CoronaActionIndia #1

Keerat Kaur Guliani
4 min readJun 11, 2020


Hello everyone.

This blog streak is meant to document my work experience and takeaways while participating in the capacity of a summer intern at IIIT-Delhi, under the guidance of Prof. Tavpritesh Sethi and Prof. Ponnurangam Kumaraguru, and working in a team of students from various institutions, including AIIMS and CSIR-IGIB, Delhi. Over the period of my internship, I plan to post several such blogs covering the work that we do.


The world has slowed down ever since the Corona pandemic has begun and it is costing us lives, whether directly or indirectly. All across the globe, leaders of States are desperately trying to come up with effective policy solutions to battle the COVID crisis, and Artificial Intelligence enthusiasts are stepping in to play their role, to analyse whatever real-time data they can lay their hands on and come up with ideas faster than the COVID spreads. Every country has a different curve and a different set of problems plaguing it.

About the Project

We, as a part of Team CoronaActionIndia, are looking for ways to model the various aspects of the Covid-19 paradigm pertinent to the Indian context, using various computational techniques. Our current focus lies in being able to come up with a set of Projections based on the models we make, representing a count of the number of expected cases in the months to follow so that Governments at the State and National level can act accordingly to implement relevant policies. The task-at-hand is an impact-oriented, real-time exercise.

Work done till now in this direction can be found in this paper.


We are employing Agent-Based Models to simulate the spread of the Covid-19 infection among the masses, and then draw the number of expected cases in India, and also State-wise, from these simulations. Agent-Based Models are especially useful when it comes to simulating epidemiological problems, such as the spread of an infectious disease because it takes into account the interaction between various members of the system at two levels:

  • Macro: At the level of the system.
  • Micro: At the level of the individual units (agents) that make up the system.

The latter is particularly important because, in the case of a pandemic such as the current one, it is the interaction between agents, as well as the agents and the environment, that directs the behaviour of the system. This is not taken into account by classical models/approaches.

Naturally, there arises a need to set certain parameters that can describe the interaction we are attempting to capture in the model, and aptly so. The table below highlights the assumptions we take in the process:

Source: The State-wise population figures are lifted from the recently published NITI Aayog data.

The basis for these assumptions is not arbitrary- it is lifted from published, verified literature on the web and so provides us with a sound basis to start. We enter these variables and design our system in a simulation software, and run simulations to get the required projections.

Results obtained have been put on this dashboard and are accessible to all.

Challenges Faced

The ease by which we’re able to explain the pipeline is deceptive- within the task, as always, lie certain challenges:

  1. An appropriate representation of Behaviour mechanisms: The trick with managing research with ABMs is to maintain a fine balance as far as the details of your model are concerned- go into enough detail so as to capture the behaviour appropriately, but not too much lest you may end up losing track of your model’s progress. The idea of trying to predict how humans are going to interact within the next few months and set certain parameters based on the same makes this task a little daunting.
  2. Validation of the model: Realise that the ground truth does not exist here- on the contrary, it is exactly what we are trying to predict, so we can avoid it in the best of our capacities in case it seems to be too overwhelming. Clearly, we have little to aid in the validation of our model (not a complete absence of methods, however- our task is to exploit the same).
  3. Pace of the Project: The number of active cases, those deceased and those recovered are changing every day- we are working in real-time and so, we need to get results in real-time too! It is definitely not as easy as trying to work in retrospect.
  4. Variation: The parameters involved assume different values for all the individual States and Union Territories. We have to keep a track of these variations, analyzing the conditions of every state so that they can contribute to our model.

Potential Impact

“I have a feeling enough testing isn’t being done.”

“I think the Government should have imposed lockdown sooner/later.”

These are statements all of us have heard in the past few months. The reason for the uncertainty is simple- we have no means to verify the basis on which governments across our country are imposing various strategies to battle COVID, and so everyone just seems to be functioning on blind faith- signs of impending chaos.

The tool we aim to develop will help us two-fold:

  1. It would give the citizens a concrete basis to question the testing and lockdown strategies imposed by the government.
  2. More importantly, it would help the government to direct its approach.

With this goal in mind, we aim to work on the aforementioned and deliver our best.

— — — — — — — — — — —

To all the readers,

It would be great to hear about your insights on the work we are doing. Healthy discourse is always welcome :)




Keerat Kaur Guliani

Research in Applied AI | Machine Intelligence & Deep Learning Enthusiast | Civil Engineering Undergrad @IIT_Roorkee