2026 year, volume 30, issue 4
Download full issue (PDF)
We outline the theory of autogenerative self-systems (TSS), i. e. systems that possess subjectivity, the capacity to feel and a self of their own. TSS derives such systems and their properties from three necessary conditions that act only jointly: generative functional systems, a deep complex network, and cumulative plasticity of its nodes. The emergence of a self-system is described by a recursive self-generation algorithm that transforms a neural network into a neural hypernetwork, and its existence by the neurogeometry of the hypernetwork and the corresponding cognitive geometry.
Keywords: self-systems, subjectivity, functional systems, cumulative plasticity, neural hypernetwork, cognitive geometry.
Download PDFSemi-online algorithms represent an intermediate approach designed to enhance the practical applicability of online and offline scheduling models. Semi-online scheduling is a modified version of online scheduling in which, in addition to the standard input data, extra information (an “Extra Piece of Information,” or EPI) is available, or a buffer is provided for data storage. This paper introduces online and semi-online approaches to partitioning (scheduling in multiprocessor systems) and bin-packing problems, discusses their areas of application and the motivations behind such research, and proposes a general classification of semi-online scheduling models. Furthermore, it outlines the primary methods used in the development and analysis of these algorithms.
Keywords: online version, semi-online version, comparative analysis, guaranteed estimate, lower bound.
Download PDFWe consider Bellman’s problem on the complexity of computing a monomial in \(m\) variables, defined as the minimum number of multiplications required, and Knuth’s problem on the complexity of computing a set of \(m\) powers of a single variable. For every fixed real number \(a\ge 2\) and every positive function \(\varphi (m)\) satisfying \(\lvert \log \varphi (m)\rvert =o(\log m)\), we establish asymptotically tight bounds for the corresponding Shannon functions, which describe the maximum possible complexity when the exponents are bounded by \(A=\varphi (m)m^a\). As \(m\to \infty\), these functions are asymptotic to \(am\) and \((a-1)m\), respectively.
Keywords: addition chains, vector addition chains, Bellman’s problem, Knuth’s problem, circuit complexity.
Download PDFAI is on everyone’s lips these days, yet at the same time it is widely misunderstood. It is often assumed, or even claimed, that AI can at least replace human intelligence and may even become autonomous. However, this lofty claim will never be fulfilled. In reality, AI is merely a tool — an intelligence aid (IA) — and simply a complement to intelligence that can be utilised with competence. Our main proposition demonstrates that human model-based thinking and behaviour cannot be emulated using current AI/IA tools within an IT-based approach. To this end, we shall discuss nine propositions relating to this theorem.
Keywords: AI questions, artificial intelligence, human intelligence, the benefits of AI, AI competence, {\sf The AI incompleteness theorem}.
Download PDFThis paper presents the results of a study on the complexity of certain decision problems—membership, equivalence, unification, and functionality—for top-down tree transducers (TDTs) with output parameters (PTDTs). Membership problem is \(\mathsf {L}\)-complete for deterministic models DTDTs and PDTDTs, belongs to \(\mathsf {LOGCFL}\) for TDTs, and it is \(\mathsf {NP}\)-complete for PTDTs. Equivalence checking of complete DTDTs and PDTDTs can be performed in cubic time. The same time \(O(n^3)\) is required for computing the most general unifier of a pair of PDTDTs as well as for computing the functional instance of a PTDT. Minimization of complete PDTDTs is \(\mathsf {NP}\)-complete in contrast with polynomial-time complexity of minimization problem for DTDTs.
Keywords: tree transducers, decision problems, output parameters, computational complexity.
Download PDFIf the input and output alphabets of an initial automaton coincide, then the number of distinct automata obtained by superposing the original automaton with itself can be either finite or infinite. The problem of determining the order of a finite initial automaton with respect to superposition is algorithmically undecidable in the general case. However, for some classes of automata, algorithms for determining the order can be found.
Keywords: Mealy automata, Post classes, algorithmic decidability.
Download PDFThe paper deals with classes of linear automata over the field \(E_2\) under composition (\(K\)-closure) and approximation (\(A\)-closure) operators. The \(K\)-generation problem is solved for the classes that form the second level of the lattice of \(A\)-closed classes. It is proved that 22 of the 23 such classes have a finite basis, while the remaining class is not \(K\)-finitely generated.
Keywords: finite automaton, linear automaton, closed class, precomplete class, composition operations, $A$-closure, $K$-finite generation, bases of closed classes.
Download PDFWe study the complexity of the addressation problem for one-dimensional idempotent cellular automata with locators. On a tape infinite in both directions, we consider a segment of \(N+1\) marked cells, outside of which all cells are in the quiescent state. Given a transmitted address, it is required to single out the cell with this address. We establish the order of complexity: with a constant number of states, the addressation problem is solvable in \(O(\log N)\) steps.
Keywords: cellular automata with locators, addressation problem, idempotency, complexity.
Download PDFThis paper considers three types of discrete dynamic systems. These are automata systems, computational systems and transition systems. The equivalence criteria have been obtained for each of these types of systems. For each considered type of systems the problem of modeling it by another kind of systems has been solved. The problem of uniqueness of such modeling has been solved also.
Keywords: Automata systems, computational systems, transition system, strong equivalence.
Download PDFPoints are scattered across a plane. The task is to count the number of points and provide the result in binary form. It has been shown that this problem can be solved using two-dimensional cellular automata with locators, in a time not exceeding \(2]\log _2 n[ + 7\), where \(n\) is the number of scattered points.
Keywords: counting the number of points, cellular automata with locators.
Download PDFThis paper considers a classification of binary one-dimensional homogeneous structures with a standard neighborhood template. Three types of structures are considered: dying, strictly periodic and infinitely growing. In dying structures every finite configuration becomes empty over time. In strictly periodic structures every finite configuration repeats itself after a period. In infinitely growing structures every finite configuration fills the whole line over time. For these three types of structures, all homogeneous structures belonging to each type have been found.
Keywords: homogeneous structures, cellular automata, a standard neighborhood template, infinite growth, strictly periodic, dying.
Download PDFWe prove that the set of upper approximations of a regular language by elements of a given finitely generated monoid of regular languages is a rational effectively constructible set.
Keywords: Regular languages, approximation of languages, rational substitution, representability problem for regular languages, monoids of regular languages.
Download PDFWe study the computational complexity of the limit-cycle existence problem for a fixed length \(k\) in Boolean networks with the parallel update scheme and local functions restricted to classes of Post’s lattice. For every fixed \(k\geqslant 2\), a complete complexity classification is obtained. The problem is shown to be NP-complete for every closed class containing at least one of the classes \(D_2\), \(F_2^\infty\), or \(F_6^\infty\). For all remaining closed classes, the problem is polynomial-time solvable.
Keywords: Boolean networks, limit cycles, Post's lattice, computational complexity, classification, NP-completeness.
Download PDFThis paper considers the application of a one-dimensional cellular automaton with locators, whose broadcast alphabet is equipped with an idempotent monoid structure, to the problem of finding the midpoint of a line segment.
Keywords: idempotent cellular automata with locators, upper and lower bounds, midpoint finding.
Download PDFWe study the computational power of one-dimensional cellular automata with locators. We formalize function computation in this model and prove that the languages recognizable in polynomial time are exactly those in PSPACE. The key step is the construction of a one-dimensional cellular automaton with locators that solves the quantified Boolean formula problem in polynomial time.
Keywords: cellular automata with locators, computational complexity, polynomial time, PSPACE.
Download PDFWe consider closed transition tours of a complete finite deterministic automaton with a fixed initial state. We state a known sharp general bound and describe a family of automata attaining it. We give a criterion for the minimum tour length to equal the number of transitions and show that this equality holds for strongly connected permutation automata and linear automata over a finite field.
Keywords: finite automaton, transition tour, linear automaton, Eulerian circuit.
Download PDFThis paper investigates countably-valued functions computed by classical automata and automata with stops. The properties of encodings that convert numbers into automaton input sequences and vice versa are examined. We identify the classes of functions computable by the specified types of automata under these encodings.
Keywords: computability, automaton.
Download PDFThis work studies completeness problems in the class of linear definite automata over the two-element field. Reduced criterial systems of closed classes are described for approximation completeness and completeness under superposition. These criteria are also used to establish the algorithmic decidability of both completeness problems for finite systems.
Keywords: linear definite automata, approximation completeness, superposition.
Download PDFThe task is to add \(n\) distinct positive integers represented by the coordinates of active cells. It is shown that a two-dimensional cellular automaton with locators solves this problem in \(5\lceil \log _2 n\rceil +1\) steps, independently of the values of the summands.
Keywords: cellular automata, locators, addition.
Download PDFWe investigate the nondeterministic state complexity of the square-root operation \(\sqrt {L}=\{\,w \mid ww\in L\,\}\) on regular languages represented by nondeterministic finite automata. For an \(n\)-state NFA accepting \(L\), it was previously known that \(\sqrt {L}\) can be accepted by an NFA with at most \(n^{3}\) states, while the best lower bound, was only \((n-1)(n-2)(n-3)\). We are going to close this gap completely and prove that \(n^{3}\) states are sufficient and necessary in the worst case.
Keywords: state complexity, language operations, nondeterministic finite automata.
Download PDFWe are developing a computational modeling method for transient biophysical processes with crisis-like transformation effects under the influence of delay effects. During invasions of aggressive species in ecosystems and COVID-19 epidemics, alternating forms of oscillatory activity regimes emerge and decay. We simulate series of invasive waves using scenario-based equations with delays and redefinable oscillation damping functions. We construct simulation models of invasions from equations for stage transitions based on hybrid automata with events that determine the change in the evolutionary adaptation mode of stages of a decaying oscillatory biophysical process from increasing and decreasing waves of sharp population activity. The hybrid model allows us to describe the conditions for the transition of a crisis situation to a new stage in a series of waves of COVID-19 or aggressive invasion.
Keywords: hybrid automata structures, oscillating invasions.
Download PDFThis paper continues the investigation into the complexity of semiautomata realization via states encoding, namely the property of linear realizability of semiautomaton. Encoding of the semiautomaton alphabets induces a Boolean operator. Arguments of the Boolean operator might be separated into two sets. The first one corresponds to the code of the input alphabet, the second one corresponds to the code of the alphabet of states. The property of the linear realizability with respect to each subset of variables can be studied separately.
This paper studies the property of linear realizability with respect to input variables. It is proven that “additive shift” property for generators of the inner semigroup \(p_0\) and \(p_1\) is a necessary and a sufficient condition for semiautomaton to be linearly realizable by input variable regardless of the code length. This means that code length increasing will not induce linear realizability if “additive shift” property is not fulfilled for generators of the inner semigroup.
Keywords: Automata theory, semiautomata, transition systems, permutation, state encoding, complexity, boolean operator.
Download PDFThis paper introduces and explores the concepts of automata-implementable and prefix-regular \(\omega\)-languages. The equality of these classes is shown. Some other properties are also shown, specifically: regularity of set of words and omni-regularity of set of superwords of automata-implementable \(\omega\)-language, the uniqueness of reduced automaton for automata-implementable \(\omega\)-language and some cardinality combinations for set of words and set of superwords of automata-implementable \(\omega\)-language are described.
Keywords: finite automaton, language, regular language, hardware-final automaton, $\omega $-language, prefix-regular $\omega $-language, automata-implementable $\omega $-language.
Download PDFIt is shown that the class of linear Moore automata equipped with an approximation-closure operator contains no precomplete classes and no basis, but does contain a continuum of closed classes.
Keywords: linear automaton, Moore automaton, approximation closure, completeness, expressibility, closed class, precomplete class.
Download PDFIt was shown earlier that the class of finite automata with one linear output is not finitely generated under bounded composition. A much narrower generating set is built from the adder, the constant and conjunctions with a delay. For automata with at most \(s\) inputs a reduced generating set of three automata is found, including a conjunction of \(s+1\) variables with a delay, and it is proved that a smaller arity of the conjunction is insufficient. An upper bound for the complexity of realization of automata by circuits over this set is obtained.
Keywords: finite automata, automata with a linear output function, composition operations, generating set, complexity of realization.
Download PDFThe report examines the problem of the completeness of functional systems of polynomial and rational functions.
Keywords: completeness problem, basis, polynomial.
Download PDFLD-Leiden locally updates the Leiden hierarchy and coordinates parallel vertex moves in dynamic weighted graphs. On nine real-world graphs (up to 50.9M vertices and 117M edges), geometric-mean speedup over repeated Leiden is 25.64, with a mean final modularity ratio of 0.996. On graphs up to 214M vertices and 3.30B edges, scaling from 1 to 64 threads gives a mean speedup of 5.22.
Keywords: dynamic graph, community detection, modularity, Leiden.
Download PDFThe paper proposes a generalization of CRMP polynomial forms, i.e., generalized polynomial forms in which any two monomials differ in the set of variables they contain. The notion of a CRMP polynomial form polarized by a vector \(\alpha \in \{-1,0,1\}^n\) is introduced. Some properties of such forms are described.
Keywords: Boolean functions, polynomial forms, CRMP forms.
Download PDFThe paper considers the problem of finding a simple path of a given length between specified vertices in a connected undirected weighted graph without multiple edges and loops, all edge weights of which are positive numbers. This problem arises when planning transport routes with guaranteed delay or when analyzing graphs that arise in microbiology. The authors propose a new algorithm for solving the problem of finding a simple path of a given length in a graph. The developed algorithm is based on the gradual transformation of the graph and the elimination of simple paths that cannot be part of the sought‑after simple path. The paper presents estimates of the time and space complexity of the new algorithm in the average and worst cases, and also presents the results of numerical experiments comparing this algorithm with the modified Yen’s algorithm.
Keywords: Yen’s algorithm, simple path search, Erdős–Rényi random graph.
Download PDFWe present a hybrid genetic algorithm for one- and multidimensional knapsack problems. Its Rust implementation combines parallel island evolution, greedy chromosome repair, and local search.
Keywords: knapsack problem, multidimensional knapsack problem, genetic algorithm, parallel computing.
Download PDFWe consider two activity measures for Boolean circuits over the standard basis: the power (the number of gates whose output is one) and the \(P\)-activity (the number of gates having one on an input or on the output). For the first measure, the Shannon function for operators with \(n\) inputs and \(m\) outputs is shown to have order \(\Theta (n+m\log n)\) for \(1\le m\le 2^n\). For the second measure, we obtain an \(\Omega (n\log n)\) lower bound for almost all Boolean functions. Both results hold for the average and the maximum measures.
Keywords: Boolean functions, Boolean circuits, circuit power, circuit activity, Shannon function.
Download PDFThe article examines the question of the maximum number of vertices of a convex integer polygon that can simultaneously lie on a Bresenham line when it is translated parallel to a rational vector. Unlike the classical linear programming problem with a non-constant linear objective function, where the extremum is achieved in no more than two vertices, in the discrete case the number of vertices simultaneously lying on a Bresenham line can be arbitrarily large.
Keywords: Bresenham’s line algorithm, convex hull, integer lattice, rational tangent.
Download PDFWe study the minimum number of multiplications and exponentiations to rational powers in the interval \((0,2)\) sufficient to compute an arbitrary system of four monomials in three variables. This number is shown to be \(\log _2 D(A)+O(1)\), where \(A\) is the nonnegative integer matrix of exponents with no zero rows, and \(D(A)\) is the maximum absolute value of all minors of \(A\).
Keywords: circuit complexity, Pippenger's problem, computational complexity, systems of monomials, λ-circuits.
Download PDFIn addition to the canonical method of defining additive characters of a finite field of composite order via the absolute trace function, an alternative approach is proposed, based on the vector-space representation of the field. For both methods, sets of character groups with distinct parameterizations are identified, and the relationships between them are established. Formulas are derived that link the differential and nonlinear properties of mappings over finite fields to the Fourier spectra of their characters.
Keywords: finite field, character, vectorial function, difference spectrum, nonlinearity, Fourier coefficients.
Download PDFA new substitution basis is given for the class of injective functions belonging together with their inverses to the generalization of the class \(\mathcal E^3\) to the case of partial functions. The basis is obtained by arithmetization of reversible counter-summator machines.
Keywords: reversible computation, substitution basis, counter-summator machines.
Download PDFProperties are obtained for functions represented by polynomials modulo the square of a prime number and for the closed class of polynomials modulo the square of a prime number.
Keywords: function on a finite set, finite ring, residue ring, polynomial, closed class.
Download PDFWe propose a new MILP presolve method that reduces nonzeros using near-parallel rows in the constraint matrix. A heuristic efficiently identifies promising row pairs. The method reduced nonzeros in \(30\%\) of instances in the MIPLIB 2017 Benchmark Set after HiGHS presolve.
Keywords: MILP presolve, discrete optimization.
Download PDFA suite of metaheuristic algorithms has been developed to solve the traveling salesperson problem on complete weighted graphs: classical simulated annealing (SA), parallel SA with solution exchange, and simulated quantum annealing (SQA). Computational experiments were conducted using benchmark graphs of varying sizes, alongside a comparative analysis with the LKH algorithm.
Keywords: Traveling Salesman Problem, TSP, metaheuristic algorithms, simulated annealing, simulated quantum annealing, LKH.
Download PDFThis paper addresses the problem of verifying the weight minimality of a bijunctive function that depends essentially on all its variables and has no affine implicits. A polynomial-time algorithm is presented for solving this problem in the case of a minimal 2-CNF.
Keywords: bijunctive function, essential variable, affine implicant, minimal 2-CNF.
Download PDFThis paper proposes a method for simulating Boolean circuits using cellular automata with locators (CALs). Given a CFE with \(k\) elements and a depth of \(d\), the corresponding CAL that simulates it will have a delay of \(4d+1\) and an area on the order of \(k^2\).
Keywords: Boolean circuits, cellular automata with locators.
Download PDFModern intelligent systems perform computational processing over a digital representation of a subject domain. The same subject domain can be represented in different forms. This paper considers the transition from an initial set of digitized information to a data structure in the context of the classical “Model – Algorithm – Program” scheme associated with A.A. Samarskii. A digital representation structure is introduced as the result of organizing the initial information according to a selected model. If \(X\) denotes the initial set of digitized information and \(R\) denotes relations and other structural connections specified by the representation, then the structure is written as \(S=(X,R)\). The role of the structure of digital representation as one of the factors in choosing a processing method, on the basis of which an algorithm for its implementation is selected, is considered. For data science problems, the sequence “digital representation – structure – method – algorithm – program” is considered.
Keywords: artificial intelligence, data science, digital data representation, data structure, databases, processing method, algorithm, mathematical modeling.
Download PDFWe consider the synthesis of Boolean circuits with simultaneous bounds on weighted complexity and switching activity, the maximum number of gates whose values differ between two input assignments. For every finite complete basis \(\mathcal B\) belonging to the union of two special classes and every Boolean function \(f(x_1,\ldots ,x_n)\), we construct a circuit \(\Sigma _f\) over \(\mathcal B\) that computes \(f\). Its weighted complexity \(L(\Sigma _f)\) and switching activity \(C(\Sigma _f)\) satisfy the following bounds for \(n=1,2,\ldots\): \[\begin {gathered} L(\Sigma _f)\le \rho (\mathcal B)\frac {2^n}{n} \left (1+\frac {6\log _2 n}{n}+O\!\left (\frac 1n\right )\right ),\quad C(\Sigma _f)=O(n), \end {gathered}\] where \(\rho (\mathcal B)\) is the minimum reduced weight of the basis elements. The first bound is asymptotically equal to the corresponding Shannon function, while the second is a linear upper bound.
Keywords: Boolean circuits, weighted complexity, switching activity, Shannon function, two-parameter circuit optimization.
Download PDFThe problem of constructing one-sided embeddings of ternary trees of formulas similar to given formulas with negations pushed down to the variables, over a basis of commutative and associative operations, into rectangular grids is considered. For an arbitrary initial formula, estimates are obtained for the depth of a similar formula and for the height of a rectangular grid admitting a canonical embedding of its tree.
Keywords: formula trees; similar formulas; embeddings; rectangular grids;.
Download PDFAn approach to scientific article preparation using independent models of content and publication requirements is proposed. The pipeline combines manuscript parsing, document generation, and validation. The implemented module produces a LaTeX project and, for DOCX input, an editable Word document. All 63 functional tests passed, covering structural processing, object preservation, and detection of predefined violations. The approach enables parsers to be reused across publications and identifies elements requiring editorial review.
Keywords: scientific article, document processing, structure, template, LaTeX, conversion validation.
Download PDFThis paper investigates the possibility of implementing arbitrary Boolean functions using cellular automata with locators. It is proved that, in a two-dimensional cellular automaton with locators, any Boolean function of \(n\) variables can be implemented by a circuit of area \(2^n+O(n2^{n/2})\), with a delay of four time steps and a new input vector supplied at every time step. Configuration counting yields a lower bound of \(\Omega (2^n)\) for almost all Boolean functions. The Shannon function for circuit area is of order \(2^n\).
Keywords: cellular automata, locators, Boolean functions, circuit area, Shannon function.
Download PDFWe consider a problem of synthesis of multi-pole irredundant contact circuits, implementing given Boolean functions between poles \(A\) and \(B\) and allowing short single fault detection tests relative to contact breaks. For each Boolean function and each test pole set \(\mathcal P\), the minimal possible length value of such a test is found. In particular, it is proved that this value does not exceed \(3\) if \(\mathcal P\ne \{\{A,B\}\}\).
Keywords: contact circuit, contact break, additional pole, single fault detection test, Boolean function.
Download PDFThe cardinality of a single fault detection test set with respect to stuck-at faults at gate outputs is studied for Boolean formulas. An asymptotic upper bound for the corresponding Shannon function is established for three bases close to the standard one. For the “Peirce arrow” and “Sheffer stroke” bases, a growing lower bound for this function is obtained, as well as an upper bound (expressed in terms of the formula’s rank) on minimal test set cardinality. Not every Boolean function admits an irredundant formula over the latter two bases.
Keywords: fault detection test set, stuck-at faults, Boolean formula.
Download PDFA modification of the quadratic Osipov–Lanchester law for forces including infantry and expendable FPV drones is proposed. Drone expenditure is described by a separate equation, which closes the model. A general integral state equation with an arbitrary drone employment profile is derived. Analytical victory conditions are obtained for a preliminary massed strike and proportional fire support.
Keywords: Osipov--Lanchester equations, combat modeling, FPV drones, unmanned aerial vehicles, differential equations.
Download PDFIn their paper “Algebraic Properties of Subquasigroups and Construction of Finite Quasigroups”, V. A. Artamonov, S. Chakrabarti, S. K. Tiwari, and V. T. Markov proposed a method for constructing finite quasigroups of order \(q=p^r>4\), where \(p\) is prime and \(r\in \mathbb {N}\), that are polynomially complete and have no proper subquasigroups. The paper proves that quasigroups obtained using the original construction do not satisfy the requirement of shaplesness introduced by S. Markovski and D. Gligoroski with co‑authors. A refinement of this construction is proposed, which makes it possible to preserve the previously established properties and obtain shapless quasigroups.
Keywords: finite quasigroups, polynomial completeness, shapeless quasigroups.
Download PDFThree polynomial multiplication methods in Kyber and NewHope are compared separately and in schemes with unchanged Lizard. Results and confidence intervals are reported.
Keywords: Kyber, NewHope, Lizard, Karatsuba, NTT.
Download PDFFor fixed \(m\ge 2\), we prove that in a uniformly random labelled \(m\)-ary quasigroup of order \(n\), the total number of proper subquasigroups converges in distribution to \(\operatorname {Pois}(1)\) as \(n\to \infty\). The probability of the existence of a proper subquasigroup of order at least two is \(O(n^{-1})\), and the probability of having no proper subquasigroups tends to \(e^{-1}\).
Keywords: m-quasigroup, subquasigroup, Latin square, Poisson distribution.
Download PDFA modification of the Shor–Rivest cryptosystem is considered, distinguished by the fact that computations in a random power basis are simulated using an optimal normal basis of type II, with acceleration achieved through fast multiplication and exponentiation.
Keywords: Shor-Rivest cryptosystem, optimal normal basis of the second type, fast multiplication, raising to a power equal to the field characteristic.
Download PDFIn this paper, we introduce and study the concept of the span of regular languages under alphabetic coding. The main result is a previously unknown dichotomy property of unit and infinite spans for languages of a special iteration-concatenation form in the case of a two-element input alphabet. It is also shown how, based on these results, one can construct an algorithm for computing the span in the general case for such an alphabet (the algorithmic decidability of this computation itself is a known fact).
Keywords: regular language, alphabetic coding, span.
Download PDFThe paper presents an algorithm for checking the shapelessness of finite quasigroups that has quadratic time and space complexity in the order of the quasigroup. In particular, a quadratic algorithm is presented for checking the absence of proper subquasigroups as one of the properties comprising the concept of shapelessness.
Keywords: quasigroup, subquasigroup, shapelessness.
Download PDFWe present an algorithm for constructing a \(k\)-qubit quantum hashing circuit with \(d\) parameters, where \(k \geq 3\) and \(2^{k-2} < d \leq 2^{k-1}\). We obtain the bounds \(N_{1} \leq 2^{k-1} + 11k - 24\) and \(N_{2} \leq 2^{k-1} + 3k - 8\) on the numbers of elementary one-qubit gates and \(\mathrm {CNOT}\) gates, respectively. For \(k \geq 5\), both bounds are lower than the corresponding bounds for the algorithm of Bergholm et al. (2005) in the same gate basis.
Keywords: quantum hashing, quantum circuit.
Download PDFIn this article, we will examine an important cryptographic problem: constructing efficient primality tests used to generate large prime numbers. We improve upon existing methods by using analogues of Fermat’s little theorem.
Keywords: primality test, finite fields, factorization, cryptographic protocols, elliptic curves, Lucas method, Williams’ method.
Download PDFWe study the product expansion property of LDPC codes. It is known that the product of two expander LDPC codes is robustly testable, which in the two-dimensional case is equivalent to product expansion. In this work, we show that the product expansion of any number of expander LDPC codes is bounded below by a positive constant. Since random LDPC codes with certain parameters are expander codes with high probability, our result covers a wide class of codes. Previously, for three or more binary codes, product expansion was known only for locally testable codes and some of their subcodes.
Keywords: tensor product of codes, LDPC codes, expander codes, product expansion.
Download PDFA prototype lightweight blockchain protocol for the Internet of Things is developed. End devices generate messages containing a data hash and an HMAC authentication code, while trusted edge nodes verify messages and form blocks. This division transfers ledger maintenance to more capable nodes. In a comparison performed on the same hardware under identical execution conditions, the prototype processed 5000 events at 4980.52 events/s versus 306.53 for the control model.
Keywords: Internet of Things, blockchain, software prototype, HMAC, edge computing, Merkle tree.
Download PDFWe consider a generalization of the well-known quadratic-linear family of Boolean functions: an arbitrary symmetric Boolean function of all variables except the one with the same index is added to the linear triangular part. It is proved that this family is proper and has the maximum possible image cardinality. Thus, the new family preserves the key property of the known particular case.
Keywords: proper family of functions, Boolean functions, symmetric function, quasigroup, image cardinality.
Download PDFWe consider the SUM-SCS problem: given a dictionary of \(n\) strings, construct a shortest string containing at least \(k\) occurrences of dictionary strings. We study the multiple query version (MSUM-SCS): for a fixed dictionary, find the lengths of shortest superstrings for \(m\) values \(k_1,\dots ,k_m\). We consider the factor-free case when non of the strings can be substring of any another. We present a quantum algorithm with \(O^*(mn^{1.5}\log k_{\max }+n^{2.5}\log k_{\max }+L)\) complexity, a classical one with \(O^*(mn^{2}\log k_{\max }+n^{3}\log k_{\max }+L)\) complexity. Here \(L\) is the total length of the strings, \(k_{\max }=\max _{1\leq r\leq m} k_r\), and \(O^*\) notation hides not only a constant, but also polylogarithmic multiple, i.e. a polynomial on \(\log m, \log n,\log \log k_{\max }, \log L\) .
Keywords: strings, quantum algorithms, query complexity, shortest common superstring.
Download PDFFor a four-qubit GHZ–W superposition, adaptive local projective measurements of two qubits yield a separable pair for every outcome. For all fixed local bases, we prove a strictly positive lower bound on its entropic robustness index.
Keywords: quantum entanglement, projective measurements, GHZ, W.
Download PDFThe paper considers \(n\)-quasigroups in the context of properties relevant to cryptographic applications. For a parameterized family of isotopic transformations, an estimate is given for the number of distinct \(n\)-quasigroups obtained from a given \(n\)-quasigroup by applying different transformations of the family.
Keywords: $n$-quasigroup, isotopy, polynomial completeness, simplicity, non-affineness, subquasigroup.
Download PDFWe consider solving QUBO problems using QAOA with fixed parameters and a circuit number of layers that grows linearly with problem size. Numerical results show that, contrary to expectations, the median number of circuit runs required to obtain a solution with a prescribed approximation ratio does not increase with problem size. The analysis of entanglement reveals an unexpected decrease of its peak value with the number of layers. Guided by this observation, we constructed a fully classical quantum-inspired algorithm without entanglement. On GSet MaxCut benchmarks with up to 20,000 variables, the ratio of the found cut to the best known value exceeds 0.95 in the vast majority of cases.
Keywords: QAOA, QUBO, MaxCut, fixed parameters, entanglement.
Download PDFThe paper deals with ternary operations that are right-invertible. For each such operation, two term operations are constructed, and the three resulting operations are shown to be orthogonal. These three orthogonal operations are then used to transform words.
Keywords: Ternary groupoid, orthogonality, reversibility, transformation.
Download PDFThe paper is devoted to the study of the product expansion coefficient of linear codes — a parameter arising in the study of coboundary expansion and constructions of locally testable and quantum codes. For an \(m\)-dimensional product of linear codes of the same length \(n\), we obtain the general lower bound \(\rho \geq 1/(\lceil m/2\rceil n^{\lceil m/2\rceil })\). In addition, in the three-dimensional case, we construct a family of codes based on Reed–Solomon codes for which the expansion coefficient is of order \(1/n^2\). Thus, for \(m=3\), the obtained lower bound is asymptotically tight.
Keywords: linear codes, product codes, product expansion, coboundary expansion.
Download PDFA content ranking method based on heterogeneous hypergraph models is proposed. A 5-partite hypergraph is constructed on music listening data. Latent behavioral clusters, serving as proxy representations of genre and mood characteristics, are extracted from the interaction matrix using SVD and the \(k\)-means method. Ranking is performed by a generalized spectral operator. In experiments on the Yambda-50M dataset, an NDCG@10 value of 0.469 is achieved, outperforming PureSVD by 7.1 % and ItemKNN by 23.6 %.
Keywords: hypergraph, recommender systems, spectral methods, music recommendation, Yambda.
Download PDFOn two LLMs (DeepSeek, MiMo) we study two types of personalization, comparing their effect with the noise level. Passive personalization (user history in the prompt) on LaMP gives a robust average gain (16–50%), but the share of “harmed” users does not exceed the noise level. For active personalization we propose reranking results by user ratings, simulated by LLM personas: personas shift ratings beyond the noise, but change the top real answer no more often than noise (DeepSeek) or only slightly more often (MiMo, +12 pp).
Keywords: large language models, personalization, reranking, LLM personas.
Download PDFThe report examines a comparative analysis of neural networks with different complex activation functions (CReLU, zReLU, ModReLU, Cardioid, and ModTanh). The analysis results showed that the ModTanh function provides the highest training efficiency.
Keywords: neural networks, activation functions, learning rate, model stability.
Download PDFWe study linear dimensionality reduction and quantization for approximate nearest neighbor search (ANNS) when the distributions of indexed vectors and queries differ. We show that discarding components that are small for indexed vectors may cause significant dot-product errors for out-of-distribution (OOD) queries. We propose a sequential decomposition into orthogonal subspaces to account for components important to OOD queries. On the 200-dimensional Text-to-Image dataset, the proposed approach reduces vector storage by a factor of 1.67 and accelerates search by a factor of 1.14 compared with PCA-based quantization.
Keywords: approximate nearest neighbor search, dimensionality reduction, out-of-distribution queries, linear transformations, quantization.
Download PDFWe study temporal localization of two actions in untrimmed sumo videos, comparing YOLO-Pose, RGB-I3D, and VideoMAE features for ActionFormer. Joint fine-tuning of VideoMAE-H and temporal adapters yields the best result: 83.57% mAP at tIoU 0.50 and 57.43% averaged over five thresholds.
Keywords: temporal action detection, sumo, computer vision, ActionFormer, VideoMAE, AdaTAD.
Download PDFTwo LLM-assisted C# migrations are studied: EnergyCalc from Xamarin.Forms to Uno Platform and HouseBuilder from Xamarin.Forms and WPF/OpenTK to Godot. The assessment separates core reuse, host creation, LLM participation, and scenario parity. EnergyCalc retained 40–55% of code and passed 260 of 300 scenarios in 200 person-hours. HouseBuilder reused 70–80% of product code and passed 450 of 600 scenarios in 330 person-hours.
Keywords: large language models, AI agents, software migration, C\#, Uno Platform, Godot.
Download PDFWe compared the capabilities of two large language models (DeepSeek and GPT-6) in the automated construction of directed graphs representing "difficult life situations" (DLS) based on their textual descriptions. Using 10 such descriptions, we generated 20 graphs and evaluated them according to both formal and content-related criteria, employing an "LLM-as-a-Judge" approach. While both models demonstrated a high degree of formal correctness, this did not guarantee the adequacy of their content.
Keywords: difficult life situation, large language model, graph modeling.
Download PDFAn interpretable method for categorical data imputation based on association rules is proposed. Experiments on Car Evaluation, Nursery, and Mushrooms benchmarks demonstrate its competitiveness with MissForest and superiority over mode imputation.
Keywords: data quality, missing values, data imputation, association rules, categorical features, Apriori.
Download PDFWe present ComNetX, an algorithm that adapts static community detection methods to dynamic graphs through local updates of a partition hierarchy. It supports topology-based and neural methods, including those using node features. We establish preservation of hierarchy nesting, conditions for modularity preservation under graph contraction, and time and memory bounds. On five graphs, speedups over repeated Leiden runs range from 6.60 to 98.72, with absolute final modularity differences of 0.0035–0.0213.
Keywords: dynamic graph, community detection, partition hierarchy, modularity.
Download PDFAn encoding method based on oriented triangle areas is proposed for deciding affine equivalence of closed planar polygonal chains. The method arranges normalized oriented-area differences generated by the vertices of the chain and their neighbors into a matrix. For two closed polygonal chains with the same number of vertices, with the vertices of each chain not all collinear, equivalence of their codes is proved equivalent to the existence of an invertible affine transformation matching their vertices up to a cyclic shift or a reversal of order.
Keywords: closed polygonal chains, oriented area, affine invariants, completeness.
Download PDFSeven time-series classifiers are compared on eight UCR datasets: ROCKET, MiniRocket and BOSS form the top group by mean rank (1.63, 1.81 and 2.56) with no significant differences within it by the Nemenyi test, and the total times of MiniRocket and BOSS in the CPU runs are 2.4 min and 11.6 h, respectively. Five models forecasting BTC and ETH prices had higher RMSE than Naive in all eight asset–horizon combinations.
Keywords: time series, classification, symbolic representation, BOSS, ROCKET, MiniRocket, computational cost, forecasting.
Download PDFWe consider lexical semantic change detection based on clustering word usages by the matrix of their semantic proximities. Large language models estimate proximity more accurately than compact specialized models but are orders of magnitude more expensive, so the proposed QUEST-cluster method uses them selectively, within Bayesian estimation of the clustering threshold. A system based on QUEST-cluster raises the average Spearman correlation with the gold ranking to 0.85, compared with 0.755 for a system based on a compact model.
Keywords: lexical semantic change, word sense induction, large language models, Bayesian threshold estimation.
Download PDFWe propose a two-level grid-tagging method for identifying text spans and grouping them into annotation elements. The method supports overlapping and nested spans. We provide an analysis of results obtained with four configurations comprising three different text representation models and a large language model. Performance is evaluated using metrics computed by annotation matching.
Keywords: two-level text annotation, pairwise classification.
Download PDFThe effect of financial time series representation on forecast quality is investigated. FuzzyLSTM is considered with an additional feature computed from price history, parametric sigmoid functions in the LSTM gates, and blending of the network forecast with the latest observed price. Sufficient conditions for the ISS stability of the recurrent component are formulated, and assumptions ensuring boundedness of the final forecast are specified. On minute, hourly, and daily BTC–USD data, the ratio of the standard LSTM RMSE to the FuzzyLSTM RMSE was \(1.24\), \(1.45\), and \(1.54\), respectively. Adding the selected technical indicators did not provide a systematic improvement in FuzzyLSTM, while, in a separate comparison, symbolic Transformer input showed no consistent advantage over numerical input.
Keywords: financial time series, forecasting, FuzzyLSTM, technical indicators, symbolic representation.
Download PDFIn DeepFusionMOT, Stage-4 updates counters but not the stored 2D memory, which is reused when projection is unavailable or LiDAR misses occur. P3DR projects the predicted 3D box as evidence; C2MC writes the accepted 2D state for later frames. On KITTI Tracking Val11, relative to DeepFusionMOT, HOTA rises from 79.876% to 80.385%, sAMOTA from 93.01% to 95.42%, and FP falls from 294 to 110. In a same-source control (RRC disabled; native or projected boxes from one 3D detector), sequence outputs match the baseline.
Keywords: 3D multi-object tracking, camera--LiDAR fusion, cross-view memory, online association.
Download PDFWe transfer hard labels from quantized RGB segmentation to LiDAR, aggregate them by geometric clusters, and refine them with an MRF using distance, normals, and curvature. RELLIS-3D mIoU on valid projections is 48.9%.
Keywords: semantic segmentation, autonomous rover, LiDAR, RGB image, U-Net, EfficientNet, quantization, DBSCAN, Markov random field, alpha-expansion.
Download PDFManual cattle counting is labour-intensive, while automation is complicated by occlusions and repeated appearances. A system combining YOLO26L, ByteTrack and ResNet50 is presented. Image selection and track merging are used for single-camera counting, and geometric verification is developed for cross-camera matching. Experiments on MmCows yielded mAP@0.5 \(\approx 0.994\) and Rank-1 Accuracy of 90.6%.
Keywords: computer vision, individual animal identification, multi-camera monitoring, animal counting.
Download PDFWe audit and refine a PPO-based virtual rover navigation prototype. The original observation lacked obstacle distances; the revised 52-dimensional vector contains 36 LiDAR features. Three independently trained policies were evaluated over 500 episodes each. The mean success rate was 46.8%, with a between-run standard deviation of 6.8 percentage points. The results apply to one simulated scene only.
Keywords: reinforcement learning, navigation, virtual rover, PPO, LiDAR.
Download PDFEight language models classified emotions in minimal Russian sentences. Alice AI LLM and Qwen 3.5-397B-A17B tied in rounded accuracy (0.751). Alice AI LLM had the highest macro-F1 (0.742). Replacing a period with a question mark in affirmative past-tense sentences changed labels more often for each model than for annotators.
Keywords: language models, emotion classification, negation, punctuation, Russian.
Download PDFHVAC equipment manufacturers transfer air handling unit data from building designs into selection software by hand. We describe a scheme for extracting this data automatically: a large language model reads the sheet image into a JSON schema and classifies sheets not recognised from text; the remaining sheet selection, parsing of unit designations by formal grammars, source merging and all decisions are done by verifiable code. Recall of system-table and specification sheets is 100 % on the training corpus and 94–99 % on a held-out sample; a recheck of 560 extracted fields found no number-reading errors.
Keywords: large language models, information extraction, formal grammars, design documentation, air handling units.
Download PDFFor a single spiking ResNet-18 trained on image classification, we track, in all 17 layers of spiking neurons during training, the dimensionality of representations and the class information carried by individual neurons. In deep layers the dimensionality first drops and only then grows, and a linear measure does not detect this compression. Per-neuron class information rises in early and middle layers during the first epochs, then declines there and rises in deep layers; the epoch of its maximum does not decrease with layer depth.
Keywords: spiking neural networks, intrinsic dimension, mutual information, population coding.
Download PDFThe paper considers representing natural-language traffic rules as attributed directed graphs \(G=(V,E,P)\). A C program was developed to construct a graph for a vehicle leaving an adjacent area; its vertices, edges, and semantic attributes form a reference example. The program and the example graph were then used as a reference for analyzing other rules with ChatGPT 5. The results were manually checked against the rules. A total of 51 distinct traffic situations were identified according to specified selection criteria.
Keywords: traffic rules, graph model, language models, knowledge formalization, road traffic situations.
Download PDFA hybrid transaction-monitoring approach combines logistic regression with expert rules. On a sample of 300,000 PaySim transactions, test recall increased from 75.32% to 96.10%, while precision decreased from 84.06% to 19.73%; the alert rate was about 0.63%. Features, rules, and threshold selection are specified. Future work introduces behavioral features and trainable score aggregation.
Keywords: machine learning, fraud detection, payment systems, expert rules, behavioral analysis, risk assessment.
Download PDFModel selection for practical deployment is governed by the trade-off between functional accuracy and cost of use. A cost-of-use estimation procedure is tailored to models for structurally complex text annotation. The aggregate estimate accounts for the full automatic annotation pipeline and returns the cost of annotating 1,000 texts. Models of differing architecture and size were compared across 21 datasets.
Keywords: structured text annotation; span detection and linking; nested spans; inference costs; model selection; operational costs.
Download PDFA hybrid system for non-Hodgkin lymphoma differential diagnosis is proposed, combining statistical and neural models with Prolog rules and Mamdani fuzzy inference. The novelty lies in a dual-loop consensus mechanism that uses semantic constraints to filter clinically invalid hypotheses. The macro-averaged \(F_1\) score on the test set was \(0.75\).
Keywords: non-Hodgkin lymphoma, hybrid intelligent systems, fuzzy logic, logic programming, machine learning, explainable AI.
Download PDFThe paper tests how the choice of the adjacency matrix affects the macro-F1 of a graph neural network. On UCI HAR, over \(20\) paired runs, a channel-metadata graph is compared with its equal-degree complement, a graph without edges, and hybrid fine-tuning. The mean difference from the complement is \(1.5\)–\(1.9\) percentage points and is significant after the Holm correction at every training fraction. The difference from the graph without edges is significant at \(5\) and \(10\%\). Hybrid fine-tuning is significantly worse than the fixed matrix at \(10\) and \(100\%\). The spread of the difference within one subsample exceeds the spread between subsamples. A sweep of initial graphs on the full training set shows no dip.
Keywords: graph neural networks, adjacency matrix, multivariate time series, generalization ability, activity recognition.
Download PDFDetecting and correcting span boundary and label errors is essential in social sciences and humanities but costly to automate. To reduce costs, we propose a method using instance hardness assessments, efficient vectorization, and one-time document encoding. Comparison with LLMs on three correction sources shows comparable quality on several auditing tasks with substantially shorter runtimes.
Keywords: annotation auditing; text spans; instance hardness; annotation correction; LLM-as-a-judge.
Download PDFAnchorBank corrects contrastive adaptation using relations to frozen anchors. We describe its controlled subspace and local projection properties and compare it with InfoNCE on four collections.
Keywords: information retrieval, contrastive learning, KL divergence, gradient projection.
Download PDFWe consider an inverse problem of choosing a fuzzy conjunctive aggregation operator in a homogeneous tree-based risk assessment model. The aggregator is required to be stable with respect to background changes and sufficiently sensitive to a local incident. In the class of strict Archimedean \(t\)-norms, the root value of a homogeneous tree and its local sensitivities are expressed through an additive generator and its profile. This leads to an optimization problem for choosing the aggregator; in the power class of profiles, the problem is reduced to one-dimensional optimization.
Keywords: fuzzy aggregation, $t$-norm, additive generator, homogeneous model, risk assessment, inverse problem, sensitivity.
Download PDFA versioned architecture of an intelligent information system is proposed for group decisions on placing analytical data warehouse objects under personal data governance. Model predictions, expert answers and final labels are tied to the object version; confirmation requires expert agreement and the presence of all mandatory roles. On a held-out set of 9,379 columns, a local model reached \(F_1=0.883\).
Keywords: data warehouse, data lineage, personal data, group decision.
Download PDFThis paper explores a synergistic (hybrid) approach to solving the Graph Coloring Problem (GCP), based on the integration of the classic heuristic DSatur algorithm and the physics-informed graph neural network. The proposed hybrid approach aims to improve coloring accuracy and stability compared to using neural network models in isolation. The results of a comparative analysis of the proposed method against classical algorithms on graphs of various sizes are presented.
Keywords: graph coloring problem, GCP, graph neural networks, GNN, PI-GNN, DSatur.
Download PDFLean 4 tactics often take candidate lemmas from vector search by the inner product of goal and lemma embeddings. We show that no fixed embedding dimension suffices to rank lemmas exactly: already on an explicit family of rewrite lemmas whose patterns consist of \(m\) blocks of \(m^2\) differences, the required dimension grows exponentially in \(m\).
Keywords: premise selection, embeddings, automated theorem proving, Lean~4, lower bounds.
Download PDFIn this paper, we define a specific criterion system for classes of functions closed under the operation of superposition. This system is used to test whether functions \(f(x,y)\) realizable by neural schemes \(\sum\) consisting of 5 elements from the set \(B=\{x+y, \frac {1}{2}x, -x, \theta (x)\}\) are universal within the class of piecewise-parallel functions .
Keywords: minimal universal functions, class of piecewise parallel functions, neural schemes.
Download PDFThis article studies generative LLM-based syntactic parsing. The impact of bracketed syntactic representation formats (GRCT and LCT) on parsing quality is investigated. Research has established that an LLM can be successfully fine-tuned for analysis on each of the formats. For both formats, according to the LAS_LLM metric, the fine-tuned Qwen3-4B model exceeds the classic neural network parsers Stanza and DeepPavlov. Moreover, the fine-tuned LLM is resistant to domain shift.
Keywords: parsing, fine-tuning, large language models, dependency tree, bracketed format.
Download PDFAn algorithm is proposed that takes as input two polynomials; in addition to the primary variable (“unknown”) \(x\), these polynomials also depend on variables \(a_1, \dots , a_s\), which act as parameters that can take on various values. The study examines the greatest common divisor of the original polynomials as a function of the values assigned to these parameters.
Keywords: polynomials with parameters, subresultant chain, greatest common divisor, computer algebra.
Download PDFThe paper addresses the lack of unified standards for the logical presentation of mathematical proofs. Drawing an analogy between mathematical text and program code, and building on the principles of Clean Code and formal logic, we propose a unified standard for writing definitions, theorem statements, and proofs. We discuss the possibility of representing a proof text constructed according to the proposed rules as a formal object — a derivation in the grammar of steps — and outline how such an object could be synthesized by a symbolic solver and checked by a verifier. The description of the standard’s rules can serve as a structured specification (a prompt) for training a specialized LLM that generates proof sketches under the control of symbolic verification.
Keywords: symbolic computation, mathematical logic, proof structure, standardization, proof synthesis, verification, large language models, formalization.
Download PDFFor first-order differential operator matrices with a nonsingular leading matrix, a deterministic algorithm dWPF is proposed that constructs a weak Popov form of minimum total coefficient degree among the results of WPF. The optimum for one column is given by a minimum spanning tree; columns are coordinated by branch and bound. The results are compared with WPF implementations in SageMath and AbstractAlgebra.jl.
Keywords: symbolic computation, operator matrices, weak Popov form, coefficient growth.
Download PDFThe triplication method for constructing strong starters in cyclic groups \(\mathbb {Z}_{3m}\) was earlier implemented with SAT/SMT solver z3 using constraints for unknowns with values in \(\{0,1,2\}\). Emulation by Boolean constraints boosted performance by several orders of magnitude.
Keywords: strong starter, triplication, SAT solver z3.
Download PDFThe study addresses the modeling and analysis of resonance characteristics in mechanical systems with moving boundaries. The focus is on the use of artificial intelligence (AI), specifically the hybrid neural network ResonanceNet (CNN + LSTM), to predict resonant frequencies, amplitudes, and prevent resonance. The computational method for solving integro-differential equations developed by the authors is applied. The problem is formulated as a boundary value problem for a partial differential equation with a moving boundary and reduced to a system of stochastic differential equations. The neural network is used offline for fast estimation; if confidence is low, a classical numerical module is activated. Comparative analysis showed that the hybrid approach reduces prediction error to 2.1% and prevents resonance in 96% of cases, which is 34% more accurate and 5 times faster than classical methods.
Keywords: viscoelasticity, moving boundaries, stochastic vibrations, resonance, artificial intelligence, neural networks, Monte Carlo method, TB-ANALYSIS 7 package.
Download PDFWe developed a brain–computer interface (BCI) that uses a class of markers new to BCIs: indices of intracortical excitability. These indices are derived from transcranial evoked potentials (TEPs) elicited by subthreshold transcranial magnetic stimulation of the motor cortex. In 18 participants, online accuracy for distinguishing motor imagery from rest using TEPs reached \(77.4\pm 12.5\) % and did not correlate with mu-rhythm BCI accuracy. The discriminative feature was attenuation of the N100 component during motor imagery. Combining TEP and mu-rhythm classifiers in a hybrid BCI yielded an accuracy of \(83.6\pm 8.6\) %.
Keywords: brain–computer interface, transcranial evoked potentials, intracortical excitability, motor imagery, mu rhythm.
Download PDFA passive finger movement stimulator was developed and tested in 24 healthy participants. In specific stimulation conditions, active attention was associated with greater P300 amplitude, higher evoked potential classification accuracy, and stronger mu-rhythm desynchronization than distracted attention. Further development of the device and EEG-based feedback for sensorimotor training and neurorehabilitation is discussed.
Keywords: passive movements, somatosensory evoked potentials, active attention, P300, mu rhythm, brain–computer interface, neurorehabilitation.
Download PDFThis paper presents a mathematical model of glucose homeostasis regulation describing the interaction between pancreatic cells, the liver, and peripheral tissues. Based on a discrete dynamical system, a system of ordinary differential equations is constructed, for which the global existence, uniqueness, and analyticity of solutions are established. The existence and uniqueness of an equilibrium point of the system are proved.
Keywords: endocrine system, hybrid automata, pancreas, glucose, insulin, glucagon, glycogen.
Download PDFThis article proposes using a horizontal spring-pendulum model with a variable stiffness coefficient to simulate the regulation of an organism’s physiological parameters via negative feedback. The system is assumed to be subject to an external force, which triggers a counteracting response from the spring that reduces the amplitude of the system’s deviation. A method for controlling the stiffness coefficient is presented, which allows for the regulation of the pendulum’s coordinates, keeping them within permissible limits.
Keywords: regulation of physiological parameters, negative feedback, horizontal spring spring-pendulum.
Download PDFMelting point prediction is crucial for chemistry and pharmaceuticals, but current methods face accuracy and efficiency challenges. We curated a harmonized dataset of 117,752 organic crystals. We compared machine learning models using molecular embeddings and graph neural networks. A hybrid TabM model with concatenated ChemBERTa and MoLFormer embeddings achieves MAE of 26.75 K in 27 minutes of training, offering an optimal balance of accuracy and computational cost compared to Chemprop (MAE 26.44 K, 45 minutes).
Keywords: melting point, organic crystals, machine learning, graph neural networks, ChemBERTa, MoLFormer, TabM, Chemprop.
Download PDFWe studied the effects of open-field (OF) exploration, seven-day enriched-environment (EE) housing, and (R)-CPP on weak non-associative place memory in mice. No significant memory enhancement was detected after OF; the mobility index increased following 1-min training. EE significantly reduced the index after 30-s, but not 1-min, training. The (R)-CPP group differed from non-injected, but not saline-injected, controls.
Keywords: weak memory, spatial novelty, behavioral tagging, enriched environment, non-associative place memory.
Download PDFUsing graph theory, we compared mouse brain functional networks built from Spearman correlations of c-Fos expression in 17 regions after conditioning to a compound visual-auditory stimulus and to each of its components. After sound and compound conditioning, the parietal associative cortex joined the hippocampus in a single module, and the compound network was weakly similar to that of its visual component.
Keywords: functional connectivity, graph theory, c-Fos, compound conditioned stimulus, parietal associative cortex, hippocampus.
Download PDFWe studied how daily experience duration (10, 30 or 60 min for four days in a novel environment) affects the stability of the spatial code in mouse hippocampal area CA1. With 10-min episodes, place-cell recurrence across days was half as high and the place-cell fraction did not grow. Comparison with the recurrence expected for independent sessions showed that conclusions about code stability depend on normalization and the null model; the same measures apply to units of artificial neural networks in continual learning.
Keywords: hippocampus, place cells, representational stability, null model.
Download PDFWe propose a method for classifying open and closed eyes based on the construction of convex linear hulls of finite point sets over the field \(\mathbb Z_3=\{0,1,2\}\). A hyperplane is called a face if it contains at least \(n\) points of the original set and the corresponding linear form takes the same nonzero value \(\lambda _a\in \{1,2\}\) at all remaining points of the set. Membership of a point in the convex linear hull is determined by the conditions \(L_a(x)\in \{0,\lambda _a\}\) for all its faces.
Keywords: convex linear hull, finite fields, $E_3^n$ space, linear forms, pattern recognition, image classification.
Download PDFUsing symbolic regression, the convergence pressure has been incorporated into Wilson’s correlation for phase equilibrium constants (K-values). The resulting correlations accurately represent the phase behavior of an oil-and-gas reservoir with a given composition across the entire pressure range.
Keywords: symbolic regression, Wilson formula, convergence pressure, distribution coefficients.
Download PDFTraditional control systems for complex technological processes require extensive expert tuning and often fail to adapt to the drift in real equipment characteristics. This study aims to develop purely data-driven predictive control algorithms, combining the end‑to‑end differentiability of the surrogate model with classical mathematical programming methods. The performance and transferability of the developed methods have been validated using data from the industrial Tennessee Eastman process simulator and a real oil refinery complex. The results demonstrate effective process control and reliable maintenance of technological parameters within specified operational constraints.
Keywords: structured state-space models, reinforcement learning, digital twin.
Download PDFAcross three experiments, 72 EEG recordings from 60 unique participants and all 3,826 tactile imagery (TI) and rest epochs were analyzed. Individual logistic regression models used regularized covariance features. In fivefold epoch-wise cross-validation, mean balanced accuracy ranged from 76.03% to 78.42% across datasets; the median of 72 estimates was 80.00%.
Keywords: electroencephalography, tactile imagery, classification, brain--computer interface.
Download PDFNeural population activity is described by low-dimensional codes, and the method decides what can be read from the code. On synthetic calcium imaging recordings we compared an autoencoder with linear and manifold learning methods. When cells encode only the animal’s position, UMAP conveys it best; when the activity carries other signals too, position is read most accurately from the autoencoder’s code by a linear decoder at 8 dimensions, and when cells encoding a signal unrelated to behavior outnumber place cells, position is predicted more accurately from its 5-dimensional code than from the codes of the other methods.
Keywords: autoencoder, dimensionality reduction, place cells, synthetic data.
Download PDFThe problem of accelerating the convergence of numerical sequences and series under complex behavior of the remainder is considered: sign changes, multiple asymptotic components, gradual loss of computational accuracy. A comparative analysis of classical and new nonlinear transformations (Levin’s transformations and their \(t\), \(u\), \(v\) variants, Levin–Sidi transformations, Sidi’s \(d^{(m)}\)-transformation, Homeier’s \(pJ\)-transformation, Lubkin’s \(W\)-transformation, Wynn’s \(\varepsilon\)-algorithm, Brezinski’s \(\theta\)-algorithm, etc.) is carried out on a wide set of test sequences with exactly known limits. Computations are performed with adjustable precision up to 100 significant decimal digits. It is shown that no single transformation is universally the best; based on the experimental results, two complementary sets are identified: a fast set comprising Drummond’s transformation, Sidi’s \(d^{(m)}\)-transformation, and Wynn’s \(\varepsilon\)-algorithm, and a high-accuracy set comprising Brezinski’s \(\theta\)-algorithm, Sidi’s \(d^{(m)}\)-transformation, and Wynn’s \(\varepsilon\)-algorithm. A special class of sequences with attracting and repelling points is identified among the test sequences, requiring further investigation.
Keywords: numerical series, convergence acceleration, Levin's transformation, Levin--Sidi transformation, Homeier's $pJ$-transformation, Brezinski's $\theta $-algorithm, finite-precision arithmetic, interval arithmetic.
Download PDFWe present a discrete stochastic model of alternative splicing of the CD44 variable region. The model describes the transport of ESRP1 and ESRP2 proteins between the nucleus and cytoplasm, their binding to pre-mRNA, and the sequential formation of splice junctions with exons v8, v9 and v10. Trajectories are simulated with the Gillespie algorithm. Parameters were fitted to junction fractions in H358 cells under control conditions and after combined ESRP1/2 knockdown; no independent validation has been performed. Fitting yielded a parameter set consistent with the data, in which the nuclear fraction of ESRP1 increases after knockdown, which may partially offset the decrease in its total amount.
Keywords: alternative splicing, CD44, RNA-binding proteins, stochastic model, Gillespie algorithm.
Download PDFWe integrated a trainable rearing detector into the Sphynx behavioral-analysis platform, combining kinematic features with spatial information from arena annotation. On a public dataset, against the consensus of two experts, the detector achieved frame-level F1 = 0.84, compared with 0.54 for the original rule-based detector and 0.85 for the third expert. Spatial features contributed substantially to classification performance.
Keywords: behavioral analysis, pose estimation, behavioral annotation, gradient boosting.
Download PDFWe present SES-GNN, a multiscale graph neural network on the solvent-excluded surface (SES) for protein binding site prediction. Instead of a uniform graph with fixed edge radius, a three-scale representation (0–2, 2–4, 4–8 Å) with Farthest Point Sampling centroid selection is built. The architecture uses 12 TransformerConv layers. The multiscale graph gives a +0.015–0.017 ROC-AUC gain over the uniform one. The method achieves ROC-AUC = 0.838 on our own PPI dataset. Ablation shows that vertex features carry the main information but realize it only through the graph structure: a pointwise classifier without message passing falls substantially below the full model, while topology alone is useless.
Keywords: protein binding site prediction, SES surface, graph neural networks.
Download PDFWe identified goal-directed approaches of mice toward a target bowl during free exploration of five arenas and assessed the association of hippocampal CA1 activity with this behavior using the INTENSE method. The groups differed in behavior: mice familiar with the bowl approached it more often than controls in the first sessions. CA1 neurons selective for such approaches were rare: their fraction was at most 0.1 %, whereas up to 4 % were selective for rearings. Neural correlates of the act may lie in its continuous spatial and kinematic components.
Keywords: goal-directed behavior, hippocampus, CA1, calcium imaging, miniscope, mutual information, INTENSE.
Download PDFThe relevance of this study stems from the need to develop objective methods for identifying cognitive load. The aim of the work is to validate a quantitative indicator for assessing cognitive load based on oculomotor data — the average difference between the actual and optimal eye movement trajectories (RFOT). The study involved 95 participants. The stimulus material included three Gorbov–Schulte tables. Eye movements were recorded using an EyeLink 1000 Plus. It was shown that in the experimental group the deviation of the actual oculomotor trajectory from the optimal one was significantly smaller than in the control group. An increase in intrinsic cognitive load led to an increase in RFOT. The obtained results indicate that the proposed indicator is sensitive to changes in cognitive load and allow the increase in RFOT to be interpreted as a marker of cognitive resource depletion.
Keywords: visual search, cognitive load, visual attention, eye tracking.
Download PDFWe compared mouse behavior and the selectivity of hippocampal CA1 neurons to behavioral variables one day after single, massed and spaced exploration of an open field with equal total exposure. No significant differences between schedules were found in the structure of behavior or in the proportions of neurons selective to behavioral variables.
Keywords: spaced learning, hippocampus, neuronal selectivity.
Download PDFWe propose a method for analyzing and controlling exercise technique from a single camera video stream: a pose estimation model converts frames into body keypoint coordinates, a feature space maps them into biomechanically interpretable characteristics, and fuzzy inference over an expert trainer’s rule base generates voice corrections in real time.
Keywords: human pose estimation, fuzzy logic, cyber-physical systems, movement biomechanics.
Download PDF
RU