Quantum-Like Networks: An Approach To Neural Behavior Through Their Mathematics and Logic 1st Edition Stephen A. Selesnick
Quantum-Like Networks: An Approach To Neural Behavior Through Their Mathematics and Logic 1st Edition Stephen A. Selesnick
Quantum-Like Networks: An Approach To Neural Behavior Through Their Mathematics and Logic 1st Edition Stephen A. Selesnick
https://ebookmeta.com/product/primary-mathematics-3a-hoerst/
https://ebookmeta.com/product/quantum-mechanics-axiomatic-
approach-and-understanding-through-mathematics-volume-1-1st-
edition-tapan-kumar-das/
https://ebookmeta.com/product/quantum-computing-a-pathway-to-
quantum-logic-design-2nd-edition-hafiz-md-hasan-babu/
https://ebookmeta.com/product/einsteins-quantum-error-an-
approach-to-rationality-1st-edition-simon-altmann/
Neural Assemblies An Alternative Approach to Classical
Artificial Intelligence Günther Palm
https://ebookmeta.com/product/neural-assemblies-an-alternative-
approach-to-classical-artificial-intelligence-gunther-palm/
https://ebookmeta.com/product/state-of-the-art-in-neural-
networks-and-their-applications-volume-1-1st-edition-ayman-s-el-
baz/
https://ebookmeta.com/product/quantum-computing-an-applied-
approach-hidary/
https://ebookmeta.com/product/quantum-communication-quantum-
networks-and-quantum-sensing-1st-edition-ivan-djordjevic/
https://ebookmeta.com/product/machine-learning-with-neural-
networks-an-introduction-for-scientists-and-engineers-bernhard-
mehlig/
Quantum-like
Networks
An Approach to Neural
Behavior through their
Mathematics and Logic
| Stephen A Selesnick
University of Missouri-St Louis, USA
World Scientific
NEW JERSEY • LONDON • SINGAPORE • BEIJING • SHANGHAI • HONG KONG • TAIPEI • CHENNAI • TOKYO
For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance
Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy
is not required from the publisher.
Printed in Singapore
For
Robert A. Taylor
and
Joe Z. Tsien
Introduction
ix
July 18, 2022 15:9 Quantum-like Networks - 9in x 6in b4780-fm page x
x Quantum-like Networks
new. The logic that results is close to the familiar Boolean logic
of the classical neural net model of McCulloch and Pitts, differing
from it in one respect only, namely in that disjunction does not
distribute over conjunction (and, necessarily, vice versa). This has
the consequence that disjunction in the logic is not truth functional:
the proposition p OR q may be valid without either p or q being
valid. Here OR is the disjunction in the logic at hand. It is this
property that is characteristic of the logic of the quantum world and
the sole reason we have used the term quantum-like to describe our
networks. To this property may be ascribed most of the discomfort
experienced by classical thinkers when confronting actual quantum
theory. Disjunction manifests in that theory as quantum superpo-
sition, and is responsible for all the apparent puzzles associated
with Schrödinger’s cat, the entanglement of quantum states, etc. In
quantum theory proper, these apparent conundrums followed from
Heisenberg’s formulation of “matrix mechanics” in 1925 and led to a
revolutionary new ontology that is still provoking head scratching
among sensitive physicists. In stark contrast to actual quantum
theory, the quantum-like properties of our model do not require
any new ontologies: rather they manifest in mundane ways, such as
parallelism of computation, and constraints upon probable outcomes.
(It is the aspect of extreme parallelism that lies behind the quest for
a quantum computer.)
Our models follow the tradition of the original McCulloch and
Pitts neural networks in that they are a mix of physical attributes
(the information handling attributes of the nodes) and the logic
associated with the global structure (Boolean, in the classical case).
In our case, both the physics-like attributes and the structural logic
are quantum-like. The physics-like attributes similarly invoke the
internal structure of the ensemble of nodes while the quantum-
like logic invokes the outer structural attributes. The physical
analogue turns out to be a finite (but large) collection of fermion-
like quanta, while the outer logic is a form of computational logic
named for Gentzen. These two approaches to the same object(s)
intermingle in interesting ways. For example, it turns out that our
model is able to simulate the operation of non-synaptic signaling
July 18, 2022 15:9 Quantum-like Networks - 9in x 6in b4780-fm page xi
Introduction xi
Acknowledgments
My thanks, above all, to Nadine Castro, whose depth of knowledge
and insight first primed my interest in the logic of brains.
I will be forever unable to fully discharge a debt of gratitude
to Robert A. Taylor MD, and the Santa Barbara Cottage Hospital
staff, whose extraordinary skill and quick action saved my own neural
networks from a fate not modeled in this work, and without which
nothing would have been modeled in it.
A truly profound debt of gratitude is owed also to Joe Z.
Tsien, whose penetrating insight, beautiful experiments and patient
tutelage of a tyro was — and remains — a major inspiration.
An enormous debt of gratitude is owed to Felicitas Ehlen for a
very long and very illuminating correspondence, which was directly
responsible for Chapter 6, though any errors are mine alone.
I am tremendously grateful to many others whose help was
instrumental. These include my collaborators on earlier work on
some of the material treated here, namely: Gareth Owen, Gualtiero
Piccinini and John Piers Rawling.
To legions of anonymous referees I extend my warmest thanks
and apologies.
My no less grateful thanks for much additional help and insights,
in the form of generous correspondence or word of mouth, to: C.
Anthony Anderson, Wallace Arthur, Sonya Bahar, Jasper Brener,
Peter Bruza, Nadine Castro, Felicitas Ehlen, Ortwin Fromm, James
Hartle, Miriam Munson, Ronald Munson, Emmanuel Pothos, and
Robert A. Taylor.
To Ivan Selesnick for manifold forms of help over a period of
years — editorial, mathematical, biomedical and typographical —
my grateful thanks. He is also responsible for Figure 5.4. My thanks
to the multitalented polyglot Haifa Nouaime for her meticulous
translations from the French of the Proust passage in Chapter 4,
and elsewhere, and lessons in Arabic for Chapter 6.
The typesetting would have been impossible without the use of
the splendid LaTeX editor Texpad. My thanks to the Texpad team:
Jawad Deo and Duncan Steele.
July 18, 2022 15:9 Quantum-like Networks - 9in x 6in b4780-fm page xiii
Introduction xiii
With the exceptions of Figures 5.2, 5.3 and 5.4 the diagrams
were made with the superb general drawing program OmniGraffle.
My thanks to the Omni Group for this remarkable program.
To say that the literature on almost every neuroscientific, logical,
linguistic and biolinguistic topic touched upon here is oceanic would
be a vast understatement: galactic would be a more appropriate
descriptor. Consequently, I must beg the forgiveness of those mul-
titudes whose work should have been cited but has been overlooked.
S. A. S.
Santa Barbara, California, USA
March 30, 2022
B1948 Governing Asia
xv
B1948 Governing Asia
Contents
Introduction ix
xvii
July 18, 2022 15:9 Quantum-like Networks - 9in x 6in b4780-fm page xviii
Contents xix
xx Quantum-like Networks
Contents xxi
Bibliography 318
Index 325
B1948 Governing Asia
Semantic Overloads
In cross-disciplinary studies such as this one, there often arises the
phenomenon of semantic overload: the same or similar words being
used with different technical meanings in different contexts. Our
subject is particularly prone to this source of confusion. To nip it in
the bud, we shall draw attention to these possibilities sooner rather
than later. To wit:
• Projection
This term is used in mathematics to describe certain operators
upon linear spaces, namely the ones that project every vector upon
a subspace of the space. In neuroscience a projection generally
refers to an axon or bundle of axons that extends from a neuron
or ensemble of neurons, to a neuron or ensemble of neurons at
some distance from the upstream neuron or neurons.
• Convergence
The mathematical notion is assumed to be familiar to the reader.
The neurological term generally applies to a number of axons
emanating from different neurons meeting at, or converging upon,
a common neuron.
• Proof
In addition to its usual usage, this term is also applied in the
area of proof theory as, epitomized by a Gentzen sequent calculus,
another word for deduction.
xxiii
July 18, 2022 15:9 Quantum-like Networks - 9in x 6in b4780-fm page xxiv
• Category
There is a strict mathematical meaning to this term and a very
much less strict linguistic meaning to this term. We shall use both,
but the differences will be clear from the contexts.
• Clique
In graph theory a clique is formally defined as a graph in which
every pair of distinct nodes are adjacent, i.e. are at the ends of a
single edge. In Chapter 5 its usage is informal.
• Motif
There is a formal graph theoretic definition which we shall not
need in this work. In Chapter 5 its use is informal.
• State
The word “state” has many connotations in the case of both
classical and quantum-like systems. In classical systems states are
usually unambiguously sets of points or elements in some ambient
parameter space, such as the phase spaces of mechanics. In the
case of actual quantum theory the word “state” has an ostensibly
different connotation. A very general view of them is that they
are certain functions defined on certain lattices, or algebras of
operators: then deep theorems, such as that of Gleason, realize
a correspondence between certain of these states — the so-called
“pure” ones — and one-dimensional subspaces, or rays, in certain
Hilbert spaces. In Chapter 1 we shall pursue an entirely logical
approach to the issue of quantum-like behavior. The central class
of objects that arises is the class of ortholattices. Thanks to
the work of R. Goldblatt, and others, this apparently denatured
logical theory has sufficient structure to provide us with built-
in versions of “states” as elements of sets, and precise analogues
of the “pure states”, or one-dimensional subspaces (“rays”), of
actual quantum theory. More generally the logical structure deals
with “propositions” which are the precise analogues of the closed
subspaces of Hilbert spaces, which have the status of propositions
in echt quantum logic.
Further confusion is threatened when we come to our main
application, which is to “parameter windows” of complex systems:
namely, subsets of Euclidean spaces, which consist of actual
July 18, 2022 15:9 Quantum-like Networks - 9in x 6in b4780-fm page xxv
Part I
1
B1948 Governing Asia
Chapter 1
Logical Foundations
There are few entries in the lexicon more abused than the word
“quantum.” It is one of the aims of this foundational chapter to
precisely elucidate why and how we came to use this term in the
context of a study of very complex systems such as those found in
living organisms. This we attempt here by rigorously rehearsing the
logic that arises when we try to cope with complexity by blocking
or ignoring the differences between states of such systems which
would produce negligibly discernible differences of effect. Roughly
speaking, we regard such states as being confusable if their effects
are indistinguishable. (We call this hiding the variables, which may
be a misnomer since it is the differences in the variables that we are
hiding.) We explain in this chapter how this leads to a quantum-
like logic, quite independently of any considerations of quantum
physics itself, except for the purposes of post facto comparisons
and interpretations. This ultimately reduces to a consideration of
subsets of Euclidean spaces from this point of view which will form
the basis of the neuronal network model introduced in subsequent
chapters.
For a purported introductory chapter this first chapter might
appear to be technically daunting, particularly to readers not
interested in the technicalities of the logic involved. For such readers
a guide will be provided in the last paragraph of this section.
The technical formalities are required, as mentioned, to justify the
unfortunate necessity for the use of the often faddish, abused and
3
July 18, 2022 12:52 Quantum-like Networks - 9in x 6in b4780-ch01 page 4
4 Quantum-like Networks
Logical Foundations 5
6 Quantum-like Networks
Logical Foundations 7
8 Quantum-like Networks
are the elements that are “possibly” in E. Cf. the discussion of the
B-modal system in section 1.2.2.
Since, in the case of the identity proximity (and denoting set
complementation by the superscript c) we have
E c = {w ∈ W : ∃ v ∈ E such that w = v}c , (1.2.5)
this becomes the subset (♦E)c . So, in the general case, and within the
limits of confusability, we obtain, as the proper reconstruction of ( )c
according to the rule above, which entails placing the ♦ operator
in front of the set to be constructed “up to confusability,” the
subset
(♦E) := ♦(♦E)c (1.2.6)
as the proper generalization of the Boolean complementation oper-
ation. For a proximity space W, ≈ the sets of the form ♦E, for
E ⊆ W , were shown by J. L. Bell [10, 11] to constitute a complete
ortholattice, with join being the ordinary set union, the concomitant
meet of two elements ♦E and ♦F being the largest subset of the
form ♦( ) contained within ♦E ∩ ♦F , and the complement given by
( ) . An ortholattice is essentially a lattice with all the properties
of a Boolean algebra except the distributive law: cf. Appendix A,
section A.1.3 for definitions and first properties. Bell calls this
lattice the lattice of “parts” of W, ≈ and we shall follow him
in denoting it by Part W since the proximity relation will never
be ambiguous. (See Definition A.1.5 for more on the meet in this
lattice.)
In the case in which the proximity is just =, then the lattice of
parts is just the Boolean algebra of all subsets of W . The “associated”
logic is what is known as the Propositional Calculus (PC) and the
association is that Boolean algebras provide models for the logic, in
the logical sense explained in section 1.2.2. There is a similar logic,
mentioned above, that is modeled by the class of ortholattices: it is
called orthologic (OL). It is a logic strictly weaker than standard
PC, meaning that everything and more that can be proved in OL
can be proved in PC. The difference is that OL does not require the
presence of the distributive law.
July 18, 2022 12:52 Quantum-like Networks - 9in x 6in b4780-ch01 page 9
Logical Foundations 9
Irredeemable complexity
hiding the variables not hiding the variables
generalizing
Proximity spaces Sets of states
out of which are constructed out of which are constructed
generalizing
Parts: E subsets: E
constituting constituting
generalizing
Ortholattices Boolean algebras
modeling modeling
generalizing
Orthologic (OL) Propositional Calculus (PC)
10 Quantum-like Networks
Logical Foundations 11
shall here adopt the policy that the former type of assertion shall
be symbolized by a simple turnstile ( ), usually bearing a subscript,
while the model-based validity assertion shall be symbolized by a
more elaborate turnstile ( , |=).
For example, ordinary PC may be characterized by morphisms
of formulas into Boolean algebras: the completeness theorem in this
case asserts that a formula is provable in PC if and only if its
image under any morphism into any Boolean algebra is the top
element. There is a similar completeness theorem for intuitionistic
logic (IL) with Boolean algebras being replaced by Heyting lattices.
Goldblatt proved a similar completeness theorem for OL, the target
lattices in this case being ortholattices. (Since Boolean algebras are
ortholattices it follows immediately from the respective completeness
theorems that any orthotheorem is also a theorem of PC, but clearly
not conversely: OL is strictly weaker than PC). As noted, in the OL
case the model theory bifurcates in the sense that there is another
kind of model that also characterizes OL, namely a Kripkean one.
From the existence of such models, one finds a different sort of
semantics arising solely from the peculiarities of disjunction, and
it is this semantics that mimics quantum behavior. This is because
in the case of the slightly stronger quantum logic itself, disjunction
is exactly “quantum” superposition, the nexus of most if not all of
the puzzlements classical thinkers experience when confronted with
quantum theory. The existence of these Kripkean models of OL led
Goldblatt to realize that OL itself may be embedded into a well-
studied modal system, namely the B-modal system of Becker [9], B
for “Brouwersche,” which well predates the advent of quantum logic
in 1936 [13]. The same result was obtained almost simultaneously but
independently by Dishkant [28]: see also [24]. The associated Kripke
models for this B-system also provide a semantics for probing the
anomalies of disjunction and it is these we shall focus on, since this
system is well known, and reveals the quantum-like behavior clearly
and simply. In this section we shall give a simplified sketch of the
path to the Modal Embedding Theorem for OL, leaving the details
to Appendix A. (Cf. also [76, 81, 82].)
July 18, 2022 12:52 Quantum-like Networks - 9in x 6in b4780-ch01 page 12
12 Quantum-like Networks
Axioms
O1. α α
O2. α β α
O3. α β β
O4. α ∼∼ α
O5. ∼∼ α α
O6. α ∼α β
Inference Rules
α β β γ
O7.
α γ
α β α γ
O8.
α β γ
α β
O9.
∼β ∼α
July 18, 2022 12:52 Quantum-like Networks - 9in x 6in b4780-ch01 page 13
Logical Foundations 13
α O β (1.2.9)
O β. (1.2.10)
Algebraic orthomodels
The first characterization of OL is in terms of ortholattices. As men-
tioned, these are lattices L, , , 0L , 1L , ( ) satisfying all the axioms
for a Boolean algebra except the axiom purporting distribution of
meet over join, or equivalently join over meet. Please see section
A.1.3 for details. Given an ortholattice L a function vL : Φ0 → L
determines a valuation upon Φ via the recursive definitions
14 Quantum-like Networks
Kripkean orthomodels
These models were first posited by Goldblatt [41] and have been
elaborated upon by Dalla Chiara [24] and others. The model in this
case comprises a proximity space W, ≈ and a function, called a
valuation : Φ → R(W ) satisfying
(α β) = (α)∩ (β) (1.2.13)
(∼ α) = (α)⊥ . (1.2.14)
Here, R(W ) denotes the complete ortholattice of propositions in
W, ≈ as described in section A.1.2, et seq.
Definition 1.2.2. A Kripke orthomodel M := W, ≈, is a
proximity space W, ≈, with a valuation : Φ → R(W ).
We will say that a formula α is:
true at the “world” w ∈ W , and write w |=M α iff w ∈ (α);
true on the set E, and write E |=M α iff w |=M α for all w ∈ E:
that is, iff E ⊆ (α);
true in the Kripke orthomodel M iff it is true at every world in
W , that is iff W |=M α;
Kripke valid, and write |= α iff it is true in all Kripke ortho-
models.
July 18, 2022 12:52 Quantum-like Networks - 9in x 6in b4780-ch01 page 15
Logical Foundations 15
Theorem 1.2.2. O α iff |= α.
B α. (1.2.19)
(The origin of the odd nomenclature in equation (1.2.17) is that
♦ := ¬ ¬ is like a strong form of double negation and the rule in
July 18, 2022 12:52 Quantum-like Networks - 9in x 6in b4780-ch01 page 16
16 Quantum-like Networks
w B α iff ∀x ≈ w, x B α (1.2.20)
w B ♦α iff ∃x ≈ w, x B α. (1.2.21)
July 18, 2022 12:52 Quantum-like Networks - 9in x 6in b4780-ch01 page 17
Logical Foundations 17
Theorem 1.2.3. B α iff α.
18 Quantum-like Networks
6.
7.
Historiallinen katsaus