– YooJung Choi (again)

Conclusions

Statistical ML “Probability” Symbolic AI “Logic” Connectionism “Deep”

PSDD

Questions?

PSDD with 15,000 nodes LearnPSDD code: https://github.com/UCLA-StarAI/LearnPSDD Other PSDD code: http://reasoning.cs.ucla.edu/psdd/ SDD code: http://reasoning.cs.ucla.edu/sdd/