Research Projects

One of the key pillars of the current scientific method is to infer the underlying process from the observational and experimental evidence. This task is challenging when the available data is large but the predicted signal size is arbitrary and small. A few noteworthy anomalies that have recently been observed in two very different domains, namely the domain of the extremely big (cosmology) and the domain of the highly energetic, serve as specific examples. We propose to scan through the publicly accessible enormous data using Artificial Neural Networks, Convolution Neural Networks, and possibly Recurrent or Bayesian Neural Networks as well. We intend to apply algorithms developed for network training in some circumstances, such as Steepest Gradient Descent back propagation, Conjugate Gradient, Levenberg-Marquardt, etc. In order to determine the most likely underlying mechanism once the existence of a signal has been established, we would employ methods like Markov Chain, Monte Carlo etc. High Energy Particle Physics (a) Address recent discrepancies (such as the muon anomalous magnetic moment, the decay of B-hadrons, the top-quark forward-backward asymmetry and others) along-with other longstanding issues (neutrino masses, existence of dark matter, baryon asymmetry, the hierarchy problem) attest to the incompleteness of the Standard Model (SM). (b) We plan to use multi-top production as a tool to investigate this using ANNs, taking various observable features and process them to enhance the power to discriminate deviations from SM. Gravity and Cosmology (a) To study the discrepancy between the Hubble constants as derived from the Cosmic Microwave Background data (Planck) on the one hand and that obtained from supernova Ia data. We propose a reanalysis of the data that may arise from a non-simplistic coupling between gravity and matter. (b) To understand Primordial magnetogenesis which remains unexplainable in the standard framework. An ab initio study is called for. (c) The resolution of the Big-Bang singularity has been a long-standing problem. While there are potential solutions in the form of Bouncing Cosmology, the exact nature of the modified gravity remains a vexing issue as it must conform to all cosmological observations many of which would need to be reinterpreted. Nuclear/Particle interface and gravity (d) Completing the circle, the much vaunted AdS/ QCD duality has the potential to recast the vexing problem of hadron spectroscopy in terms of gravity duals. To solve the inverse problem of finding the gravity analog, we plan to accommodate hadron two-point functions as a supervised learning, and use the observed masses (at zero temperature) as the training data. Building the NN as a discretized version of the bulk action, the dilaton field and the metric would as network weights. This has the potential to predict the masses of as yet unobserved hadrons.