Over the last decade, new-age technologies like Artificial Intelligence (AI) and Machine Learning (ML) have taken centre stage all over the globe. This has also steered the uptake of more professionals who have the technical know-how to develop applications that can enhance the efficiency of public and private organisations. 

Interestingly, this puts things in perspective. According to the US Bureau of Labor Statistics, the employment of computer and information research scientists (particularly those who are adept at AI) is predicted to grow by a staggering 21% from 2021 to 2031. 

Back home, there has always been demand for professionals with degrees in science and technology. With AI and ML being a much sought-after skill today, universities, too, are taking cognisance and contributing immensely to the growth of the industry through deep research and introduction of skill-forward curriculums. 

The Indian Institute of Science (IISc), located in the country’s Silicon Valley, Bengaluru has always been regarded as a leading institution for research and higher education in science, engineering, design, and management. Among the many initiatives, they have a state-of-the-art AI and ML centre in collaboration with Kotak Mahindra Bank. This centre offers courses at several levels – Bachelor’s, Master’s and short-term in areas such as AI, ML, deep learning, fintech, reinforcement learning, image processing and computer vision, among others.  

It’s time to take a walkthrough of the several path breaking innovations powered by IISc that have the potential to provide an impetus to the industry, one step at a time. 

Creating a real-time simulation 

For the unversed, real-time simulation is a computer model of a physical system that can execute at the same rate as an actual wall clock time. This isn’t the easiest feat to achieve, but the Department of Computer Science in IISc Bengaluru has been consistently conducting experiments in the application of machine learning in scientific computing, with assistance from Shell India. 

A research group, Scientific Machine Learning and Operations (STARS), led by Sashikumaar Ganeshan, Associate Professor and Chair, Department of Computational and Data Sciences (CDS), IISc Bengaluru has developed an AI-based framework that could aid in the real-time environment-aware high-fidelity simulation of a wind farm using Deep Learning, Model Order Reduction, and High-performance computing. 

However, like everything else, it isn’t devoid of challenges. One of the biggest issues that crops up is facing difficulty in real-time, high-fidelity simulations of fluid flows. Furthermore, it is also computationally expensive for complex geometries and data assimilation is problematic. What’s more, these models need to be retrained for change in simulation parameters that impacts the down time of the framework.

To address these challenges, the team led by Ganeshan presented a concept called digital shadowing that attempts to augment AI into the existing computational model, making it more accurate and robust. 

“This technology helps make predictions more accurate and reduces the system’s downtime. The approach in which machine learning is used in scientific computing is called Physics-informed neural networks (PINNs). This approach helps to solve the Partial differential equations (PDEs) using the neural network. This allows us to solve PDEs without using conventional methods, as they have many challenges,” he explains. 

They also leverage neural networks to solve the equations. That’s because such a model can eliminate high computational costs. The focus here is only on transferring the crucial information to build the model, which is termed as federated learning.

Another pressing issue is the occurrence of singularly perturbed problems that may produce spurious oscillations due to numerical discretization. The team attempted to analyse the effect of Artificial Neural Networks (ANN)’s hyperparameters on trying to predict the stability for SUPG-based Finite Element Methods (FEM). 

While these can be tackled by using SUPG, these schemes require stabilization parameters, which does not have a closed-form solution for higher dimensions. That’s exactly why they proposed a solution to develop an AI framework that could predict stabilisation parameters that are based on training data. Alongside, the team also analysed the effect of various hyperparameters on the final prediction outcome. 

To put things in perspective, IISc has developed multiple NN frameworks such as SPDE-ConvNet, SPDE-Net, AiStab-FEM which predicts the stability parameters of these systems. 

Computer vision and deep learning

Apart from the above-mentioned initiative, the team led by R. Venkatesh Babu, a professor at CDS has proposed a crowd counting model that maps a given crowd scene to its density. What is crowd analysis governed by? It could be inter-occlusion between people due to extreme crowding. Another reason could be a striking similarity in the appearance of people and background elements as well as large variability of camera viewpoints. 

Most of this is tackled by using multi-scale CNN architectures, recurrent networks and late fusion of features from multi-column CNN with different receptive fields. The research team at IISc has proposed switching convolutional neural networks that leverage variation of crowd density within an image, in order to enhance the accuracy and localization of the predicted crowd count. 

Here’s what the IISc website mentions. Patches from a grid within a crowd scene are relayed to independent CNN regressors based on crowd count prediction quality of the CNN established during training. The independent CNN regressors are designed to have different receptive fields and a switch classifier is trained to relay the crowd scene patch to the best CNN regressor. 

That’s not all — the team has also proposed solutions like 3D reconstruction from 2D, 3D human pose estimation, and HDR deghosting. 

In the space of machine learning, Venkatesh and team have come up with Domain Adaptation and Generalization as a solution to the absence of data in India. Adapting the model without minimal data to make it useful is a possibility. 

Furthermore, the team proposed the first model to DeepFuse. In his article titled ‘DeepFuse: A Deep Unsupervised Approach for Exposure Fusion with Extreme Exposure Image Pairs’, Venkatesh explains that this is a novel deep learning architecture for fusing static multi-exposure images. 

“The current multi-exposure fusion (MEF) approaches use handcrafted features to fuse input sequences. However, the weak hand-crafted representations are not robust to varying input conditions. Moreover, they perform poorly for extreme exposure image pairs. Thus, it is highly desirable to have a method that is robust to varying input conditions and capable of handling extreme exposure without artefacts,” he adds. 

Space exploration

IISc has also made strides in space exploration with the help of its team led by Chandra R. Murthy, a professor in the department of Electrical Communication Engineering at the institute. 

IISc along with ISRO has developed a sustainable process for making brick-like structures on the moon. This works by utilising lunar soil, bacteria and guar beans to consolidate the soil into possible load-bearing structures. What’s more, these “space bricks” could eventually be used to assemble structures for habitation on the moon’s surface, the researchers suggest.

Most of us already know that the cost incurred to send even a little material to outer space is astounding. However, the process developed by ISRO and IISc uses urea and lunar oil to minimise the cost. 

The IISc website gives out all the details about the process. First, the bacteria was mixed with lunar soil. Then, required urea and calcium sources along with gum extracted from locally-sourced guar beans was added. The guar gum helped increase the strength of the material by serving as a scaffold for carbonate precipitation. The final product obtained after a few days of incubation was found to possess significant strength and machinability.

The last word 

With significant research by IISc in areas that were often difficult to tackle earlier or neglected, it is a positive sign that AI, ML and related technologies are serving in a big way to make a difference to humankind. 

They have many more projects in the pipeline. All we have to do is wait and watch for them to make the right noise!