31-May-2022: Machine learning helps predict new materials for nano alloys, semiconductors & rare earths

Scientists have used Machine learning to develop a design map of alloys at the nanoscale which can help predict the match of pairs of metals that can form bimetallic nanoalloys.

These nano alloys, also called core-shell nanocluster alloys, in which one metal forms the core and another stays on the surface as a shell, are a new frontier in the quest of scientists for new materials and have applications in biomedicine and other areas.

It is important to know under what conditions core-shell structures are formed in the nanocluster alloys and which metal forms the core, and which stays on the surface as a shell. A number of factors like cohesive energy difference, atomic radius difference, surface energy difference and electronegativity of the two atoms may play a part in the core and shell preference of the atoms.

The periodic table has 95 metals of different categories ranging from alkalis to alkaline earth, which can potentially form 4465 pairs. It is experimentally impossible to determine how they behave in forming nanocluster alloys. But computers can be programmed to predict the behaviour of these pairs and more through ‘machine learning’. The machine is taught to recognise patterns by feeding in a number of patterns with well-defined attributes. The more the data fed into the computer, the more accurate will be the recognition of an unknown data by the computer.

However, scientists faced a stumbling block here because of the limited number of experimentally synthesised binary nanoclusters with clear identification of the chemical ordering of constituents, and few core−shell combinations studied theoretically. Machine Learning could not be applied with confidence on small data set of sizes less than or around 100.

Researchers at the S N Bose Centre for Basic Sciences, an autonomous institute of the Department of Science and Technology, circumvented this problem by calculating the Surface-to-core relative energy on a variety of possible binary combinations of alkali metals, alkaline earth, basic metals, transition metals and p-block metals to create a large data-set of 903 binary combinations.

In their paper published in the Journal of Physical Chemistry, they investigated the key attributes driving the core−shell morphology using the statistical tool of machine learning applied on this large data set. Core-shell structures with lighter metals having lower atomic numbers in the core were classified as Type 1, and those having the heavier metals in the core were classified as Type 2. A number of attributes were built to characterise each data point in the set. The performance of the ML model was tallied with existing experimental data, and the ML model was proved to be reliable.

Having thus established confidence in the ML model, the dominant attributes driving the core-shell pattern were now analysed. It was found that the relative importance of the key factors depends on the subset combinations like alkali metal- alkaline earth, transition metal–transition metal etc. It was also found that if the difference in the cohesive energies between the two types of atoms is very small, the nanoclusters constitute a random mix of both the metals, and if the difference in the cohesive energies is very large, the atoms get segregated into a structure having two faces with one face of A atoms and another face of B atoms called the Janus structure named after two-faced Greek God.

Thus the attempt to connect ML with nanoscience was successful in tracing the mixing patterns of metal atoms in nanoclusters and formed a basis for the design map, which can help select the pairs of metals for nanocluster alloys. This design map developed by the scientists will be tested out in the nano laboratories at the Moscow State University and also at the S.N Bose Centre.

Another domain of study for the S.N Bose team has been the heterogeneous structure formed at the junction of two dissimilar semiconductors. They have established that using machine learning, hetero-structure types used in hetero-junctions of two semiconductors which are at the heart of devices like LEDs, solar cells and photovoltaic devices, can be predicted fairly accurately.

The ML model designed by the S.N Bose team predicted 872 unknown semiconductor hetero-structures of type 2 where the electrons and holes align themselves in A semiconductor and B semiconductor, respectively, giving rise to a desirable hetero-structure for semiconductor gadgets.

S.N Bose Centre has used machine learning to search for cheaper substitutes of naturally occurring rare earth material. Rare earth compounds with permanent magnetic properties are used in loudspeakers and computer hard drives. Of these, 17 elements of the periodic table like Neodymium, Lanthanum and so on are found sparsely on the earth’s crust, and their supply is monopolised by the countries where their mines happen to be located. By painstakingly creating a database of rare earth compounds and their attributes and then constructing a machine-learning model, they have predicted a list of potential candidates for permanent magnets whose cost will be less than $100 per Kg.

This work carried out through the ‘National Supercomputing Mission’ has added a whole new drive to humankind’s quest for new material.

18-May-2021: Machine learning helps pick out stars in a crowd

Indian Astronomers have developed a new method based on Machine Learning that can identify cluster stars-- assembly of stars physically related through common origin, with much greater certainty. The method can be used on clusters of all ages, distances, and densities. The method has been used to identify hundreds of additional stars for six different clusters up to 18000 light-years away and uncover peculiar stars.

Studying stars and how they evolve is the cornerstone of astronomy. But understanding them is difficult since they are observed at different ages. A star cluster is, therefore, a great place to study stars. All stars in a star cluster have approximately the same age and chemistry, so any differences seen can be attributed to the peculiarities in individual stars with certainty. As the clusters are part of the Milky Way, there are many stars between us and the cluster, so it isn’t easy to identify and select the stars of a particular cluster.

A team of Astronomers from Indian Institute of Astrophysics (IIA), an autonomous institute of the Department of Science & Technology, Government of India used European Space Agency (ESA)’s recently released Gaia Early Data Release 3 (EDR3) which gives very accurate information about the brightness, parallax, and proper motion of more than a billion stars with an accuracy of 1 milli-arc-second (equivalent to seeing a person standing on the moon) to pick out the stars that are cluster members.

IIA team identified the crucial measurements for this task and understood the complex relationship between these parameters, using a machine learning technique called Probabilistic Random Forest. This uses a combination of parallax, proper motion, temperature, brightness and other parameters to classify each star as a cluster member or a non-member. The IIA team trained their algorithm using the most likely members from a model called the Gaussian Mixture Model, which can identify clumps of co-moving stars. The Probabilistic Random Forest algorithm then learns how to identify a typical cluster member star and efficiently takes out stars that share only similar proper motions or only similar velocities as the cluster itself. They used 10 parameters to identify members, after performing a trade study of all available parameters in the catalogue.

IIA team used the catalogue of members to identify the hottest stars in the six clusters using ultraviolet images from Ultra-Violet Imaging Telescope (UVIT) on the Indian space observatory ‘AstroSat’. This work has been published in the scientific journal ‘Monthly Notices of the Royal Astronomical Society’. Their work has already resulted in discovering hot subdwarf-B type stars (compact stars that are very rare) in open cluster King 2. A research paper on the same has been accepted for publication in the ‘Journal of Astrophysics and Astronomy’. The tool helped confirm that these stars are indeed part of the cluster, though showing unexpected properties.

The newly developed method can now identify cluster stars with much greater certainty and pinpoint individual stars that behave differently from their siblings. The team will apply the algorithms to more clusters in the future.

“Manual identification of stars belonging to a star-cluster is a daunting task owing to an armload of data to be analyzed. The new Artificial Intelligence based algorithm is very promising in automating and greatly speeding this process and may also find uses in other areas of analysis of patterns in biology and materials science,” said Prof Ashutosh Sharma, Secretary, DST.

12-Feb-2021: PSA Professor K Vijay Raghavan calls for unlocking entrepreneurship in design, training youth in machine learning

Principal Scientific Adviser to the Government of India Professor K Vijay Raghavan under scored the need for breaking open entrepreneurship in design across sectors to scale up technologies and boost local manufacturing for sustainable, inclusive development at his tech talk on ‘Reboot, Reinvent & Resilience – Road ahead’ organised on the occasion of the celebration of the 34th TIFAC Foundation Day.

“For scaling up exponentially in focused areas, manufacturing should be distributed with design at the core of it. Prototyping and manufacturing of products can be done locally by entrepreneurs. Design companies anchors and links to Indian academia should flourish so that they have the confidence to make products of any kind anywhere in the world. We should move to a stage where we can make the products, design them and export the design competing globally,” Professor Vijay Raghavan stressed.

He also said that for distribution of knowledge as the basis of power, mathematics, statistics, and computer science learning should be brought to scale and machine learning based decisions which are consequence of those kinds of learning taken to our population in general. Research needs to be amplified to places like schools, colleges, and universities where 90% of our students go. “National Research Foundation (NRF) announced in some details in the budget this time could help training in research in general and science and technology in particular reach number of people,” Prof. Vijay Raghavan added.

He called for the partnership of the entire scientific community in the road ahead for rebooting, reinventing & resilience to strengthen the innovation and research ecosystem of the country.

TIFAC is an autonomous organisation under DST which carries out technology foresight exercise, facilitates and supports technology development, prepares technology linked business opportunity reports and implements mission-mode programmes.

8-Feb-2021: Integrating concepts at the intersection of algebra & geometry could provide better machine learning algorithms

Scientists may soon develop robust algorithms that can provide more efficient machine learning applications by focusing on concepts that lie at the intersection of algebra and geometry.

Hariharan Narayanan, Assistant Professor, Tata Institute of Fundamental Research Mumbai, a recipient of this year’s Swarna Jayanti fellowship instituted by the Department of Science & Technology, Govt. of India, wishes to create machine learning algorithms that can learn from observations and make improved predictions based on mathematical objects known as manifolds and Lie groups. This can lead to improved modelling of data arising from certain sources, such as visual observations.

Machine learning can be broadly defined as a discipline whose goal is to enable a computer to make inferences from observed data about future observations. There are two directions in which progress is crucial to make progress in machine learning. The first is making inferences from very few observations. The second is dealing with complex data, which has come to prominence through recent applications in vision, imaging like Cryo-electron Microscope and the World Wide Web.

The use of manifolds and Lie groups can help addressing both of these issues and may lead to algorithms that make better predictions in real-life applications.