CyberSec Research

The industrialization of catalytic processes hinges on the availability of reliable kinetic models for design, optimization, and control. Traditional mechanistic models demand extensive domain expertise, while many data-driven approaches often lack interpretability and fail to enforce physical consistency. To overcome these limitations, we propose the Physics-Informed Automated Discovery of Kinetics (PI-ADoK) framework. By integrating physical constraints directly into a symbolic regression approach, PI-ADoK narrows the search space and substantially reduces the number of experiments required for model convergence. Additionally, the framework incorporates a robust uncertainty quantification strategy via the Metropolis-Hastings algorithm, which propagates parameter uncertainty to yield credible prediction intervals. Benchmarking our method against conventional approaches across several catalytic case studies demonstrates that PI-ADoK not only enhances model fidelity but also lowers the experimental burden, highlighting its potential for efficient and reliable kinetic model discovery in chemical reaction engineering.

The integrability problem of rational first-order ODEs $y^{\prime}=\frac{M(x,y)}{N(x,y)}$, where $M,N \in \mathbb{R}[x,y]$ is a long-term research focus in the area of dynamical systems, physics, etc. Although the computer algebra system such as Mathematica, Maple has developed standard algorithms to tackle its first integral expressed by Liouvillian or special function, this problem is quite difficult and the general method requires specifying a tight degree bound for the Darboux polynomial. Computing the bounded degree first integral, in general, is very expensive for a computer algebra system\cite{duarte2021efficient}\cite{cheze2020symbolic} and becomes impractical for ODE of large size. In \cite{huang2025algorithm}, we have proposed an algorithm to find the inverse of a local rational transformation $y \to \frac{A(x,y)}{B(x,y)}$ that transforms a rational ODE to a simpler and more tractable structure $y^{\prime}=\sum_{i=0}^nf_i(x)y^i$, whose integrability under linear transformation $\left\{x \to F(x),y \to P(x)y+Q(x)\right\}$ can be detected by Maple efficiently \cite{CHEBTERRAB2000204}\cite{cheb2000first}. In that paper we have also mentioned when $M(x,y),N(x,y)$ of the reducible structure are not coprime, canceling the common factors in $y$ will alter the structure which makes that algorithm fail. In this paper, we consider this issue. We conclude that with the exact tight degree bound for the polynomial $A(x,y)$ given, we have an efficient algorithm to compute such transformation and the reduced ODE for "quite a lot of" cases where $M,N$ are not coprime. We have also implemented this algorithm in Maple and the code is available in researchgate.

Continuous attractor networks (CANs) are widely used to model how the brain temporarily retains continuous behavioural variables via persistent recurrent activity, such as an animal's position in an environment. However, this memory mechanism is very sensitive to even small imperfections, such as noise or heterogeneity, which are both common in biological systems. Previous work has shown that discretising the continuum into a finite set of discrete attractor states provides robustness to these imperfections, but necessarily reduces the resolution of the represented variable, creating a dilemma between stability and resolution. We show that this stability-resolution dilemma is most severe for CANs using unimodal bump-like codes, as in traditional models. To overcome this, we investigate sparse binary distributed codes based on random feature embeddings, in which neurons have spatially-periodic receptive fields. We demonstrate theoretically and with simulations that such grid-cell-like codes enable CANs to achieve both high stability and high resolution simultaneously. The model extends to embedding arbitrary nonlinear manifolds into a CAN, such as spheres or tori, and generalises linear path integration to integration along freely-programmable on-manifold vector fields. Together, this work provides a theory of how the brain could robustly represent continuous variables with high resolution and perform flexible computations over task-relevant manifolds.

In this paper, we advance local search for Satisfiability Modulo the Theory of Nonlinear Real Arithmetic (SMT-NRA for short). First, we introduce a two-dimensional cell-jump move, called \emph{$2d$-cell-jump}, generalizing the key operation, cell-jump, of the local search method for SMT-NRA. Then, we propose an extended local search framework, named \emph{$2d$-LS} (following the local search framework, LS, for SMT-NRA), integrating the model constructing satisfiability calculus (MCSAT) framework to improve search efficiency. To further improve the efficiency of MCSAT, we implement a recently proposed technique called \emph{sample-cell projection operator} for MCSAT, which is well suited for CDCL-style search in the real domain and helps guide the search away from conflicting states. Finally, we design a hybrid framework for SMT-NRA combining MCSAT, $2d$-LS and OpenCAD, to improve search efficiency through information exchange. The experimental results demonstrate improvements in local search performance, highlighting the effectiveness of the proposed methods.

We give a quantifier elimination procedure for one-parametric Presburger arithmetic, the extension of Presburger arithmetic with the function $x \mapsto t \cdot x$, where $t$ is a fixed free variable ranging over the integers. This resolves an open problem proposed in [Bogart et al., Discrete Analysis, 2017]. As conjectured in [Goodrick, Arch. Math. Logic, 2018], quantifier elimination is obtained for the extended structure featuring all integer division functions $x \mapsto \lfloor{\frac{x}{f(t)}}\rfloor$, one for each integer polynomial $f$. Our algorithm works by iteratively eliminating blocks of existential quantifiers. The elimination of a block builds on two sub-procedures, both running in non-deterministic polynomial time. The first one is an adaptation of a recently developed and efficient quantifier elimination procedure for Presburger arithmetic, modified to handle formulae with coefficients over the ring $\mathbb{Z}[t]$ of univariate polynomials. The second is reminiscent of the so-called "base $t$ division method" used by Bogart et al. As a result, we deduce that the satisfiability problem for the existential fragment of one-parametric Presburger arithmetic (which encompasses a broad class of non-linear integer programs) is in NP, and that the smallest solution to a satisfiable formula in this fragment is of polynomial bit size.

Motivated by a flurry of recent work on efficient tensor decomposition algorithms, we show that the celebrated moment matrix extension algorithm of Brachat, Comon, Mourrain, and Tsigaridas for symmetric tensor canonical polyadic (CP) decomposition can be made efficient under the right conditions. We first show that the crucial property determining the complexity of the algorithm is the regularity of a target decomposition. This allows us to reduce the complexity of the vanilla algorithm, while also unifying results from previous works. We then show that for tensors in $S^d\mathbb{C}^{n+1}$ with $d$ even, low enough regularity can reduce finding a symmetric tensor decomposition to solving a system of linear equations. For order-$4$ tensors we prove that generic tensors of rank up to $r=2n+1$ can be decomposed efficiently via moment matrix extension, exceeding the rank threshold allowed by simultaneous diagonalization. We then formulate a conjecture that states for generic order-$4$ tensors of rank $r=O(n^2)$ the induced linear system is sufficient for efficient tensor decomposition, matching the asymptotics of existing algorithms and in fact improving the leading coefficient. Towards this conjecture we give computer assisted proofs that the statement holds for $n=2, \dots, 17$. Next we demonstrate that classes of nonidentifiable tensors can be decomposed efficiently via the moment matrix extension algorithm, bypassing the usual need for uniqueness of decomposition. Of particular interest is the class of monomials, for which the extension algorithm is not only efficient but also improves on existing theory by explicitly parameterizing the space of decompositions. Code for implementations of the efficient algorithm for generic tensors and monomials are provided, along with several numerical examples.

We explore new approaches for finding matrix multiplication algorithms in the commutative setting by adapting the flip graph technique: a method previously shown to be effective for discovering fast algorithms in the non-commutative case. While an earlier attempt to apply flip graphs to commutative algorithms saw limited success, we overcome both theoretical and practical obstacles using two strategies: one inspired by Marakov's algorithm to multiply 3x3 matrices, in which we construct a commutative tensor and approximate its rank using the standard flip graph; and a second that introduces a fully commutative variant of the flip graph defined via a quotient tensor space. We also present a hybrid method that combines the strengths of both. Across all matrix sizes up to 5x5, these methods recover the best known bounds on the number of multiplications and allow for a comparison of their efficiency and efficacy. Although no new improvements are found, our results demonstrate strong potential for these techniques at larger scales.

We propose a strategy for the generation of fast and accurate versions of non-commutative recursive matrix multiplication algorithms. To generate these algorithms, we consider matrix and tensor norm bounds governing the stability and accuracy of numerical matrix multiplication. We start by a unification on known max-norm bounds on matrix multiplication stability and then extend them to further norms and more generally to recursive bilinear algorithms and the alternative basis matrix multiplication algorithms. Then our strategy has three phases. First, we reduce those bounds by minimizing a growth factor along the orbits of the associated matrix multiplication tensor decomposition. Second, we develop heuristics that minimize the number of operations required to realize a bilinear formula, while further improving its accuracy. Third, we perform an alternative basis sparsification that improves on the time complexity constant and mostly preserves the overall accuracy. For instance this strategy allows us to propose a non-commutative algorithm for multiplying 2x2-matrices using 7 coefficient products. This algorithm reaches simultaneously a better accuracy in practice compared to previously known such fast ___2x2x2:7___ Strassen-like algorithms and a time complexity bound with the best currently known leading term (obtained via alternative basis sparsification). We also present detailed results of our technique on other recursive matrix multiplication algorithms, such as Smirnov's ___3x3x6:40___ family of algorithms.

The Peterson hit problem in algebraic topology is to explicitly determine the dimension of the quotient space $Q\mathcal P_k = \mathbb F_2\otimes_{\mathcal A}\mathcal P_k$ in positive degrees, where $\mathcal{P}_k$ denotes the polynomial algebra in $k$ variables over the field $\mathbb{F}_2$, considered as an unstable module over the Steenrod algebra $\mathcal{A}$. Current approaches to this problem still rely heavily on manual computations, which are highly prone to errors due to the intricate nature of the underlying calculations. To date, no efficient algorithm implemented in any computer algebra system has been made publicly available to tackle this problem in a systematic manner. Motivated by the above, in this work, which is considered as Part I of our project, we first establish a criterion based entirely on linear algebra for determining whether a given homogeneous polynomial is "hit". Accordingly, we describe the dimensions of the hit spaces. This leads to a practical and reliable computational method for determining the dimension of $Q\mathcal{P}_k$ for arbitrary $k$ and any positive degrees, with the support of a computer algebra system. We then give a concrete implementation of the obtained results as novel algorithms in \textsc{SageMath}. As an application, our algorithm demonstrates that the manually computed result presented in the recent work of Sum and Tai [15] for the dimension of $Q\mathcal{P}_5$ in degree $2^{6}$ is not correct. Furthermore, our algorithm determines that $\dim(Q\mathcal{P}_5)_{2^{7}} = 1985,$ which falls within the range $1984 \leq \dim(Q\mathcal{P}_5)_{2^{7}} \leq 1990$ as estimated in [15].

Understanding and modeling nonlinear dynamical systems is a fundamental problem across scientific and engineering domains. While deep learning has demonstrated remarkable potential for learning complex system behavior, achieving models that are both highly accurate and physically interpretable remains a major challenge. To address this, we propose Structured Kolmogorov-Arnold Neural ODEs (SKANODEs), a novel framework that integrates structured state-space modeling with the Kolmogorov-Arnold Network (KAN). SKANODE first employs a fully trainable KAN as a universal function approximator within a structured Neural ODE framework to perform virtual sensing, recovering latent states that correspond to physically interpretable quantities such as positions and velocities. Once this structured latent representation is established, we exploit the symbolic regression capability of KAN to extract compact and interpretable expressions for the system's governing dynamics. The resulting symbolic expression is then substituted back into the Neural ODE framework and further calibrated through continued training to refine its coefficients, enhancing both the precision of the discovered equations and the predictive accuracy of system responses. Extensive experiments on both simulated and real-world systems demonstrate that SKANODE achieves superior performance while offering interpretable, physics-consistent models that uncover the underlying mechanisms of nonlinear dynamical systems.

The quest for non-commutative matrix multiplication algorithms in small dimensions has seen a lot of recent improvements recently. In particular, the number of scalar multiplications required to multiply two $4\times4$ matrices was first reduced in \cite{Fawzi:2022aa} from 49 (two recursion levels of Strassen's algorithm) to 47 but only in characteristic 2 or more recently to 48 in \cite{alphaevolve} but over complex numbers. We propose an algorithm in 48 multiplications with only rational coefficients, hence removing the complex number requirement. It was derived from the latter one, under the action of an isotropy which happen to project the algorithm on the field of rational numbers. We also produce a straight line program of this algorithm, reducing the leading constant in the complexity, as well as an alternative basis variant of it, leading to an algorithm running in $\frac{19}{16} n^{2+\frac{\log_2 3}{2}} +o\left(n^{2+\frac{log_2 3}{2}}\right)$ operations over any ring containing an inverse of 2.

We propose Polyra Swarms, a novel machine-learning approach that approximates shapes instead of functions. Our method enables general-purpose learning with very low bias. In particular, we show that depending on the task, Polyra Swarms can be preferable compared to neural networks, especially for tasks like anomaly detection. We further introduce an automated abstraction mechanism that simplifies the complexity of a Polyra Swarm significantly, enhancing both their generalization and transparency. Since Polyra Swarms operate on fundamentally different principles than neural networks, they open up new research directions with distinct strengths and limitations.

Optimal control in general, and flatness-based control in particular, of robotic arms necessitate to compute the first and second time derivatives of the joint torques/forces required to achieve a desired motion. In view of the required computational efficiency, recursive $O(n)$-algorithms were proposed to this end. Aiming at compact yet efficient formulations, a Lie group formulation was recently proposed, making use of body-fixed and hybrid representation of twists and wrenches. In this paper a formulation is introduced using the spatial representation. The second-order inverse dynamics algorithm is accompanied by a fourth-order forward and inverse kinematics algorithm. An advantage of all Lie group formulations is that they can be parameterized in terms of vectorial quantities that are readily available. The method is demonstrated for the 7 DOF Franka Emika Panda robot.

This paper introduces the Primender sequence, a novel integer sequence defined by a hybrid rule that combines classical primality with modular digit-based conditions. Specifically, a number n is included in the sequence if it is prime or ends with a prime number of unit digit or any length. In other words, numbers which are primes or have at least one prime suffix. The resulting sequence exhibits a deterministic yet non-trivial structure, blending number-theoretic properties with symbolic patterning. We propose the Primender sequence as a benchmark for evaluating the symbolic reasoning capabilities of Large Language Models (LLMs). The study is motivated by the need for interpretable, rule-based testbeds that can assess an LLM's ability to infer hidden rules, validate mathematical hypotheses, and generalize symbolic logic at scale. A key hypothesis explored is: Whenever a number in the Primender sequence is exactly one more than the largest prime less than or equal to it, the difference between it and the previous number in the sequence is also 1. We design a structured prompt and evaluation framework to test this hypothesis across multiple state-of-the-art LLMs, including ChatGPT, Copilot, DeepSeek, Gemini, Grok, and LLaMA. The models are tasked with identifying the underlying rule, validating the hypothesis, and generating the next 100,000 terms of the sequence. Comparative metrics such as rule inference accuracy, hypothesis evaluation, sequence validity, and symbolic explanation quality are used to assess model performance. This work contributes a novel mathematical construct and a reproducible methodology for benchmarking LLMs in symbolic reasoning, hypothesis testing, and scalable pattern generalization - bridging the domains of number theory, artificial intelligence, and software engineering.

The multistep solving strategy consists in a divide-and-conquer approach: when a multivariate polynomial system is computationally infeasible to solve directly, one variable is assigned over the elements of the base finite field, and the procedure is recursively applied to the resulting simplified systems. In a previous work by the same authors (among others), this approach proved effective in the algebraic cryptanalysis of the Trivium cipher. In this paper, we present a new implementation of the corresponding algorithm based on a Depth-First Search strategy, along with a novel complexity analysis leveraging tree structures. We further introduce the notion of an "oracle function" as a general predictive tool for deciding whether the evaluation of a new variable is necessary to simplify the current polynomial system. This notion allows us to unify all previously proposed variants of the multistep strategy, including the classical hybrid approach, by appropriately selecting the oracle function. Finally, we apply the multistep solving strategy to the cryptanalysis of the low-latency block cipher Aradi, recently introduced by the NSA. We present the first full round algebraic attack, raising concerns about the cipher's actual security with respect to its key length.

This paper studies the concept and the computation of approximately vanishing ideals of a finite set of data points. By data points, we mean that the points contain some uncertainty, which is a key motivation for the approximate treatment. A careful review of the existing border basis concept for an exact treatment motivates a new adaptation of the border basis concept for an approximate treatment. In the study of approximately vanishing polynomials, the normalization of polynomials plays a vital role. So far, the most common normalization in computational commutative algebra uses the coefficient norm of a polynomial. Inspired by recent developments in machine learning, the present paper proposes and studies the use of gradient-weighted normalization. The gradient-weighted semi-norm evaluates the gradient of a polynomial at the data points. This data-driven nature of gradient-weighted normalization produces, on the one hand, better stability against perturbation and, on the other hand, very significantly, invariance of border bases with respect to scaling the data points. Neither property is achieved with coefficient normalization. In particular, we present an example of the lack of scaling invariance with respect to coefficient normalization, which can cause an approximate border basis computation to fail. This is extremely relevant because scaling of the point set is often recommended for preprocessing the data. Further, we use an existing algorithm with coefficient normalization to show that it is easily adapted to gradient-weighted normalization. The analysis of the adapted algorithm only requires tiny changes, and the time complexity remains the same. Finally, we present numerical experiments on three affine varieties to demonstrate the superior stability of our data-driven normalization over coefficient normalization. We obtain robustness to perturbations and invariance to scaling.

We study an important special case of the differential elimination problem: given a polynomial dynamical system $\mathbf{x}' = \mathbf{g}(\mathbf{x})$ and a polynomial observation function $y = f(\mathbf{x})$, find the minimal differential equation satisfied by $y$. In our previous work, for the case $y = x_1$, we established a bound on the support of such a differential equation and shown that it can be turned into an algorithm via the evaluation-interpolation approach. The main contribution of the present paper is a generalization of the aforementioned result in two directions: to allow any polynomial function $y = f(\mathbf{x})$, not just a single coordinate, and to allow $\mathbf{g}$ and $f$ depend on unknown symbolic parameters. We conduct computation experiments to evaluate the accuracy of our new bound and show that the approach allows to perform elimination for some cases out of reach for the state of the art software.

A complete reduction on a difference field is a linear operator that enables one to decompose an element of the field as the sum of a summable part and a remainder such that the given element is summable if and only if the remainder is equal to zero. In this paper, we present a complete reduction in a tower of $\Sigma^*$-extensions that turns to a new efficient framework for the parameterized telescoping problem. Special instances of such $\Sigma^*$-extensions cover iterative sums such as the harmonic numbers and generalized versions that arise, e.g., in combinatorics, computer science or particle physics. Moreover, we illustrate how these new ideas can be used to reduce the depth of the given sum and provide structural theorems that connect complete reductions to Karr's Fundamental Theorem of symbolic summation.