Shaurya Shriyam
| Home |
| Research |
| Courses |
| Publications |
| Blog |
|
Research |
My research interests lie in the application areas (robotics, logistics, healthcare, urban transport) of planning, inference, and learning algorithms aimed towards social good or industrial automation. More elaborately speaking, I am keenly interested in simulation modeling, heuristic optimization, multi-agent planning, reinforcement learning, sequential decision-making for public systems, swarm intelligence simulation, causal inference for social good, game-theoretic insights into relevant engineering problems, and network science methodologies. I am open to multidisciplinary collaborative projects. |
|
Research
|
Welcome to the \(``\)Laboratory of Empirical Algorithmics for Modeling Analytics\("\) abbreviated as LEAMA and pronounced as lemma! We are temporarily located at Block \(3\) Room \(369\). We also occupy some space in PSL & VB \(1\)st floor. Good LaTeX\(\mathbin{/}\)Overleaf skills and a fair amount of coding experience in one out of MATLAB, Python, R, or Julia are the bare minimum requirements for joining our research group if there is an open position.
A preamble: Complex decision-making scenarios are a common occurrence when dealing with industrial and robotic systems; resource distribution systems such as food distribution, judiciary, transportation, public health, and education systems; socioeconomic multi-layered network processes arising in organizations, institutions, and global multilateral agencies; problems involving resource mapping and distribution under constraints and uncertainty.
The standard solution procedures consist mainly of two parts: modeling and optimization. It is not possible to carefully study any of the above problems without delving into both. However, our focus tends to be rather more towards developing novel data-driven solutions for the modeling phase because in most cases, the real challenge is to capture the salient features of the real-world process by building a model at the right fidelity. This mainly involves a trade-off between building too detailed a model that may be highly time-consuming as well as computationally intractable and building too simple a model that may not furnish any novel insight.
Our primary goal is to understand decision problems of relevance to policymakers and industry leaders, especially in the context of countries of the Global South in general and India in particular while providing novel insights. We broadly try to focus on two types of decision problems: those that arise during the deployment of multi-robot systems in industries as well as those occurring during the optimal deployment of limited public health resources to maximize social good. For both, but especially the first, we also try to develop \(``\)physics-informed bio-inspired\("\) algorithms in an effort to augment the (usually applied) more rigorous yet also less flexible mathematical frameworks with more naturalized holistic frameworks.
The focus of our lab is on leveraging the recent rise in available computing power for research purposes to experiment with different algorithmic designs and modeling frameworks as opposed to doing a purely theoretical analysis or simply deploying existing analytical techniques using black-box applications-oriented approaches. To make this point explicit, we use the phrase \(``\)empirical algorithmics.\("\) The purpose of such algorithmic research methodologies undertaken by us is to strike an optimum balance between purely theoretical and simply applied research. Moreover, analytics refers to using data to make effective decisions. By \(``\)modeling analytics,\("\) we refer to the use of data for developing a model in such a way that after being coupled with a suitable inference\(\mathbin{/}\)learning engine (which under the hood is simply going to be an optimization solver), it may furnish satisfactory insights and provide effective assistance to the end-users in selecting optimal strategies.
Our research along the optimization direction mainly focuses on the following points. Firstly, such research studies as mentioned above furnish relevant and useful results only when we design the objective function carefully by taking into account all the deciding factors as well as the involved uncertainty. Secondly, since there is no shortage of optimization methods available in the literature, a major research component is to model the concerned scenario, thus making it amenable to a particular class of optimization techniques in such a way that high-quality solutions may be obtained in reasonable amount of time. Thirdly, although the mathematical programming methods and gradient-based methods form the bedrock of optimization theory, there are excellent reasons to alternatively consider incorporating metaheuristic optimization approaches.
|
|
Research
|
|