Species development experiences numerous forms of regulations at different embryonic stages because of numerous proteins capable of altering the underlying signaling dynamics of biological processes. Also, these processes are inherently noisy and often deviate away from the stipulated strength, yet the development of species demonstrates remarkable plasticity. Our research adopts stochastic and deterministic approaches to unravel the mechanisms that assure remarkable reproducibility of species development despite the presence of alternative sources of perturbations and noises of different forms. We adopt alternative context-based quantitative approaches.

• Ordinary Differential Equation (ODE) approach to study the dynamical and steady-state analysis of the signaling pathways

• Partial Differential Equation (PDE) approach for spatial analysis of scaling of morphogen mediated patterning

• Stochastic modeling of underlying regulatory networks of patterning signals and gene expression

The detection and estimation of randomly-scattered objects distributed in time or space are frequent in a broad range of applications in science and engineering, including communications theory, signal processing, ecology, and biological modeling. Many systems consider random point processes distributed randomly in the continuum (time or space) to describe the underlying phenomenon. Often it is necessary to estimate the rate of random events, where random observation by an observer attempts to detect the target of interest. In many systems, both target objects and the detection attempts to detect those are randomly distributed in a continuum. Intensity estimation of the randomly arriving targets in these systems is crucial for many critical applications.

We apply the statistical signal processing approach, precisely the estimation and detection theory concepts, to estimate the unknown rate parameter of random events and characterize the estimators as necessary.

Cells are excellent computational devices tuned and calibrated against different forms of perturbations by billions of years of evolutionary steps. For a robust and highly reproducible growth and maintenance of underlying biophysical processes, many signaling pathways continuously transmit and process information to guide the cellular processes and gene expression underneath. We capitalize on these signaling networks, their connectivity, and other forms of regulation to perform computation capable of classifying objects, applicable in intelligent drug design, DNA computing, and aqueous deployment for pollutant detections.

Specifically, we are interested in the physics of these processes and employ a multi-disciplinary approach at the interface of physics, math, and biology. Our focus is to design a biomolecular neural network (BNN) and extend it to develop task-specific models.

Natural Language Processing (NLP) tasks in non-dominant and low-resource languages have not experienced significant progress. Although pre-trained BERT models are available, GPU-dependency, large memory requirement, and data scarcity often limit their applicability. So, countries lacking substantial socio-economic capacity and technological infrastructures are lagging. The current trend of NLP research evolves mainly around a few dominant languages, leaving NLP research for many low-resource languages unattended or less explored when there is a surge of textual toxicity in social media. Interestingly, many textual classification tasks do not require a rigorous use of linguistic semantics. So, models structured well against the semantics, for instance, the BERT models, may not always be the most optimal choice in NLP tasks less dependent on language semantics.

Our research in NLP primarily focuses on the viable trade-off between the deployability, scarce annotated data, and DNN models’ accuracy in NLP tasks for low-resource languages and environments.

Advancements in deep learning models have propelled the emergence of computer vision, requiring computers or digital devices to assess images and videos and perform object detections automatically. These applications require an exhaustive exploration of an image at pixel levels to extract the necessary information to facilitate object recognition in applications from self-driving cars to security, disease detection, to medical imaging.

We harness existing image processing techniques to transform images into digital form and analyze them at pixel levels. We are also interested in model-based image analysis that combines convolutional neural networks and image processing techniques to perform image denoising and analysis of low-photon imaging.