Deep Learning is a computational model for learning goodrepresentations of data. The computationalmodel is typicallya neural network. Recently, deep learning has had tremendoussuccess in performing tasks such as image and speechrecognition, and carry the potential of impacting every areaof science.

This article aims tomake deep learning accessible to a broaderaudience including technically-minded people who are noncodersor mathematically unsophisticated.

The phrase artificial intelligence has become common in ourcurrent day discourse. Fuelled by successes in machine learning,and applications interacting with us in speech and naturallanguage, many commentators have made a leap of faiththat behind these successes is a thinking machine, and thishas even stoked fears of machines overcoming humankind.In this two-part article, we look at how close we are to theoriginal quest for creating “machines who think”. In the firstpart, we trace the evolution ofmechanical computers and alsothe notion of the mind up to the era before digital computersappeared on the horizon.

In the first week of year 2020 we got news that AI now outperformsdoctors in detecting breast cancer. This is in line witha continuous stream of news coming from the world of diagnosis,and has lent credence to the sentiment that AI is poisedto overcome humankind. However some perceptive observershave commented that recent advances are largely due to themassive increase in both availability of data and computingpower. Moreover, it is only a narrow task of classificationthat has led the news blitz. Classification can be thought ofas a stimulus-response process. Human intelligence is muchmore broad. In particular humans often display a stimulusdeliberation-response cycle. There is much that goes on in the

“thinking” phase that was the original aim of AI before thedata and speed started dominating applications. The secondof the two part article on AI traces the evolution in the fieldsince the Dartmouth conference, and takes stock of where weare on the road to thinking machines.

Machine learning has been applied to various fields and is envisagedas the technology of the future. We discuss here, theapplications of machine learning methods to represent potentialenergy surfaces – an important aspect of chemical dynamics.We illustrate the process of machine learning usingsimple examples, and demonstrate how it can be extended tocomplicated problems.

The problem of finding the capacitance of a thin conductingdisc has continued to engage the attention of the scientistsfor more than 150 years. Not that the problem has not beensolved. Rather it is the novelty of the method that is the motivatingfactor. In the literature there are only a few methodsfor obtaining the capacitance of a thin conducting disc. Themethods use elaborate mathematical schemes. But in suchmethods no attention has been paid to closely examine thecharge distribution on the disc itself, and try to see if we canuse this charge distribution to build up a conducting surfaceon which charge distribution can be found by some suitablemathematical method such as Conformal Mapping. In ourmethod we adopt this scheme of solution.

“Earth provides enough to satisfy every man’s needs, but notevery man’s greed” – Mahatma Gandhi

The vast number of species prevailing on planet Earth is theresult of evolutionary processes that have been operating sincelife originated about 3.5 billion years ago. As new speciesevolved, a small number of species that became misfits inthe prevailing environment became extinct (background extinction).However, the rapid increase in human populationand humanity’s greed for luxurious living have resulted inmarked environmental degradation, particularly in the recentdecades, increasing species extinction hundred or eventhousand-fold compared to background extinctions, thus precipitatingthe ‘sixth mass extinction’ crisis. Unlike the pastfive mass extinctions that were due to natural catastrophes,the sixth mass extinction would be exclusively the result ofhuman activities. Habitat loss and its degradation, overexploitationof bioresources and climate change have been themain drivers of the sixth mass extinction crisis. Amongsthuman-induced environmental changes, climate change is goingto affect humanity more than any other changes. Apartfrom exterminating a large number of both terrestrial andaquatic species, these changes bring down crop productivityand quality substantially, thus seriously compromising ecosystem services essential for human welfare. Mitigating humaninducedenvironmental changes has become one of the highestpriorities for the humanity to sustain biodiversity and humanwelfare.

Continuing to explore the intriguing world of the Indian paperwasp Ropalidia marginata for one last time, here we willfocus on the function of fighting behaviour in two additionalcontexts (i) the hyper-aggression of the potential queen duringqueen succession and (ii) during encounters with nonnestmatewasps. We will see again that the function of fightingis different in different contexts. We have already seentwo different functions of fighting in two different contexts—to decide who will be the queen and who will be the workerin the context of founding new nests, and to regulate foragingin mature colonies by conveying colony hunger levels toforagers. Here we will see that the function of the potentialqueen’s hyper-aggression is to boost her own ovarian developmentand the function of aggression towards non-nestmates isto keep them away, and if necessary, to kill! As before, ourprimary focus will be on how to design simple experimentsthat will help answer a direct question, while minimising theneed for expensive equipment or other facilities.

Diophantine geometry is a fascinating interplay between algebra,geometry and number theory. The author was introducedto it in a blackboard-less one-on-one talk given inthe aerodrome at IIT Kanpur during a thunderstorm! Thespeaker was Professor V. Srinivas of the TIFR. At that pointin time, the latter was a graduate student at the University ofChicago and happened to be in Kanpur.