Source author record

Subhadeep Roy

Subhadeep Roy appears in the imported research catalog. Authorship, coauthor and topic links are available while profile ownership is still unclaimed.

ResearcherUnclaimed source record

cond-mat.dis-nn cond-mat.stat-mech Artificial Intelligence Computer Vision cond-mat.mtrl-sci cond-mat.soft physics.soc-ph

Catalog footprint

What is connected

10works

7topics

4close collaborators

Actions

Connect this record

Open graph Browse works

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2026arXiv

Prototypicality Bias Reveals Blindspots in Multimodal Evaluation Metrics

Automatic metrics are now central to evaluating text-to-image models, often substituting for human judgment in benchmarking and large-scale filtering. However, it remains unclear whether these metrics truly prioritize semantic correctness or instead favor visually and socially prototypical images learned from biased data distributions. We identify and study prototypicality bias as a systematic failure mode in multimodal evaluation. We introduce a controlled contrastive benchmark ProtoBias (Prototypical Bias), spanning Animals, Objects, and Demography images, where semantically correct but non-prototypical images are paired with subtly incorrect yet prototypical adversarial counterparts. This setup enables a directional evaluation of whether metrics follow textual semantics or default to prototypes. Our results show that widely used metrics, including CLIPScore, PickScore, and VQA-based scores, frequently misrank these pairs, while even LLM-as-Judge systems exhibit uneven robustness in socially grounded cases. Human evaluations consistently favour semantic correctness with larger decision margins. Motivated by these findings, we propose ProtoScore, a robust 7B-parameter metric that substantially reduces failure rates and suppresses misranking, while running at orders of magnitude faster than the inference time of GPT-5, approaching the robustness of much larger closed-source judges.

preprint2022arXiv

Model for creep failure with healing

To understand the general properties of creep failure with healing effects, we study a mean-field fiber bundle model with probabilistic rupture and rejoining processes. The dynamics of the model are determined by two factors: bond breaking and the formation of new bonds. Steady states are realized due to the balance between breaking and healing beyond a critical healing factor, below which the bundle breaks completely. Correlation between the fluctuating value of strain generated in the model with time at the steady-state leads to a characteristic time that diverges in a scale-free manner as we approach the critical healing factor. Transient behaviors in strain rate also involve a power law with a non-universal exponent.

preprint2022arXiv

Modeling crack propagation in heterogeneous materials: Griffith's law, intrinsic crack resistance and avalanches

Various kinds of heterogeneity in solids including atomistic discreteness affect the fracture strength as well as the failure dynamics remarkably. Here we study the effects of an initial crack in a discrete model for fracture in heterogeneous materials, known as the fiber bundle model. We find three distinct regimes for fracture dynamics depending on the initial crack size. If the initial crack is smaller than a certain value, it does not affect the rupture dynamics and the critical stress. While for a larger initial crack, the growth of the crack leads to a breakdown of the entire system, and the critical stress depends on the crack size in a power-law manner with a nontrivial exponent. The exponent, as well as the limiting crack size, depend on the strength of heterogeneity and the range of stress relaxation in the system.

preprint2021arXiv

Opinion dynamics: Public and private

We study here the dynamics of opinion formation in a society where we take into account of the internally held beliefs and externally expressed opinions of the individuals, which are not necessarily the same at all times. While these two components can influence one another, their difference, both in dynamics and in the steady state, poses interesting scenarios in terms of the transition to consensus in the society and characterizations of such consensus. Here we study this public and private opinion dynamics and the critical behavior of the consensus forming transitions, using a kinetic exchange model.

preprint2020arXiv

Crack Localization and the Interplay between Stress Enhancement and Thermal Noise

We study the competition between thermal fluctuations and stress enhancement in the failure process of a disordered system by using a local load sharing fiber bundle model. The thermal noise is introduced by defining a failure probability that constitutes the temperature and elastic energy of the fibers. We observe that at a finite temperature and low disorder strength, the failure process, which nucleate in the absence of any thermal fluctuation, becomes spatially uncorrelated when the applied stress is sufficiently low. The dynamics of the model in this limit lies closely to the universality class of ordinary percolation. When applied stress is increased beyond a threshold value, localized fractures appear in the system that grow with time. We identify the boundary between the localized and random failure process in the space of temperature and applied stress, and find that the threshold of stress corresponding to the onset of localized crack growth increases with the increase of temperature.

preprint2020arXiv

Size distribution of emitted energies in local load sharing fiber bundles

We study the local load sharing fiber bundle model and its energy burst statistics. While it is known that the avalanche size distribution of the model is exponential, we numerically show here that the avalanche size ($s$) and the corresponding energy burst ($E$) in this version of the model have a non-linear relation ($E\sim s^γ$). Numerical results indicate that $γ\approx 2.5$ universally for different failure threshold distributions. With this numerical observation, it is then possible to show that the energy burst distribution is a power law, with a universal exponent value of $-(γ+1)$.

preprint2019arXiv

Creep failure in a threshold activated dynamics: Role of temperature during a sub-critical loading

Creep is a time-dependent deformation of solids at relatively low stresses, leading to the breakdown with time. Here we propose a simple model for creep failure of disordered solids, in which temperature and stress are controllable. Despite its simplicity, this model can reproduce most experimental observations. Time dependence of the strain rate is well fitted with power laws resembling the Omori-Utsu and the inverse Omori laws in the primary and the tertiary creep regimes, respectively. Distribution of the creep lifetime obeys the log-normal distribution, and the average creep lifetime decays in a scale-free manner with the increasing stress. The above results are in good agreement with experiments. Additionally, the mean avalanche size as a function of temperature exhibits a series of jumps, and finite-size scaling implies the existence of phase transitions.

preprint2016arXiv

A Recipe for Composite Materials: An Approach through Fiber Bundle Model

Strengthening of materials and preventing abrupt fracture are really challenging jobs in the field of engineering and material science. Such problems can be resolved by using composite materials. In this work, we have studied the fracture process of a composite material in light of fiber bundle model with different elastic constants as well as different random threshold breaking strength of fibers. The critical width of the threshold distribution ($δ_c$), for which abrupt failure occurs, is studied both analytically and numerically with increasing number of components $(k)$ in the composite and it is shown that $δ_c$ is inversely related to $k$. Corresponding phase diagram for the model suggests decrease in the tendency of abrupt fracture as number of components in the composite increase.

preprint2015arXiv

Nucleation versus percolation: Scaling criterion for failure in disordered solids

One of the major factors governing the mode of failure in disordered solids is the effective range $R$, over which the stress field is modified following a local rupture event. In random fiber bundle model, considered as a prototype of disordered solids, we show that the failure mode is nucleation dominated in the large system size limit, as long as $R$ scales slower than $L^ζ$, with $ζ=2/3$. For a faster increase in $R$, the failure properties are dominated by the mean-field critical point, where the damages are uncorrelated in space. In that limit, the precursory avalanches of all sizes are obtained even in the large system size limit. We expect these results to be valid for systems with finite (normalizable) disorder.

preprint2014arXiv

Criticality in Fiber Bundle Model

We report a novel critical behavior in the breakdown of an equal load sharing fiber bundle model at a dispersion $δ_c$ of the breaking threshold of the fibers. For $δ< δ_c$, there is a finite probability $P_b$, that rupturing of the weakest fiber leads to the failure of the entire system. For $δ\geq δ_c$, $P_b = 0$. At $δ_c, P_b \sim L^{-η}$, with $η\approx 1/3$, where $L$ is the size of the system. As $δ\rightarrow δ_c$, the relaxation time $τ$ diverges obeying the finite size scaling law: $τ\sim L^β(|δ-δ_c| L^α)$ with $α, β= 0.33 \pm 0.05$. At $δ_c$, the system fails, at the critical load, in avalanches (of rupturing fibers) of all sizes $s$ following the distribution $P(s) \sim s^{-κ}$, with $κ= 0.50 \pm 0.01$. We relate this critical behavior to brittle to quasi-brittle transition.