arXiv subjects

Soham Biswas

Publications and source records attributed to Soham Biswas.

At least 19 recordsLinked to original sources

arXiv2026

Metadata-Aware Multi-Prompt Reasoning for Zero-Shot Accident Understanding

In this paper, we address the problem of zero-shot understanding of accidents from surveillance videos by identifying when an impact event occurs, what type of impact it is, and where in the frame it occurs using natural language. We propose a three-stage pipeline that decomposes the accident understanding into when, what, and where. The first stage extracts a short temporal window around the impact using vision-language similarity. In the second stage, we perform metadata-driven multi-prompt reasoning with five complementary views (baseline, motion, geometry, contrast, and tiebreaker) and resolve disagreement via an entropy-gated pairwise adjudicator. Finally, we localize the impact of an open-vocabulary detector queried on the predicted accident type and scene layout, and aggregate detections across keyframes using a score-weighted centroid. Our pipeline achieves a substantial improvement in the harmonic-mean score over a centre-of-frame baseline on the zero-shot ACCIDENT @ CVPR benchmark. We show that decomposing zero-shot video understanding into temporal localization, semantic classification, and spatial grounding enable more reliable reasoning with vision-language models than direct prompting alone.

cs.CV↗

arXiv2026

Correlation analysis of the dispersion of SARS-CoV-2 in Mexico

In this paper, we propose a method to analyze correlations in pandemic-related data across different geographical regions, relying on the analysis of correlations for non-stationary time series, which are typical of pandemic data. Unlike traditional epidemiological approaches focused on medical and modeling perspectives during a pandemic, our method emphasizes post-pandemic analysis to assess how societal responses; such as lockdowns, travel restrictions, mobility patterns, and vaccination campaigns, manifest in the collective behavior of regions. These insights can inform future public health strategies and enhance understanding of the complex dynamics underlying pandemic spread and control.

stat.AP↗

arXiv2024

Symmetry operations and Critical Behaviour in Classical to Quantum Stochastic Processes

Recently, a novel construction scheme for generating quantum analogs of classical stochastic processes has been introduced. Here, we use this scheme in order to generate a large class of self-contained quantum extensions of a classical Markov chain process using symmetry operations. We show that the relaxation processes unfold very differently for the different quantum extensions. This is supported by monitoring the coherence, the probability of reaching the equilibrium, the decay of the number of domain walls and the purity. Unexpectedly, we find a rather ambiguous relation between the coherence measure based on the L1-norm and the speed of the relaxation process. Finally we find that the finite size scaling of the coherence measure exists for both short and long times and the value of the critical exponent is different for the short and long time.

quant-ph↗

arXiv2024

Separate length scale for coarsening and for fractal formation by persistent sites

We present the first example where length scale for the growth of ordered regions and the correlation length for the two point correlations of persistent sites scale differently with time. We do so by studying a global spin exchange dynamics in one dimension where a selected spin interacts with its two nearest domains. We found domain growth exponent $z=2.47\pm 0.03$ and the persistence exponent $θ=0.445\pm0.002$, making $zθ> d=1$. Unlike any previous study, we found correlation length of two point correlation of persistent sites grows in a power law with exponent $ζ=1.00\pm 0.03$ by studying the fractal structure created by the persistent sites at the different stages of the dynamics and shown that fractal dimension is not related with any growth exponents.

cond-mat.stat-mech↗

arXiv2022

From Classical to quantum stochastic process

In this paper for the first time, we construct quantum analogs starting from classical stochastic processes, by replacing random which path decisions with superpositions of all paths. This procedure typically leads to non-unitary quantum evolution, where coherences are continuously generated and destroyed. In spite of their transient nature, these coherences can change the scaling behavior of classical observables. Using the zero temperature Glauber dynamics in a linear Ising spin chain, we find quantum analogs with different domain growth exponents. In some cases, this exponent is even smaller than for the original classical process, which means that coherence can play an important role to speed up the relaxation process.

cond-mat.stat-mech↗

arXiv2021

Ballistic annihilation in one dimension : A critical review

In this article we review the problem of reaction annihilation $A+A \rightarrow \emptyset$ on a real lattice in one dimension, where $A$ particles move ballistically in one direction with a discrete set of possible velocities. We first discuss the case of pure ballistic annihilation, that is a model in which each particle moves simultaneously at constant speed. We then review ballistic annihilation with superimposed diffusion in one dimension. This model consists of diffusing particles each of which diffuses with a fixed bias, which can be either positive or negative with probability $1/2$, and annihilate upon contact. When the initial concentration of left and right moving particles is same, the concentration $c(t)$ decays as $t^{-1/2}$ with time, for pure ballistic annihilation. However when the diffusion is superimposed decay is faster and the concentration $c(t) \sim t^{-3/4}$. We also discuss the nearest-neighbor distance distribution as well as crossover behavior.

cond-mat.stat-mech↗

arXiv2019

Zero temperature ordering dynamics in two dimensional BNNNI model

We investigate the dynamics of a two dimensional bi-axial next nearest neighbour Ising (BNNNI) model following a quench to zero temperature. The Hamiltonian is given by $H = -J_0\sum_{i,j=1}^L [(S_{i,j}S_{i+1,j}+S_{i,j}S_{i,j+1}) -κ(S_{i,j}S_{i+2,j} + S_{i,j}S_{i,j+2})]$ . For $κ<1$, the system does not reach the equilibrium ground state and keep evolving in active states for ever. For $κ\geq 1$, though the system reaches a final state, but it do not reach the ground state always and freezes to a striped state with a finite probability like two dimensional ferromagnetic Ising model and ANNNI model. The overall dynamical behaviour for $κ> 1$ and $κ=1$ is quite different. The residual energy decays in a power law for both $ κ>1$ and $κ=1$ from which the dynamical exponent $z$ have been estimated. The persistence probability shows algebraic decay for $κ> 1$ with an exponent $θ= 0.22 \pm 0.002$ while the dynamical exponent for ordering $z=2.33\pm 0.01$. For $κ=1$, the system belongs to a completely different dynamical class with $θ= 0.332 \pm 0.002$ and $z=2.47\pm 0.04$. We have computed the freezing probability for different values of $κ$. We have also studied the decay of autocorrelation function with time for different regime of $κ$ values. The results have been compared with that of the two dimensional ANNNI model.

cond-mat.stat-mech↗

arXiv2016

Correlation Matrix Spectra: A Tool for Detecting Non-apparent Correlations?

It has been shown that, if a model displays long-range (power-law) spatial correlations, its equal-time correlation matrix of this model will also have a power law tail in the distribution of its high-lying eigenvalues. The purpose of this letter is to show that the converse is generally incorrect: a power-law tail in the high-lying eigenvalues of the correlation matrix may exist even in the absence of equal-time power law correlations in the original model. We may therefore view the study of the eigenvalue distribution of the correlation matrix as a more powerful tool than the study of correlations, one which may in fact uncover structure, that would otherwise not be apparent. Specifically, we show that in the Totally Asymmetric Simple Exclusion Process, whereas there are no clearly visible correlations in the steady state, the eigenvalues of its correlation matrix exhibit a rich structure which we describe in detail.

cond-mat.stat-mech↗

arXiv2016

Novel Dynamical Phenomena in Magnetic systems

Dynamics of Ising models is a much studied phenomenon and has emerged as a rich field of present-day research. An important dynamical feature commonly studied is the quenching phenomenon below the critical temperature. In this thesis we have studied the zero temperature quenching dynamics of different Ising spin systems. First we have studied the zero temperature quenching dynamics of two dimensional Ising spin system with competating interactions. Then we have studied the effect of randomness or disorder on the quenching dynamics of Ising spin system. We have studied the effect of the nature of randomness on zero temperature quenching dynamics of one dimensional Ising model on two type of complex networks. A model for opinion dynamics also has been proposed in this thesis, in which the binary opinions of the individuals are determined according to the size of their neighboring domains. This model can be equivalently defined in terms of Ising spin variables and the various quantities studied have one to one correspondence with magnetic systems. Introducing disorder in this model through a parameter called rigidity parameter $ρ$ (probability that people are completely rigid and never change their opinion), the transition to a heterogeneous society at $ρ= 0^{+}$ is obtained. The Model (Model I) has been generalized introducing a parameter named as size sensitivity parameter to modify the dynamics of the proposed model and a macroscopic crossover in time is observed for the intermediate values of this parameter.

cond-mat.stat-mech↗

arXiv2016

Ballistic annihilation with superimposed diffusion in one dimension

We consider a one-dimensional system with particles having either positive or negative velocity, which annihilate on contact. To the ballistic motion of the particle, a diffusion is superimposed. The annihilation may represent a reaction in which the two particles yield an inert species. This model has been the object of previous work, in which it was shown that the particle concentration decays faster than either the purely ballistic or the purely diffusive case. We report on previously unnoticed behaviour for large times, when only one of the two species remains and also unravel the underlying fractal structure present in the system. We also consider in detail the case in which the initial concentration of right-going particles is $1/2+\varepsilon$, with $\varepsilon\neq0$. It is shown that a remarkably rich behaviour arises, in which two crossover times are observed as $\varepsilon \to 0$.

cond-mat.stat-mech↗

arXiv2014

Universal features of exit probability in opinion dynamics models with domain size dependent dynamics

We study the exit probability for several binary opinion dynamics models in one dimension in which the opinion state (represented by $\pm 1$) of an agent is determined by dynamical rules dependent on the size of its neighbouring domains. In all these models, we find the exit probability behaves like a step function in the thermodynamic limit. In a finite system of size $L$, the exit probability $E(x)$ as a function of the initial fraction $x$ of one type of opinion is given by $E(x) = f[(x-x_c)L^{1/ν}]$ with a universal value of $ν= 2.5 \pm 0.03$. The form of the scaling function is also universal: $f(y) = [\tanh(λy +c) +1]/2$, where $λ$ is found to be dependent on the particular dynamics. The variation of $λ$ against the parameters of the models is studied. $c$ is non-zero only when the dynamical rule distinguishes between $\pm 1$ states; comparison with theoretical estimates in this case shows very good agreement.

cond-mat.stat-mech↗

arXiv2014

Exit probability in inflow dynamics: nonuniversality induced by range, asymmetry and fluctuation

Probing deeper into the existing issues regarding the exit probability (EP) in one dimensional dynamical models, we consider several models where the states are represented by Ising spins and the information flows inwards. At zero temperature, these systems evolve to either of two absorbing states. The exit probability $E(x)$, which is the probability that the system ends up with all spins up starting with $x$ fraction of up spins is found to have the general form $E(x) = x^α/\left[x^α+ (1-x)^α\right]$. The exit probability exponent $α$ strongly depends on $r$, the range of interaction, the symmetry of the model and the induced fluctuation. Even in a nearest neighbour model, nonlinear form of EP can be obtained by controlling the fluctuations and for the same range, different models give different results for $α$. Non-universal behaviour of the exit probability is thus clearly established and the results are compared to existing studies in models with outflow dynamics to distinguish the two dynamical scenarios.

cond-mat.stat-mech↗

arXiv2013

A stochastic opinion dynamics model with domain size dependent dynamic evolution

We introduce a stochastic model of binary opinion dynamics in one dimension. The binary opinions $\pm 1$ are analogous to up and down Ising spins and in the equivalent spin system, only the spins at the domain boundary can flip. The probability that a spin at the boundary is up is taken as $P_{up} = \frac {s_{up}} {s_{up} + δs_{down}}$ where $s_{up} (s_{down})$ denotes the size of the domain with up (down) spins neighbouring it. With $x$ fraction of up spins initially, a phase transition is observed in terms of the exit probability and the phase boundary is obtained in the $δ-x$ plane. In addition, we investigate the coarsening behaviour starting from a completely random state; conventional scaling is observed only at the phase transition point $δ= 1$. The scaling behaviour is compared to other dynamical phenomena; the model apparently belongs to a new dynamical universility class as far as persistence is concerned although the dynamical exponent, equal to one, is identical to a similar model with no stochasticity.

cond-mat.stat-mech↗

arXiv2012

Effect of the nature of randomness on quenching dynamics of Ising model on complex networks

Randomness is known to affect the dynamical behaviour of many systems to a large extent. In this paper we investigate how the nature of randomness affects the dynamics in a zero temperature quench of Ising model on two types of random networks. In both the networks, which are embedded in a one dimensional space, the first neighbour connections exist and the average degree is four per node. In the random model A, the second neighbour connections are rewired with a probability $p$ while in the random model B, additional connections between neighbours at Euclidean distance $l ~ (l >1)$ are introduced with a probability $P(l) \propto l^{-α}$. We find that for both models, the dynamics leads to freezing such that the system gets locked in a disordered state. The point at which the disorder of the nonequilibrium steady state is maximum is located. Behaviour of dynamical quantities like residual energy, order parameter and persistence are discussed and compared. Overall, the behaviour of physical quantities are similar although subtle differences are observed due to the difference in the nature of randomness.

cond-mat.stat-mech↗

arXiv2011

Opinion dynamics model with domain size dependent dynamics: novel features and new universality class

A model for opinion dynamics (Model I) has been recently introduced in which the binary opinions of the individuals are determined according to the size of their neighboring domains (population having the same opinion). The coarsening dynamics of the equivalent Ising model shows power law behavior and has been found to belong to a new universality class with the dynamic exponent $z=1.0 \pm 0.01$ and persistence exponent $θ\simeq 0.235$ in one dimension. The critical behavior has been found to be robust for a large variety of annealed disorder that has been studied. Further, by mapping Model I to a system of random walkers in one dimension with a tendency to walk towards their nearest neighbour with probability $ε$, we find that for any $ε> 0.5$, the Model I dynamical behaviour is prevalent at long times.

physics.soc-ph↗

arXiv2011

Novel ballistic to diffusive crossover in the dynamics of a one dimensional Ising model with variable range of interaction

The idea that the dynamics of a spin is determined by the size of its neighbouring domains was recently introduced (S. Biswas and P. Sen, Phys. Rev. E {\bf 80}, 027101 (2009)) in a Ising spin model (henceforth, referred to as model I). A parameter $p$ is now defined to modify the dynamics such that a spin can sense domain sizes up to $R = pL/2$ in a one dimensional system of size $L$. For the cutoff factor $p$ \to 0$, the dynamics is Ising like and the domains grow with time $t$ diffusively as $ t^{1/z}$ with $z=2$, while for $p=1$, the original model I showed ballistic dynamics with $z \simeq 1$. For intermediate values of $p$, the domain growth, magnetisation and persistence show model I like behaviour up to a macroscopic crossover time $ t_1 \sim pL/2$. Beyond $t_1$, characteristic power law variations of the dynamic quantities are no longer observed. The total time to reach equilibrium is found to be $t = apL + b(1-p)^3L^2$, from which we conclude that the later time behaviour is diffusive. We also consider the case when a random but quenched value of $p$ is used for each spin for which ballistic behaviour is once again obtained.

cond-mat.stat-mech↗

arXiv2010

Complex Networks: effect of subtle changes in nature of randomness

In two different classes of network models, namely, the Watts Strogatz type and the Euclidean type, subtle changes have been introduced in the randomness. In the Watts Strogatz type network, rewiring has been done in different ways and although the qualitative results remain same, finite differences in the exponents are observed. In the Euclidean type networks, where at least one finite phase transition occurs, two models differing in a similar way have been considered. The results show a possible shift in one of the phase transition points but no change in the values of the exponents. The WS and Euclidean type models are equivalent for extreme values of the parameters; we compare their behaviour for intermediate values.

cond-mat.stat-mech↗

arXiv2009

A new model of binary opinion dynamics: coarsening and effect of disorder

We propose a model of binary opinion in which the opinion of the individuals change according to the state of their neighbouring domains. If the neighbouring domains have opposite opinions, then the opinion of the domain with the larger size is followed. Starting from a random configuration, the system evolves to a homogeneous state. The dynamical evolution show novel scaling behaviour with the persistence exponent $θ\simeq 0.235$ and dynamic exponent $z \simeq1.02 \pm 0.02$. Introducing disorder through a parameter called rigidity coefficient $ρ$ (probability that people are completely rigid and never change their opinion), the transition to a heterogeneous society at $ρ= 0^{+}$ is obtained. Close to $ρ=0$, the equilibrium values of the dynamic variables show power law scaling behaviour with $ρ$. We also discuss the effect of having both quenched and annealed disorder in the system.

cond-mat.stat-mech↗