## Dynamical Recognizers: Real-time Language Recognition by Analog Computers (1996)

### Cached

### Download Links

- [acoma.santafe.edu]
- [ftp.santafe.edu]
- [ftp.santafe.edu]
- DBLP

### Other Repositories/Bibliography

Venue: | Theoretical Computer Science |

Citations: | 56 - 4 self |

### BibTeX

@ARTICLE{Moore96dynamicalrecognizers:,

author = {Cristopher Moore},

title = {Dynamical Recognizers: Real-time Language Recognition by Analog Computers},

journal = {Theoretical Computer Science},

year = {1996},

volume = {201},

pages = {99--136}

}

### Years of Citing Articles

### OpenURL

### Abstract

We consider a model of analog computation which can recognize various languages in real time. We encode an input word as a point in R d by composing iterated maps, and then apply inequalities to the resulting point to test for membership in the language. Each class of maps and inequalities, such as quadratic functions with rational coefficients, is capable of recognizing a particular class of languages; for instance, linear and quadratic maps can have both stack-like and queue-like memories. We use methods equivalent to the VapnikChervonenkis dimension to separate some of our classes from each other, e.g. linear maps are less powerful than quadratic or piecewise-linear ones, polynomials are less powerful than elementary (trigonometric and exponential) maps, and deterministic polynomials of each degree are less powerful than their non-deterministic counterparts. Comparing these dynamical classes with various discrete language classes helps illuminate how iterated maps can...

### Citations

10921 |
Computers and Intractability: A Guide to the Theory of NP-Completeness
- Garey, Johnson
(Show Context)
Citation Context ...lowing way: in R d with d fixed, the VC dimension of the family \Phi f \Gamma1 w (H yes ) fi fi jwjsl \Psi is O(l) for polynomial maps and fixed for linear maps (this also follows from the results in =-=[13]-=-), while L 7 and L unary require a VC dimension of at least n l and l respectively. Unfortunately, arguments about independent sets don't seem capable of proving our conjecture that L Dyck = 2 Lin(Z),... |

3836 |
J.D.: Introduction to automata theory, languages, and computation
- Hopcroft, Motwani, et al.
(Show Context)
Citation Context ...(n k )) = BP(PR ) (Blum, Shub and Smale [2]) NPiecePoly(R)TIME(O(n k ))SPACE(O(n k )) = BP(NDPR ) (Cucker and Matamala [8]) PieceLin(R)TIME(O(n k ))SPACE(O(n k )) = BP(P ! lin ) (Meer [24] and Koiran =-=[19]-=-) PieceLin(R)TIME(O(n k ))SPACE(O(1)) = NET \Gamma P (Siegelmann and Sontag [33]) PiecePoly k (Z)TIME(n)SPACE(O(1)) = PiecePoly k (Z) (dynamical recognizers) and similarly for other complexity classes... |

2343 | Computational Complexity
- Papadimitriou
- 1994
(Show Context)
Citation Context ...i and ignore the others, and check that v = w i . (Another language we could use here is fw"k j w k = 1g.) Several corollaries follow from theorem 7, using arguments almost identical to those use=-=d in [32]-=- for the deterministic real-time languages TIME(n): Corollary 1. Poly, Poly k for all k, and PieceLin are properly contained in NPoly, NPoly k , and NPieceLin respectively, for both U = Z and R. Corol... |

1328 |
Nonlinear oscillations, dynamical systems, and bifurcations of vector fields
- Guckenheimer, Holmes
- 1983
(Show Context)
Citation Context ...alculating a polynomial function consists of a fixed number of multiplications and additions. Multi-tape Turing machines can add m-digit numbers in time O(m) and multiply them in O(m log m log log m) =-=[18]-=-. The number of digits of the result is O(km). Iterating such a polynomial n times on initial values x 0 with a fixed number of digits, then, gives us O(k n ) digits and a total time O(k n n log n). A... |

624 |
Learnability and the vapnik-chervonenkis dimension
- Blumer, Ehrenfeucht, et al.
- 1989
(Show Context)
Citation Context ...ere is an interesting connection between theorems 7, 9 and 10 and computational learning theory. The Vapnik-Chervonenkis dimension has been used to quantify the difficulty of learning sets by example =-=[3]-=-; for a family of sets F, the VC dimension is the size of the largest independent family S ae F . Then our arguments about independent sets can be re-stated in the following way: in R d with d fixed, ... |

377 |
On a theory of computation and complexity over the real numbers
- Blum, Shub, et al.
- 1989
(Show Context)
Citation Context ...n what kinds of maps are allowed. Thirdly, there is an increasing amount of interest in models of analog computation, such as Blum, Shub, and Smale's flowchart machines with polynomial maps and tests =-=[2]-=- and other models [24] with linear or trigonometric maps as their elementary operations. In this context, dynamical recognizers form a hierarchy of analog computers with varying sets of elementary ope... |

248 |
Seminumerical Algorithms
- Knuth
- 1998
(Show Context)
Citation Context ...ber of components of each map [13], branching could conceivably be simulated by polynomials of degree jk. Finally, we note that combining the above with results of Cucker and Grigoriev [9] and Koiran =-=[19, 20]-=- give us bounds on Poly and NPoly tighter than theorem 10, as well as bounds on PieceLin(R) and NPieceLin(R). Recall [30] that a machine has polynomial advice if it has access to an oracle whose advic... |

210 | The induction of dynamical recognizers
- Pollack
- 1991
(Show Context)
Citation Context ...2 = P. 30 9 Relationships with other models of analog computation. There are several differences between Blum, Shub and Smale's (BSS) analog machines [2], Siegelmann and Sontag's (SS) neural networks =-=[33]-=-, and dynamical recognizers. First, BSS-machines can branch on polynomial inequalities during the course of the computation. Except for NLin, our recognizers have completely continuous dynamics except... |

114 |
Transductions and Context-Free Languages. Teubner Studienbucher
- Berstel
- 1979
(Show Context)
Citation Context ...osed under inverse homomorphism,sffl-free homomorphism, and intersection with a regular language. (For a formal treatment of trios and other families of languages closed under various operations, see =-=[2]-=-.) Then we have shown that Theorem 1. NLin;NPoly and NPieceLin are trios. Proof. Lemma 3 applies since all these classes are closed under composition. Lemmas 1, 5, and 6 also apply. The interleave of ... |

103 |
The dynamics of discrete-time computation, with application to recurrent neural networks and finite state machine extraction
- Casey
- 1996
(Show Context)
Citation Context ...maps in the same way. Henceforth we will simply refer to Lin(Z), Poly(Z), etc. 5.1 Queues and stacks We now explore the specific abilities of the first few classes. A k-tape real-time queue automaton =-=[6]-=- is a finite-state machine with access to k queues. The queues are first-in-first-out (FIFO), so that the machine can add symbols at one end (say the right), but only read them at the other (say the l... |

90 |
Unpredictability and undecidability in dynamical systems
- Moore
- 1990
(Show Context)
Citation Context ...l can it do? For instance, we will show that linear and quadratic maps can have both stack-like (last in, first out) and queue-like (first in, first out) memories. Secondly, a number of recent papers =-=[26, 28, 7, 1]-=- have shown that various kinds of iterated maps (piecewise-linear, differentiable, C 1 , analytic, etc.) in low dimensions are capable of various kinds of computation, including simulation of universa... |

87 | Analog computation via neural networks - Siegelmann, Sontag - 1994 |

74 | JL: Language as a dynamical system
- Elman
- 1995
(Show Context)
Citation Context ...elow that dynamical recognizers can be thought of as off-line BSS-machines with constant space. Finally, recurrent neural networks are being studied as models of language recognition [31] for regular =-=[12]-=-, context-free [10, 12], and context-sensitive [34] languages, as well as fragments of natural language [?], where grammars are represented dynamically rather than symbolically. The results herein the... |

72 | Lower Bounds for Approximation by Nonlinear Manifolds - Warren - 1968 |

39 | Learning to count without a counter: A case study of dynamics and activation landscapes in recurrent networks
- Wiles, Elman
- 1995
(Show Context)
Citation Context ... recognizers can be thought of as off-line BSS-machines with constant space. Finally, recurrent neural networks are being studied as models of language recognition [36] for regular [16], context-free =-=[13, 41]-=-, and context-sensitive [39] languages, as well as fragments of natural language [14], where grammars are represented dynamically rather than symbolically. The results herein then represent upper and ... |

29 | Closed-form analytic maps in one and two dimensions can simulate Turing machines
- Koiran, Moore
- 1999
(Show Context)
Citation Context ...e, fnja n 2 L ae g is a quasiperiodic sequence and so L ae = 2 CCF. Alternately, consider the language L copy = fwawg, which is in QA 1 and Lin(Z) but not CCF (this follows easily from the results in =-=[23]). Corolla-=-ry 2. NTIME(O(n)) ` NPieceLin(Z)"NPoly 2 (Z) and TIME(n) ` PieceLin(Z) " NPoly 2 (Z). Proof. We have shown that (deterministic) piecewise-linear and non-deterministic quadratic maps can simu... |

27 |
Computing over the reals with addition and order
- Koiran
- 1994
(Show Context)
Citation Context ... iterates the map and x 0 = xw . Doing this with integer coefficients as in theorem 3 requires one more variable. Elementary functions in two dimensions can also simulate Turing machines in real time =-=[21]-=-. Several questions suggest themselves. Is emptiness decidable for Lin(Z) if d = 1? Is membership still P-complete for PieceLin(Z) if ds2, or for PieceLin(Q) if d = 1? Is membership P-hard for Poly k ... |

25 |
Quasi-realtime languages
- Book, Greibach
- 1970
(Show Context)
Citation Context ... deques in the literature, but I prefer to call them quacks. FSAs with access to a finite number of quacks can simulate multi-tape Turing machines in real time, and vice versa [22]. Book and Greibach =-=[4]-=- showed that in the non-deterministic case, real time (n) is equivalent to linear time (O(n)); furthermore, NTIME(O(n)) is precisely the images of languages in CCF under alphabetic homomorphisms, and ... |

25 | Descriptive complexity theory over the real numbers
- Grädel, Meer
- 1996
(Show Context)
Citation Context ...l x falls into as f a is iterated; for f a (x) = 2x mod 1 this is just x 0 's binary digit sequence. If H yes = [1=2; 1], then, L ae = fa t js t = 1g. Both these maps, have complete symbolic dynamics =-=[16]-=-, i.e. there is an x 0 for every possible itinerary; so we can get any L ae ae a we want by properly choosing x 0 . Corollary. The class C(R) properly contains C(Z) for C = Lin, NLin, PieceLin, NPiece... |

24 | On the power of real Turing machines over binary inputs
- Grigoriev
- 1997
(Show Context)
Citation Context ... recognizers can be thought of as off-line BSS-machines with constant space. Finally, recurrent neural networks are being studied as models of language recognition [31] for regular [12], context-free =-=[10, 12]-=-, and context-sensitive [34] languages, as well as fragments of natural language [?], where grammars are represented dynamically rather than symbolically. The results herein then represent upper and l... |

17 | A recurrent network that performs a context-sensitive prediction task - Steijvers, Grünwald - 1996 |

16 |
Computability properties of low-dimensional dynamical systems
- Koiran, Cosnard, et al.
- 1994
(Show Context)
Citation Context ...ating an additional map f comp , in the same class as the f a , until x falls into some subset H halt . Thirdly, restrict BSS-machines to their Boolean part BP and to digital nondeterminism, e.g. DNP =-=[8]-=-. And finally, define an off-line BSS-machine as one who receives its input dynamically in the first n steps, and which has a bound SPACE(f(n)) on the number of variables it can use during the computa... |

16 |
Real-Time Definable Languages
- Rosenberg
- 1967
(Show Context)
Citation Context ...s off-line BSS-machines with constant space. Finally, recurrent neural networks are being studied as models of language recognition [31] for regular [12], context-free [10, 12], and context-sensitive =-=[34]-=- languages, as well as fragments of natural language [?], where grammars are represented dynamically rather than symbolically. The results herein then represent upper and lower limits on the grammatic... |

14 |
On digital nondeterminism
- Cucker, Matamala
- 1996
(Show Context)
Citation Context ...re j is the number of components of each map [13], branching could conceivably be simulated by polynomials of degree jk. Finally, we note that combining the above with results of Cucker and Grigoriev =-=[9]-=- and Koiran [19, 20] give us bounds on Poly and NPoly tighter than theorem 10, as well as bounds on PieceLin(R) and NPieceLin(R). Recall [30] that a machine has polynomial advice if it has access to a... |

12 |
number models under various sets of operations
- Meer, Real
- 1993
(Show Context)
Citation Context ...l can it do? For instance, we will show that linear and quadratic maps can have both stack-like (last in, first out) and queue-like (first in, first out) memories. Secondly, a number of recent papers =-=[26, 28, 7, 1]-=- have shown that various kinds of iterated maps (piecewise-linear, differentiable, C 1 , analytic, etc.) in low dimensions are capable of various kinds of computation, including simulation of universa... |

11 |
New real-time simulations of multihead tape units
- Leong, Seiferas
- 1981
(Show Context)
Citation Context ...are allowed. Thirdly, there is an increasing amount of interest in models of analog computation, such as Blum, Shub, and Smale's flowchart machines with polynomial maps and tests [2] and other models =-=[24]-=- with linear or trigonometric maps as their elementary operations. In this context, dynamical recognizers form a hierarchy of analog computers with varying sets of elementary operations. We show below... |

9 |
Qrt fifo automata, breadth-first grammars and their relations
- Cherubini, Citrini, et al.
- 1991
(Show Context)
Citation Context ...l can it do? For instance, we will show that linear and quadratic maps can have both stack-like (last in, first out) and queue-like (first in, first out) memories. Secondly, a number of recent papers =-=[26, 28, 7, 1]-=- have shown that various kinds of iterated maps (piecewise-linear, differentiable, C 1 , analytic, etc.) in low dimensions are capable of various kinds of computation, including simulation of universa... |

7 |
An infinite hierarchy of intersections of context-free languages
- Liu, Weiner
- 1973
(Show Context)
Citation Context ... much worth studying. It ought to be possible to prove space hierarchy theorems within each time class analogous to those for Turing machines (constant space is universal if unlimited time is allowed =-=[25]-=-). Gradel and Meer have given a logical description of the polynomial time, constant space class [14]. In figure 3 we summarize the inclusions between language classes, both dynamical and discrete, th... |

6 |
Intersections of some families of languages, in
- BRANDENBURG
- 1985
(Show Context)
Citation Context ...w that these containments are proper, consider the language of palindromessL pal = fwaw R g, where w R means w in reverse order (we assume w is in an alphabet not including a). This language is known =-=[5]-=- not to be in NQA; we will show it is in Lin(Z). By using push left w = (pop left w ) \Gamma1 , we can push symbols on to the left end of the queue, i.e. the most significant digits of q; do this for ... |

6 |
Differential fields, machines over the real numbers and automata
- Michaux
- 1991
(Show Context)
Citation Context ...maps can read the top stack symbol of (deterministic) PDAs, while for non-obstinate PDAs quadratic maps need to guess. The fact that CF ae NPieceLin(Z) and DCF ae PieceLin(Z) was essentially shown in =-=[27]-=- and [1]. The closure of CF (DCF, OCF) under intersection and union is the class of concurrent (deterministic, obstinate) context-free languages, or CCF (CDCF, COCF). They are recognized by (determini... |

5 |
A note on undecidable properties of formal languages
- Greibach
- 1968
(Show Context)
Citation Context ... [29], in a dynamical, rather than digital, way. 3 Discrete computation classes We will relate our dynamical classes to the following language classes from the standard theory of discrete computation =-=[17, 30]-=-: Reg, the regular languages, are recognizable by finite-state automata (FSAs), and representable by expressions using concatenation, union, and the Kleene stars(iteration 0 or more times). For instan... |

5 |
The rational index: a complexity measure for languages
- Boasson, Courcelle, et al.
- 1981
(Show Context)
Citation Context ... natural complete problems? (I thank the referee for suggesting this question.) 8) What relation, if any, do our classes Lin, Poly and Elem have to functionvalued complexity measures such as those in =-=[5]-=- and [35]? The VC-dimension argument, for instance, can be seen as a refinement of the latter; they count the number of equivalence classes where two words are equivalent if they can be followed by th... |

4 |
Learning and Extracting Finite-State Automata with Second-Order Recurrent Neural Networks
- Giles, Miller, et al.
- 1992
(Show Context)
Citation Context ...lly, define an off-line BSS-machine as one who receives its input dynamically in the first n steps, and which has a bound SPACE(f(n)) on the number of variables it can use during the computation. (In =-=[14]-=- these are called separated input and output or SIO-BSS-machines.) Then we can look at these classes in a unified way: PiecePoly(R)TIME(O(n k ))SPACE(O(n k )) = BP(PR ) (Blum, Shub and Smale [2]) NPie... |

3 |
Smooth one-dimensional maps of the interval and the real line capable of universal computation." Santa Fe Institute Working Paper 9301
- Moore
- 1993
(Show Context)
Citation Context ... [29], in a dynamical, rather than digital, way. 3 Discrete computation classes We will relate our dynamical classes to the following language classes from the standard theory of discrete computation =-=[17, 30]-=-: Reg, the regular languages, are recognizable by finite-state automata (FSAs), and representable by expressions using concatenation, union, and the Kleene stars(iteration 0 or more times). For instan... |

1 |
Computational complexity of piecewise linear maps of the interval
- Bartlett, Garzon
(Show Context)
Citation Context |

1 |
G.Z.Sun, "Using prior knowledge in an NNPDA to learn languages
- Das, Giles
- 1993
(Show Context)
Citation Context ...ll refer to these problems as emptiness and membership respectively; we will show that even for the simplest classes, they are undecidable or intractable. For definitions of Pand NP-completeness, see =-=[11]-=-. 29 Theorem 21. Emptiness is undecidable for Lin(Z) if ds2, and for Elem(Z) for all d. Proof. Post's Correspondence Problem (PCP) is the following: given a list of words w i and u i , is there a sequ... |

1 |
Bounding the Vapnik-Chevonenkis dimension of concept classes parametrized by real numbers
- Goldberg, Jerrum
- 1995
(Show Context)
Citation Context ...e.g. L 1 ` L 2 if and only if L 1 "L 2 = ;). Corollary 3. For languages in Lin(Z), it is undecidable whether L is regular, context-free, DCF, QA, NQA, etc. Proof. This follows from Greibach's The=-=orem [15, 17]-=-, which states that virtually any non-trivial property is undecidable for a class which is closed under concatenation with a regular language (concatenation of languages with disjoint alphabets suffic... |

1 |
A weak version of the Blum, Shub and Smale model." DIMACS
- Koiran
- 1994
(Show Context)
Citation Context ...d double-ended queues or deques in the literature, but I prefer to call them quacks. FSAs with access to a finite number of quacks can simulate multi-tape Turing machines in real time, and vice versa =-=[22]-=-. Book and Greibach [4] showed that in the non-deterministic case, real time (n) is equivalent to linear time (O(n)); furthermore, NTIME(O(n)) is precisely the images of languages in CCF under alphabe... |

1 |
Generalized one-sided shifts and maps of the interval." Nonlinearity 4
- Moore
- 1991
(Show Context)
Citation Context ..., with varying sorts of nonlinearities. Perhaps these are `baby steps' 4 toward understanding the cognitive processes of experience, imagination, and communication, so important to our everyday lives =-=[29]-=-, in a dynamical, rather than digital, way. 3 Discrete computation classes We will relate our dynamical classes to the following language classes from the standard theory of discrete computation [17, ... |

1 |
Her er en lille gris. Gyldendal
- Nilsson, Hald
- 1994
(Show Context)
Citation Context ... particular subset of the space; the set of words we accept forms a language recognized by the system. We will call such systems dynamical recognizers; they were formally defined by Jordan Pollack in =-=[31]-=-. To define them formally, we will use the following notations (slightly different from his): A is the set of finite words in an alphabet A, with ffl the empty word. If w is a word in A , then jwj is ... |

1 |
Analog Computation Via Neural Networks. " Theoretical Computer Science 131
- Siegelmann, Sontag
- 1994
(Show Context)
Citation Context ... " 0 @ " v=2Lu f \Gamma1 v (H yes ) 1 A Since any such intersection is therefore non-empty, the set of sets f \Gamma1 v (H yes ) over any finite set of words w is independent. Now a theorem =-=of Warren [35]-=- states that m polynomials of degree k can divide R d into at most (4emk=d) d components if msd. If this number is less than 2 m , then not all these sets can be independent. Suppose ae is polynomial ... |

1 |
A class of measures on formal languages." Acta Informatica 9
- Paradaens, Vincke
- 1977
(Show Context)
Citation Context ... complete problems? (I thank the referee for suggesting this question.) 8) What relation, if any, do our classes Lin, Poly and Elem have to functionvalued complexity measures such as those in [5] and =-=[35]-=-? The VC-dimension argument, for instance, can be seen as a refinement of the latter; they count the number of equivalence classes where two words are equivalent if they can be followed by the same su... |