Compressing Polarized Boxes Beniamino Accattoli Carnegie Mellon University

B. Accattoli (CMU)

Compressing Polarized Boxes

1 / 39

linear logic and boxes

Proof nets: the graphical syntax for linear logic. Brought new deep perspectives about normalization: 1

Optimal reductions;

2

Implicit computational complexity;

3

Explicit substitutions;

4

Strong normalization.

Key tool: boxes for the promotion rule, the heart of the system. This work: a new understanding of boxes, via polarity.

B. Accattoli (CMU)

Compressing Polarized Boxes

2 / 39

Multiplicative Linear Logic (MLL)

Identity rules:

Multiplicative rules:

B. Accattoli (CMU)

` A⊥ , A

ax

` Γ, A ` ∆, B ⊗ ` Γ, ∆, A ⊗ B

Compressing Polarized Boxes

` Γ, A

` A⊥ , ∆ cut ` Γ, ∆

` Γ, A, B ` ` Γ, A ` B

3 / 39

Proof nets for MLL

` A⊥ , A

ax

ax A⊥

π :

σ :

` Γ, A

` ∆, A⊥ ` Γ, ∆

π? cut

σ? A cut A⊥

Γ



π?

π :

A

Γ

` Γ, A, B

`

` Γ, A ` B

`

B

A`B

π :

σ :

` Γ, A

` ∆, B

` Γ, ∆, A ⊗ B

B. Accattoli (CMU)

A

π? Γ ⊗

A

B



σ? ∆

A⊗B

Compressing Polarized Boxes

4 / 39

Cut-elimination for MLL

ax

A A⊥ cut

A

A⊥ B ⊥

A

`

B



A

→ax

→`

cut

P⊥

Q⊥ P cut

Q

cut

No duplication/erasure of subnets ⇒ Everything works fine

B. Accattoli (CMU)

Compressing Polarized Boxes

5 / 39

Multiplicative Exponential Linear Logic (MELL)

MLL + Exponential rules:

B. Accattoli (CMU)

` Γ, A d ` Γ, ?A

`?Γ, A ! `?Γ, !A

` Γ, ?A, ?A c ` Γ, ?A

`Γ w ` Γ, ?A

Compressing Polarized Boxes

6 / 39

Exponential Cut-elimination Consider the following cut with contraction: π

ρ

:

:

` ?A⊥ , ?A⊥ , Γ

`?∆, A c ! `?∆, !A ` ?A⊥ , Γ cut `?∆, Γ

Its elimination requires to duplicate ρ: ρ ρ :

`?∆, A ! `?∆, !A

:

`?∆, A ! `?∆, !A

π : ⊥

` ?A , ?A⊥ , Γ

`?∆, ?A⊥ , Γ `?∆, ?∆, Γ c .. . c `?∆, Γ

cut

cut

Similarly, weakening induces erasure of sub-proofs. B. Accattoli (CMU)

Compressing Polarized Boxes

7 / 39

Na¨ıve proof nets for MELL π :

`Γ w ` Γ, ?A

π?

w

Γ

π

?A

π?

:

A

` Γ, A d ` Γ, ?A

d ?A

π

π?

:

` Γ, ?A, ?A c ` Γ, ?A

Γ

?A ?A

π

π?

:

` ?Γ, A ! ` ?Γ, !A

B. Accattoli (CMU)

?A

c



Compressing Polarized Boxes

A

! !A

8 / 39

How to eliminate cuts? Na¨ıve translation of promotion: π

π?

:

` ?Γ, A ! ` ?Γ, !A

A

!



!A

Given this cut in a generic net: ?A⊥ ?A⊥

A

c

!

?A⊥

cut

!A

There is no way of recovering the sub-proof to duplicate. Then !-rules are represented as boxes: π π?

:

` ?Γ, A ! ` ?Γ, !A B. Accattoli (CMU)

A

! ?Γ

Compressing Polarized Boxes

!A 9 / 39

Exponential cut elimination implemented using boxes w ?A



A⊥

P

d

cut!A ?A⊥

?A⊥

cut

A

?∆

A⊥

A

P

cut Γ

!A

!

... ?B1 ?Bk

!B cut

! ?Γ

→c

! ... ?A⊥ ?A⊥ !A cut cut

! !A . . .

c

c

... ?B1 ?Bk

→

P ! A

B. Accattoli (CMU)

w

Γ

P !

...

?B1 . . . ?Bk

→d

!

?A⊥ ?A⊥

c

w

→w

! ... !A cut ?B ?Bk 1

?∆

Compressing Polarized Boxes

!B cut

!

?Γ 10 / 39

Boxes

Boxes solve the problem of defining cut-elimination. However, the solution is drastic, equivalent to give up. Some fragments seem to have an inherent notion of box. Where does the problem lie? Is there a logic feature that internalizes boxes?

B. Accattoli (CMU)

Compressing Polarized Boxes

11 / 39

Last rule 1 Main problem: in proof nets there is no last rule. Re-consider: π :

ρ : `?∆, !A

!

` ?A⊥ , Γ `?∆, Γ

c



`?∆, !A

cut

π :

`?∆, A !

`?∆, !A

`?∆, A

` ?A⊥ , ?A⊥ , Γ

`?∆, A

ρ :

ρ : !

` ?A⊥ , ?A⊥ , Γ

`?∆, ?A⊥ , Γ `?∆, ?∆, Γ . . . `?∆, Γ

cut

cut

c c

In sequent calculus: rule occurrence r 7→ sub-proof ending on r . No such thing in proof nets! B. Accattoli (CMU)

Compressing Polarized Boxes

12 / 39

Last rule 2

Intuition: Internalizing a notion of last rule will internalize boxes Partially internalized boxes: Olivier Laurent’s polarized MELL. Abstract last rule = last positive rule. This work: totally internalized boxes for MELLP. Expressiveness: MELLP codes classical logic/λµ-calculus.

B. Accattoli (CMU)

Compressing Polarized Boxes

13 / 39

Outline

1

Polarized MELL

2

Compressing polarized boxes

B. Accattoli (CMU)

Compressing Polarized Boxes

14 / 39

Polarization Formulas: P, Q ::= X N, M ::= X ⊥

| 1 | P ⊗Q | ⊥ | N `M

| !N | ?P

Sequents: `Γ;P

or

`Γ;

Multiplicative rules: ` Γ; P

` ∆, P ⊥ ; [Q] cut ` Γ, ∆; [Q] ` Γ; [P] ⊥ ` Γ, ⊥; [P]

` Γ, N, M; [P] ` ` Γ, N ` M; [P] B. Accattoli (CMU)

ax

` P ⊥; P

`; 1

1

` Γ; P ` ∆; Q ⊗ ` Γ, ∆; P ⊗ Q

Compressing Polarized Boxes

15 / 39

Laurent’s MELLP: adding exponentials Exponential rules: ` Γ; [P] w ` Γ, N; [P]

` Γ; P ` Γ, ?P;

d

` Γ, N, N; [P] c ` Γ, N; [P]

` Γ, N; ` Γ; !N

!

Difference with linear logic: Promotion, contraction, and weakening do not need the ? modality. Important: Only positives are duplicated/erased. Positives are last rules and every positive will have a box. B. Accattoli (CMU)

Compressing Polarized Boxes

16 / 39

`; 1

1

1

` P ⊥; P

1 ax

ax

π?

` Γ; P ` ∆; Q ⊗ ` Γ, ∆; P ⊗ Q

P⊥

P

P

Q

Γ

θ? ⊗



P⊗Q

` Γ; P

` ∆, P ⊥ ; [Q] ` Γ, ∆; [Q]

π? cut

θ? ⊥ cut P

P Γ



[Q]

π? ` Γ, N, M; [P] ` Γ, N ` M; [P]

`

N Γ

`

M [P]

N`M

B. Accattoli (CMU)

Compressing Polarized Boxes

17 / 39

w

` Γ, N; [P]

` Γ; P ` Γ, ?P;

π?

w

` Γ; [P]

N

Γ

[P]

π? d

P

d

Γ

?P

π? ` Γ, N, N; [P] ` Γ, N; [P]

c

N Γ

N

c

[P]

N

` Γ, N; ` Γ; !N

π? !

! Γ

B. Accattoli (CMU)

N

Compressing Polarized Boxes

!N

18 / 39

Positive Trees ` Γ; P

` ∆, P ⊥ ; [Q] cut ` Γ, ∆; [Q] ` Γ; [P] ⊥ ` Γ, ⊥; [P]

` Γ, N, M; [P] ` ` Γ, N ` M; [P]

ax

` P ⊥; P

`; 1

1

` Γ; P ` ∆; Q ⊗ ` Γ, ∆; P ⊗ Q

` Γ; [P] w ` Γ, N; [P]

` Γ; P ` Γ, ?P;

d

` Γ, N, N; [P] c ` Γ, N; [P]

` Γ, N; ` Γ; !N

!

Note: positives have a forest structure. B. Accattoli (CMU)

Compressing Polarized Boxes

19 / 39

Positive Tree Positive connectives: 1, ⊗, !. Explicit boxes for ! ⇒ induced box for every positive: H

H



! N1

!

N ... k

N1

Q

P

⊗ P⊗Q

P⊥



ax P

P⊥

N ... k

Q

P



ax P

1 ⊗



1

P⊗Q

My contribution: explicit boxes for ! can be made implicit.

B. Accattoli (CMU)

Compressing Polarized Boxes

20 / 39

Generalized rewriting rule

Laurent uses the positive tree to generalize box rules: T (+) w

+

→w

...

cut N1

B. Accattoli (CMU)

w

...

w

N1

...

Nk

Nk

Compressing Polarized Boxes

21 / 39

Polarized cut-elimination 1

ax P⊥ cut

P

ax

N

N⊥ cut

N

P⊥ Q



`

P

→ax+

Q

P



P

→ax−

→`

N

P⊥

Q⊥ P cut

Q

cut

cut

B. Accattoli (CMU)

Compressing Polarized Boxes

22 / 39

Polarized cut-elimination 2 N⊥

N⊥

E d

!N cut

w P⊥

cut

!

P

→d

... N1

+

w

T (+)

→w

...

c P

P cut

w

N1 . . . Nk

P⊥ P⊥

+

...

T (+)

→c

... N1

P cut

T (+) ...

+

+

Q

... M1 Mh

B. Accattoli (CMU)

cut

P

+

T (+) ... N1

Nk

→

E !

cut ... M1 Mh

Compressing Polarized Boxes

c

c

Nk

E

T (+) ...

P cut

N1

!

Nk

Nk

P⊥ P⊥



... N1

Nk

N1

E

N cut

P

...

+

Nk

T (+) ... N1

Nk

23 / 39

Outline

1

Polarized MELL

2

Compressing polarized boxes

B. Accattoli (CMU)

Compressing Polarized Boxes

24 / 39

Matching property ` Γ; P

` ∆, P ⊥ ; [Q] cut ` Γ, ∆; [Q] ` Γ; [P] ⊥ ` Γ, ⊥; [P]

` Γ, N, M; [P] ` ` Γ, N ` M; [P]

ax

` P ⊥; P

`; 1

1

` Γ; P ` ∆; Q ⊗ ` Γ, ∆; P ⊗ Q

` Γ; [P] w ` Γ, N; [P]

` Γ; P ` Γ, ?P;

d

` Γ, N, N; [P] c ` Γ, N; [P]

` Γ, N; ` Γ; !N

!

Matching property: every !-rule is enabled by a d-rule. B. Accattoli (CMU)

Compressing Polarized Boxes

25 / 39

Materializing the matching property 1 Consider: w !

1 d

w !

w

1 d

w

Problem: without box the content is disconnected. Idea: let’s materialize the matching property with an additional edge. w !

i

1 d

w

The content and the positive sub-graphs are now connected. The induced box: the positive tree plus the negative trees on it.

B. Accattoli (CMU)

Compressing Polarized Boxes

26 / 39

Quotient and weakenings Let’s do it again: w !

1 d

w !

w

1 d

w

We do not recover the original box: w !

i

1 d

w

w !

1 d

w

Interpretation: we are quotienting proof nets with explicit boxes. Remark: weakenings are not attached! ⇒ improvement over Francois Lamarche’s essential nets. B. Accattoli (CMU)

Compressing Polarized Boxes

27 / 39

Box borders

Let’s do it again: w !

1 d

w

w !

1 d

i

1

1

d

d

!

!

w

i

Remark: we are not attaching the border of the box. ⇒ improvement over Ian Mackie’s interaction nets technique.

B. Accattoli (CMU)

Compressing Polarized Boxes

28 / 39

Implicit boxes Recipe: Take a cut-free proof net. Matching: every !-box has a unique dereliction at level 0. Remove the explicit box and add the matching edge. Then: The induced boxes define a net with explicit boxes. Induced boxes are locally reconstructable. There is a simple correctness criterion (i.e. not ad-hoc). It is a canonical representation (i.e. no choice). B. Accattoli (CMU)

Compressing Polarized Boxes

29 / 39

Summing up

In a cut-free proof net the explicit box of a ! can be replaced by a single edge in a canonical and more parallel way.

B. Accattoli (CMU)

Compressing Polarized Boxes

30 / 39

Cuts Cuts introduce a problem: ax

ax cut

!

d

The positive sub-graph is no longer connected. Let’s iterate the same idea: ax

!

B. Accattoli (CMU)

ax cut

i i

d

Compressing Polarized Boxes

31 / 39

Example ax ax

w w c N M ` ! d N`M ?!P

P

⊗ d

P

?(P ⊗ P)



w !

w cut

w

1

c ! d

?1

cut

ax ax

w w c N M ` ! d N`M

i

⊗ d i

w !

w cut

w

P

⊗ d

P

?(P ⊗ P)

?!P ⊥



ax ax

w w `

1 d

i

1 d

w w

1

c ! d

1 d

1

?1

cut

w cut ⊥

ax ax

w w `

!

i

⊗ d i

w !

w

i

1 d

cut

w

1 cut

cut

Implicit box: one dereliction plus the cuts at level 0. Induced box: positive tree plus negative sub-trees. Novelty: ` commutes with box borders! B. Accattoli (CMU)

Compressing Polarized Boxes

32 / 39

Cut elimination ax

X⊥ cut

X

!

P⊥ Q

X

Q

P

`

→ax−

⊗ !

→`

i !0

→ax+ !

X⊥

N

N⊥

d

!

c P⊥

!

P cut

P⊥ P⊥

+

+

i

P

cut

!

i

P

+

N1

N1

w →w Nk

!0

i i

!00

+ c

i

(+) ...

N cut

(+) ...

P

i



→d

i

(+) ... cut

!

Nk

w

B. Accattoli (CMU)

P cut

→c

... N1

i

cut

(+)

cut

i

!00

P⊥ P⊥

Q

i

i

X⊥

i

cut

!

ax X cut

Q⊥ P

P⊥

cut

!

i

N⊥

X⊥

!



X

...

c ...

Nk

w

N1 . . . Nk

!

Compressing Polarized Boxes

33 / 39

Side effects 1 Cut elimination has ’side effects’. Consider the weakening rule: w P



cut

!

P

+

w

(+) ... N1

→w Nk

...

w

N1 . . . Nk

!

i

It automatically pushes the created weakenings out of boxes: w !

+ cut i

(+)

...

... N1

→w

!

Nk

...

w

w

... N1 Nk

Similarly for contraction.

B. Accattoli (CMU)

Compressing Polarized Boxes

34 / 39

No commutative cuts

There is no commutative rule. It is included inside the axiom and dereliction rules. The axiom rule: ax

X⊥ cut

X

!

X

X

→ax− !

i

and its action through box borders: P

ax

... P cut

!

B. Accattoli (CMU)

+

(+) ... N1

Nk

→ax−

i

Compressing Polarized Boxes

+ P

(+)

N1

... Nk

!

35 / 39

No commutative cuts 2 The dereliction rule: N⊥ i !0

N

d

!

N⊥ i

→d

cut

!00

N cut

!0

i

i i

!00

and an example of its action:

i !0

d0

... cut

!00

B. Accattoli (CMU)

w !

i

1 d

w cut

→d

i i !0

1 d

...

!00

i

Compressing Polarized Boxes

36 / 39

Cut-elimination On η-expanded nets: Cut elimination is strongly normalizing (SN). Confluence requires assoc., comm., and neutrality for contractions. Then: Church-Rosser modulo and SN modulo hold. The general case: More and wilder ’side effects’. Difficult critical pairs and known techniques do not work. By-product: new simpler proof of SN for linear logic (RTA 2013).

B. Accattoli (CMU)

Compressing Polarized Boxes

37 / 39

Conclusions An alternative representation of boxes: Simple; Canonical; More parallel; Provided of a correctness criterion; A local reconstruction of boxes; Results on the dynamics.

New perspective on polarity. It already lead to new understanding of SN for linear logic.

B. Accattoli (CMU)

Compressing Polarized Boxes

38 / 39

THANKS!

B. Accattoli (CMU)

Compressing Polarized Boxes

39 / 39

Compressing Polarized Boxes

Boxes solve the problem of defining cut-elimination. However, the solution is drastic, equivalent to give up. Some fragments seem to have an inherent notion of box. Where does the problem lie? Is there a logic feature that internalizes boxes? B. Accattoli (CMU). Compressing Polarized Boxes. 11 / 39 ...

383KB Sizes 1 Downloads 219 Views

Recommend Documents

Compressing Polarized Boxes
compact, and natural representation of boxes: in an expressive polarized ...... negative arborescence of the external ! is given by the axioms, the contraction and ...

Compressing Polarized Boxes
classical system LC [25], [26] and it is built around the concept of polarity. ... key technical point here is a representation of implicit boxes as additional edges ...

Jumping Boxes
... does not require any linear logic notation; (ii) our approach does not de- ...... (Dup) t[u/x]. → t{y|x}[u/x][u/y]. |t|x ≥ 2 & y fresh where t[v/y] stands for t[v1/y1].

Wooden Boxes - Zongkers
card theme. In this chapter, you'll learn tech- niques for cutting .... add more clamps, spacing them about 3 in. apart. THE DECORATIVE BANDING THAT RUNS.

Compressing Forwarding Tables
Abstract—With the rise of datacenter virtualization, the number of entries in forwarding tables is expected to scale from several thousands to several millions. Unfortunately, such forwarding table sizes can hardly be implemented today in on-chip m

Method for organizing and compressing spatial data
Aug 13, 2010 - “Wireless Solutions with SpatialFX, Any Client, Anywhere,” Object/. FX Corporation; Aug. .... segmenting of spatial databases to improve data access times ... Other objects and advantages of the present invention will become ...

pdf boxes
There was a problem previewing this document. Retrying... Download. Connect more apps... Try one of the apps below to open or edit this item. pdf boxes.

nesting boxes template.pdf
Page 1 of 1. Page 1 of 1. nesting boxes template.pdf. nesting boxes template.pdf. Open. Extract. Open with. Sign In. Main menu. Displaying nesting boxes ...

Boxes of Love.pdf
Page 1 of 1. Grade Level Boys Girls. Prek-Kindergarten Toothpaste toothbrushes. 1. st Grade Soap, washcloths body wash, bath puffs. 2. nd Grade fuzzy socks New, small stuffed. animals. 3. rd Grade shampoo,. conditioner. hair brushes, barrettes. 4. th

On compressing social networks
far less compressible than Web graphs yet closer to host graphs and exploiting link ... code with parameter 4 (which we found to be the best in our ex- periments) [7]. .... called min-wise independent family suffices [10]; in practice, even pairwise 

Nonequilibrium superconductivity in spin-polarized ...
They correspond to Emax 6 and max 12 for 0.5 meV. The kinetic equations are then solved in the Xi variables by using standard library routines for systems of nonlinear equations. The u functions enter- ing the DOS and the coherence factors have to be

Nonequilibrium superconductivity in spin-polarized ...
fully explained by electronic spin-flip processes induced by spin-orbit ... ported in Ref. 19 showed a clear signature of a small but ..... If we add a quasiparticle ...

Download Qualitative Polarized Light Microscopy ...
Download Qualitative Polarized Light. Microscopy (Microscopy Handbooks 09). Ebook Download. Download Qualitative Polarized Light Microscopy (Microscopy ...

Compressing Regular Expression Sets for Deep Packet Inspection
from intrusion detection systems to firewalls and switches. While early systems classified traffic based only on header-level packet information, modern systems are capable of detecting malicious patterns within the actual packet payload. This deep p

THP-monthly-cal-boxes-oct_VintageglamStudio.pdf
There was a problem previewing this document. Retrying... Download. Connect more apps... Try one of the apps below to open or edit this item.

The Convergence of Difference Boxes
Jan 14, 2005 - On the midpoint of each side write the (unsigned) difference between the two numbers at its endpoints. 3. Inscribe a new square in the old one, ...

Compressing Regular Expression Sets for Deep Packet ...
an evolutionary search based on Genetic Programming: a large popula- tion of expressions ... models the effort required for applying all expressions in R to a given string. 1 http://www.snort.org ..... web-php.rules.pcre. 16. 400 105. 3360.

Compressing Deep Neural Networks using a ... - Research at Google
tractive model for many learning tasks; they offer great rep- resentational power ... differs fundamentally in the way the low-rank approximation is obtained and ..... 4Specifically: “answer call”, “decline call”, “email guests”, “fast

Mining Compressing Patterns in a Data Stream
18 Apr 2013 - problem of mining compressing sequential patterns from a data stream. We propose an ... Mining frequent patterns is an important research topic in data mining. It has been shown that frequent ...... Part of the work was done when the fi