Asynchronous Spiking Neural P Systems with Structural Plasticity

Size: px

Start display at page:

Download "Asynchronous Spiking Neural P Systems with Structural Plasticity"

Jack Ryan
5 years ago
Views:

1 Asynchronous Spiking Neural P Systems with Structural Plasticity Francis George Carreon-Cabarle 1, Henry N. Adorna 1, Mario J. Pérez-Jiménez 2 1 Department of Computer Science, University of the Philippines Diliman Quezon city, 1101, Philippines; 2 Department of Computer Science and AI University of Sevilla Avda. Reina Mercedes s/n, 41012, Sevilla, Spain fccabarle@up.edu.ph, hnadorna@dcs.upd.edu.ph, marper@us.es 02 to 06 Feb th Brainstorming Week on Membrane Computing Seville, Spain

2 SNP systems with structural plasticity In short, SNPSP systems; Dynamism is only applied for synapses: can create and delete synapses; Introduced in Asian CMC 2013 (Chengdu, China), then improved and extended; 1 Sequential SNPSP systems; 2 1 Cabarle, F.G.C., Adorna, H.N., Pérez-Jiménez, M.J., Song, T.: Spiking neural P systems with structural plasticity (in press). Neural Computing and Applications (2015) 2 Cabarle, F.G.C., Adorna, H.N., Pérez-Jiménez, M.J.: Sequential Spiking neural P systems with structural plasticity (submitted). Neural Computing and Applications (2015)

3 SNP systems with structural plasticity In short, SNPSP systems; Dynamism is only applied for synapses: can create and delete synapses; Introduced in Asian CMC 2013 (Chengdu, China), then improved and extended; 1 Sequential SNPSP systems; 2 1 Cabarle, F.G.C., Adorna, H.N., Pérez-Jiménez, M.J., Song, T.: Spiking neural P systems with structural plasticity (in press). Neural Computing and Applications (2015) 2 Cabarle, F.G.C., Adorna, H.N., Pérez-Jiménez, M.J.: Sequential Spiking neural P systems with structural plasticity (submitted). Neural Computing and Applications (2015)

4 SNP systems with structural plasticity In short, SNPSP systems; Dynamism is only applied for synapses: can create and delete synapses; Introduced in Asian CMC 2013 (Chengdu, China), then improved and extended; 1 Sequential SNPSP systems; 2 1 Cabarle, F.G.C., Adorna, H.N., Pérez-Jiménez, M.J., Song, T.: Spiking neural P systems with structural plasticity (in press). Neural Computing and Applications (2015) 2 Cabarle, F.G.C., Adorna, H.N., Pérez-Jiménez, M.J.: Sequential Spiking neural P systems with structural plasticity (submitted). Neural Computing and Applications (2015)

5 SNP systems with structural plasticity In short, SNPSP systems; Dynamism is only applied for synapses: can create and delete synapses; Introduced in Asian CMC 2013 (Chengdu, China), then improved and extended; 1 Sequential SNPSP systems; 2 1 Cabarle, F.G.C., Adorna, H.N., Pérez-Jiménez, M.J., Song, T.: Spiking neural P systems with structural plasticity (in press). Neural Computing and Applications (2015) 2 Cabarle, F.G.C., Adorna, H.N., Pérez-Jiménez, M.J.: Sequential Spiking neural P systems with structural plasticity (submitted). Neural Computing and Applications (2015)

6 SNP systems with structural plasticity A spiking neural P system with structural plasticity (SNPSP system) of degree m 1 is a construct of the form Π = (O, σ 1,..., σ m, syn, out), where: O = {a} is the singleton alphabet (a is called spike); σ 1,..., σ m are neurons of the form (n i, R i ), 1 i m; n i 0 indicates the initial number of spikes in σ i ; R i is a finite rule set of σ i with two forms: 1. Spiking rule: E/a c a, where E is a regular expression over O, c 1; 2. Plasticity rule: E/a c αk(i, N), where E is a regular expression over O, c 1, α {+,, ±, }, k 1, and N {1,..., m} {i}; syn {1,..., m} {1,..., m}, with (i, i) / syn for 1 i m (synapses between neurons); out {1,..., m} indicate the output neuron.

7 SNPSP systems example step 1: syn = {(j, k), (k, l)} i a a ±1(i, {j, k}) l a k j a a a a a a

8 SNPSP systems example step 1: syn = {(j, k), (k, l)} i step 2: syn = {(j, k), (k, l),(i, j) } a ±1(i, {j,k}) l k j a a a a a a a

9 SNPSP systems example step 1: syn = {(j, k), (k, l)} step 2: syn = {(j, k), (k, l),(i, j) } step 3: syn = {(j, k), (k, l)} i a ±1(i, {j,k}) l k a j a a a a a a

10 SNPSP systems example step 1: syn = {(j, k), (k, l)} step 2 i : syn = {(j, k), (k, l),(i, k) } a ±1(i, {j,k }) l k a j a a a a a a

11 SNPSP systems example step 1: syn = {(j, k), (k, l)} step 2 : syn = {(j, k), (k, l),(i, j) } step 3: syn = {(j, k), (k, l)} i a ±1(i, {j,k}) l a k j a a a a a a

12 In this work Asynchronous SNPSP systems: Show SLIN = N tot SNP SP asyn (bound p ), p 1, by: N totsnp SP asyn (bound p) SLIN, p 1. SLIN N totsnp SP asyn (bound p), p 1. NRE = N tot W SNP SP asyn Under a normal form.

13 In this work Asynchronous SNPSP systems: Show SLIN = N tot SNP SP asyn (bound p ), p 1, by: N totsnp SP asyn (bound p) SLIN, p 1. SLIN N totsnp SP asyn (bound p), p 1. NRE = N tot W SNP SP asyn Under a normal form.

14 In this work Asynchronous SNPSP systems: Show SLIN = N tot SNP SP asyn (bound p ), p 1, by: N totsnp SP asyn (bound p) SLIN, p 1. SLIN N totsnp SP asyn (bound p), p 1. NRE = N tot W SNP SP asyn Under a normal form.

15 In this work Asynchronous SNPSP systems: Show SLIN = N tot SNP SP asyn (bound p ), p 1, by: N totsnp SP asyn (bound p) SLIN, p 1. SLIN N totsnp SP asyn (bound p), p 1. NRE = N tot W SNP SP asyn Under a normal form.

16 In this work Asynchronous SNPSP systems: Show SLIN = N tot SNP SP asyn (bound p ), p 1, by: N totsnp SP asyn (bound p) SLIN, p 1. SLIN N totsnp SP asyn (bound p), p 1. NRE = N tot W SNP SP asyn Under a normal form.

17 In this work Asynchronous SNPSP systems: Show SLIN = N tot SNP SP asyn (bound p ), p 1, by: N totsnp SP asyn (bound p) SLIN, p 1. SLIN N totsnp SP asyn (bound p), p 1. NRE = N tot W SNP SP asyn Under a normal form.

18 SLIN = N tot SNP SP asyn (bound p ), p 1 N tot SNP SP asyn (bound p ) SLIN, p 1. Construct a right-linear grammar G, such that Π generates the length set of the regular language L(G). Let us denote by C the set of all possible configurations of Π, with C 0 being the initial configuration. The right-linear grammar G = (C, {a}, C 0, P ), where the production rules in P are as follows: (1) C C, for C, C C where Π has a transition C C in which the output neuron does not spike; (2) C ac, for C, C C where Π has a transition C C in which the output neuron spikes; (3) C λ, for any C C in which Π halts.

19 SLIN = N tot SNP SP asyn (bound p ), p 1 SLIN N tot SNP SP asyn (bound p ), p 1. ADD module simulating l i : (ADD(1) : l j, l k ) of a strongly monotonic register machine. l i a a 1 a q /a a for 1 q p li 1 l a ±1(li 1, {l j, l k }) j l k

20 Achieving universality Additional ingredient: weighted synapses 3 Synapse set is now of the form syn {1, 2,..., m} {1, 2,..., m} N. If σ i applies a rule E/a c a p, and the synapse (i, j, r) exists (i.e. the weight of synapse (i, j) is r) then σ j receives p r spikes. Normal form: if σ i has standard rule, then σ i is simple, i.e. R i = 1; 3 Pan, L., Zeng, X., Zhang, X.: Spiking Neural P Systems with Weighted Synapses. Neural Processing Letters vol. 35, pp (2012)

21 Achieving universality Additional ingredient: weighted synapses 3 Synapse set is now of the form syn {1, 2,..., m} {1, 2,..., m} N. If σ i applies a rule E/a c a p, and the synapse (i, j, r) exists (i.e. the weight of synapse (i, j) is r) then σ j receives p r spikes. Normal form: if σ i has standard rule, then σ i is simple, i.e. R i = 1; 3 Pan, L., Zeng, X., Zhang, X.: Spiking Neural P Systems with Weighted Synapses. Neural Processing Letters vol. 35, pp (2012)

22 Achieving universality Additional ingredient: weighted synapses 3 Synapse set is now of the form syn {1, 2,..., m} {1, 2,..., m} N. If σ i applies a rule E/a c a p, and the synapse (i, j, r) exists (i.e. the weight of synapse (i, j) is r) then σ j receives p r spikes. Normal form: if σ i has standard rule, then σ i is simple, i.e. R i = 1; 3 Pan, L., Zeng, X., Zhang, X.: Spiking Neural P Systems with Weighted Synapses. Neural Processing Letters vol. 35, pp (2012)

23 Achieving universality Additional ingredient: weighted synapses 3 Synapse set is now of the form syn {1, 2,..., m} {1, 2,..., m} N. If σ i applies a rule E/a c a p, and the synapse (i, j, r) exists (i.e. the weight of synapse (i, j) is r) then σ j receives p r spikes. Normal form: if σ i has standard rule, then σ i is simple, i.e. R i = 1; 3 Pan, L., Zeng, X., Zhang, X.: Spiking Neural P Systems with Weighted Synapses. Neural Processing Letters vol. 35, pp (2012)

24 NRE = N tot W SNP SP asyn ADD module simulating l i : (ADD(r) : l j, l k ) of register machine. l i a a l i 1 a a r 2 l j a ±1(li 2, {l j, l k }, 1) li 2 l k

25 NRE = N tot W SNP SP asyn SUB module simulating l i : (SUB(r) : l j, l k ) of register machine. l i l 1 4 S r s i a a a p 1(li 1, {r}, 1) r a(a 2 ) + /a 3 for 3 S r p < 8 S r ± S r (r, S r, 4 S r + s) a 8 Sr ±1(li 1 a ± S r (r, S r, 5 S r + s), {l j}, 1) a 9 Sr ±1(li 1, {l k}, 1) l j l k

26 NRE = N tot W SNP SP asyn FIN module. a(a 2 ) + /a 2 a 1 a a l h

27 In summary Asynchronous SNPSP systems: Shown SLIN = N tot SNP SP asyn (bound p ), p 1, by: N totsnp SP asyn (bound p) SLIN, p 1. SLIN N totsnp SP asyn (bound p), p 1. NRE = N tot W SNP SP asyn Under a normal form.

28 In summary Asynchronous SNPSP systems: Shown SLIN = N tot SNP SP asyn (bound p ), p 1, by: N totsnp SP asyn (bound p) SLIN, p 1. SLIN N totsnp SP asyn (bound p), p 1. NRE = N tot W SNP SP asyn Under a normal form.

29 In summary Asynchronous SNPSP systems: Shown SLIN = N tot SNP SP asyn (bound p ), p 1, by: N totsnp SP asyn (bound p) SLIN, p 1. SLIN N totsnp SP asyn (bound p), p 1. NRE = N tot W SNP SP asyn Under a normal form.

30 In summary Asynchronous SNPSP systems: Shown SLIN = N tot SNP SP asyn (bound p ), p 1, by: N totsnp SP asyn (bound p) SLIN, p 1. SLIN N totsnp SP asyn (bound p), p 1. NRE = N tot W SNP SP asyn Under a normal form.

31 In summary Asynchronous SNPSP systems: Shown SLIN = N tot SNP SP asyn (bound p ), p 1, by: N totsnp SP asyn (bound p) SLIN, p 1. SLIN N totsnp SP asyn (bound p), p 1. NRE = N tot W SNP SP asyn Under a normal form.

32 In summary Asynchronous SNPSP systems: Shown SLIN = N tot SNP SP asyn (bound p ), p 1, by: N totsnp SP asyn (bound p) SLIN, p 1. SLIN N totsnp SP asyn (bound p), p 1. NRE = N tot W SNP SP asyn Under a normal form.

33 Final remarks Asynchronous SNP systems using extended rules are universal; Open question: 4 Asynchronous SNP systems using standard rules only are universal? Their conjecture: No. Plasticity rules can produce at most one spike each step; Plasticity rules with weighted synapses can produce more than one spike each step; Our (non)universality results provide some hint to support their conjecture α {±, } can be another form of synchronization: what if we remove this? A uniform construction? i.e. SUB module construction is independent on a given M. 4 Cavaliere, M., Egecioglu, O., Woodworth, S., Ionescu, I., Păun, G.: Asynchronous spiking neural P systems: Decidability and undecidability. DNA 2008, LNCS vol. 4848, pp (2008)

34 Final remarks Asynchronous SNP systems using extended rules are universal; Open question: 4 Asynchronous SNP systems using standard rules only are universal? Their conjecture: No. Plasticity rules can produce at most one spike each step; Plasticity rules with weighted synapses can produce more than one spike each step; Our (non)universality results provide some hint to support their conjecture α {±, } can be another form of synchronization: what if we remove this? A uniform construction? i.e. SUB module construction is independent on a given M. 4 Cavaliere, M., Egecioglu, O., Woodworth, S., Ionescu, I., Păun, G.: Asynchronous spiking neural P systems: Decidability and undecidability. DNA 2008, LNCS vol. 4848, pp (2008)

35 Final remarks Asynchronous SNP systems using extended rules are universal; Open question: 4 Asynchronous SNP systems using standard rules only are universal? Their conjecture: No. Plasticity rules can produce at most one spike each step; Plasticity rules with weighted synapses can produce more than one spike each step; Our (non)universality results provide some hint to support their conjecture α {±, } can be another form of synchronization: what if we remove this? A uniform construction? i.e. SUB module construction is independent on a given M. 4 Cavaliere, M., Egecioglu, O., Woodworth, S., Ionescu, I., Păun, G.: Asynchronous spiking neural P systems: Decidability and undecidability. DNA 2008, LNCS vol. 4848, pp (2008)

36 Final remarks Asynchronous SNP systems using extended rules are universal; Open question: 4 Asynchronous SNP systems using standard rules only are universal? Their conjecture: No. Plasticity rules can produce at most one spike each step; Plasticity rules with weighted synapses can produce more than one spike each step; Our (non)universality results provide some hint to support their conjecture α {±, } can be another form of synchronization: what if we remove this? A uniform construction? i.e. SUB module construction is independent on a given M. 4 Cavaliere, M., Egecioglu, O., Woodworth, S., Ionescu, I., Păun, G.: Asynchronous spiking neural P systems: Decidability and undecidability. DNA 2008, LNCS vol. 4848, pp (2008)

37 Final remarks Asynchronous SNP systems using extended rules are universal; Open question: 4 Asynchronous SNP systems using standard rules only are universal? Their conjecture: No. Plasticity rules can produce at most one spike each step; Plasticity rules with weighted synapses can produce more than one spike each step; Our (non)universality results provide some hint to support their conjecture α {±, } can be another form of synchronization: what if we remove this? A uniform construction? i.e. SUB module construction is independent on a given M. 4 Cavaliere, M., Egecioglu, O., Woodworth, S., Ionescu, I., Păun, G.: Asynchronous spiking neural P systems: Decidability and undecidability. DNA 2008, LNCS vol. 4848, pp (2008)

38 Final remarks Asynchronous SNP systems using extended rules are universal; Open question: 4 Asynchronous SNP systems using standard rules only are universal? Their conjecture: No. Plasticity rules can produce at most one spike each step; Plasticity rules with weighted synapses can produce more than one spike each step; Our (non)universality results provide some hint to support their conjecture α {±, } can be another form of synchronization: what if we remove this? A uniform construction? i.e. SUB module construction is independent on a given M. 4 Cavaliere, M., Egecioglu, O., Woodworth, S., Ionescu, I., Păun, G.: Asynchronous spiking neural P systems: Decidability and undecidability. DNA 2008, LNCS vol. 4848, pp (2008)

39 Final remarks Asynchronous SNP systems using extended rules are universal; Open question: 4 Asynchronous SNP systems using standard rules only are universal? Their conjecture: No. Plasticity rules can produce at most one spike each step; Plasticity rules with weighted synapses can produce more than one spike each step; Our (non)universality results provide some hint to support their conjecture α {±, } can be another form of synchronization: what if we remove this? A uniform construction? i.e. SUB module construction is independent on a given M. 4 Cavaliere, M., Egecioglu, O., Woodworth, S., Ionescu, I., Păun, G.: Asynchronous spiking neural P systems: Decidability and undecidability. DNA 2008, LNCS vol. 4848, pp (2008)

40 Final remarks Asynchronous SNP systems using extended rules are universal; Open question: 4 Asynchronous SNP systems using standard rules only are universal? Their conjecture: No. Plasticity rules can produce at most one spike each step; Plasticity rules with weighted synapses can produce more than one spike each step; Our (non)universality results provide some hint to support their conjecture α {±, } can be another form of synchronization: what if we remove this? A uniform construction? i.e. SUB module construction is independent on a given M. 4 Cavaliere, M., Egecioglu, O., Woodworth, S., Ionescu, I., Păun, G.: Asynchronous spiking neural P systems: Decidability and undecidability. DNA 2008, LNCS vol. 4848, pp (2008)

41 Thank you for your attention! End!

Notes on Spiking Neural P Systems and Finite Automata

Notes on Spiking Neural P Systems and Finite Automata Francis George C. Cabarle 1, Henry N. Adorna 1, Mario J. Pérez-Jiménez 2 1 Department of Computer Science University of the Philippines Diliman Diliman,