Search

Classifying player actions from soccer videos is a challenging problem, which has become increasingly important in sports analytics over the years. Most state-of-the-art methods employ highly complex offline networks, which makes it difficult to deploy such models in resource constrained scenarios. Here, in this paper we propose a novel end-to-end knowledge distillation based transfer learning network pre-trained on the Kinetics400 dataset and then perform extensive analysis on the learned framework by introducing a unique loss parameterization. We also introduce a new dataset named SoccerDB1 containing 448 videos and consisting of 4 diverse classes each of players playing soccer. Furthermore, we introduce an unique loss parameter that help us linearly weigh the extent to which the predictions of each network are utilized. Finally, we also perform a thorough performance study using various changed hyperparameters. We also benchmark the first classification results on the new SoccerDB1 dataset obtaining 67.20% validation accuracy. Apart from outperforming prior arts significantly, our model also generalizes to new datasets easily. The dataset has been made publicly available at: https://bit.ly/soccerdb1, Comment: Accepted to 10th Springer PReMI 2023

Diverting, and controlling, elastic vibrations impacting upon infrastructure is a major challenge for seismic hazard mitigation, and for the reduction of machine noise and vehicle vibration in the urban environment. Seismic metamaterials (SMs), with their inherent ability to manipulate wave propagation, provide a key route for overcoming the technological hurdles involved in this challenge. Engineering the structure of the SM serves as a basis to tune and enhance its functionality, and inspired by split rings, swiss-rolls, notch-shaped and labyrinthine designs of elementary cells in electromagnetic and mechanical metamaterials, we investigate altering the structure geometries of SMs with the aim of creating large bandgaps \textcolor{black}{in a subwavelength regime}. We show that square stiff inclusions, perform better in comparison to circular ones, whilst keeping the same filling fraction. En route to enhancing the bandgap, we have also studied the performance of SMs with different constituent materials; we find that steel columns, as inclusions, show large bandgaps, however, the columns are too large for steel to be a feasible material in practical or financial terms. Non-reinforced concrete would be preferable for industry level scaling up of the technology because, concrete is cost-effective, easy to cast directly at the construction site and easy to provide arbitrary geometry of the structure. As a part of this study, we show that concrete columns can also be designed to exhibit bandgaps if we cast them within a soft soil coating surrounding the protected area for various civil structures like a bridge, building, oil pipelines etc.

Stochastic thermodynamics is an important development in the direction of finding general thermodynamic principles for non-equilibrium systems. We believe stochastic thermodynamics has the potential to benefit from the measure-theoretic framework of stochastic differential equations. Towards this, in this work, we show that Fluctuation Theorem (FT) is a special case of the Girsanov theorem, which is an important result in the theory of stochastic differential equations. We report that by employing Girsanov transformation of measures between the forward and the reversed dynamics of a general class of Langevin dynamic systems, we arrive at the Integral Fluctuation Relation. Following the same approach, we derive the FT also for the overdamped case. Our derivation is applicable to both transient and steady state conditions and can also incorporate diffusion coefficients varying as a function of state and time. We expect that the proposed method will be an easy route towards deriving the FT irrespective of the complexity and non-linearity of the system.

Starting with a micropolar formulation, known to account for nonlocal microstructural effects at the continuum level, a generalized Langevin equation (GLE) for a particle, describing the predominant motion of a localized region through a single displacement degree-of-freedom (DOF), is derived. The GLE features a memory dependent multiplicative or internal noise, which appears upon recognising that the micro-rotation variables possess randomness owing to an uncertainty principle. Unlike its classical version, the new GLE qualitatively reproduces the experimentally measured fluctuations in the steady-state mean square displacement of scattering centers in a polyvinyl alcohol slab. The origin of the fluctuations is traced to nonlocal spatial interactions within the continuum. A constraint equation, similar to a fluctuation dissipation theorem (FDT), is shown to statistically relate the internal noise to the other parameters in the GLE.

A global optimization framework, acronymed COMBEO (Change OfMeasure Based Evolutionary Optimization), is proposed. An important aspect in the development is a set of derivative-free additive directional terms obtainable through a change of measures en route to the imposition of any stipulated conditions aimed at driving the realized design variables (particles) to the global optimum. The generalized setting offered by the new approach also enables several basic ideas, used with other global search methods such as the particle swarm or the differential evolution, to be rationally incorporated in the proposed setup via a change of measures. The global search may be further aided by imparting to the directional update terms additional layers of random perturbations such as scrambling and selection. Depending on the precise choice of the optimality conditions and the extent of random perturbation, the search can be readily rendered either greedy or more exploratory. As numerically demonstrated, the new proposal appears to provide for a more rational, more accurate and faster alternative to most available evolutionary optimization schemes, prominent amongst which are the differential evolution and the particle swarm methods., Comment: 38 pages; 2 tables; 4 figures. arXiv admin note: substantial text overlap with arXiv:1403.1680

A new global stochastic search, guided mainly through derivative-free directional information computable from the sample statistical moments of the design variables within a Monte Carlo setup, is proposed. The search is aided by imparting to a directional update term, which parallels the conventional Gateaux derivative used in a local search for the extrema of smooth cost functionals, additional layers of random perturbations referred to as 'coalescence' and 'scrambling'. A selection scheme, constituting yet another avenue for random perturbation, completes the global search. The direction-driven nature of the search is manifest in the local extremization and coalescence components, which are posed as martingale problems that yield gain-like update terms upon discretization. As anticipated and numerically demonstrated, to a limited extent, against the problem of parameter recovery given the chaotic response histories of a couple of nonlinear oscillators, the proposed method apparently provides for a more rational, more accurate and faster alternative to most available evolutionary schemes, prominently the particle swarm optimization., Comment: 35 pages, 2 figures; being submitted to Physica D

Despite the cheap availability of computing resources enabling faster Monte Carlo simulations, the potential benefits of particle filtering in revealing accurate statistical information on the imprecisely known model parameters or modeling errors of dynamical systems, based on limited time series data, have not been quite realized. A major numerical bottleneck precipitating this under-performance, especially for higher dimensional systems, is the progressive particle impoverishment owing to weight collapse and the aim of the current work is to address this problem by replacing weight-based updates through additive ones. Thus, in the context of nonlinear filtering problems, a novel additive particle update scheme, in its non-iterative and iterative forms, is proposed based on manipulations of the innovation integral in the governing Kushner-Stratonovich equation. Numerical evidence for the identification of nonlinear and large dimensional dynamical systems indicates a substantively superior performance of the non- iterative version of the EnKS vis-\`a-vis most existing filters. The costlier iterative version, though conceptually elegant, mostly appears to effect a marginal improvement in the reconstruction accuracy over its non-iterative counterpart. Prominent in the reported numerical comparisons are variants of the Ensemble Kalman Filter (EnKF) that also use additive updates, albeit with many inherent limitations of a Kalman filter.