Ma­chines at the MALL Learn the Ways of the World

IISc Lab is work­ing to give ma­chines a world­view

The Economic Times - - Disruption: Startups & Tech - J.Vig­nesh@ times­group.com

Ben­galuru: In one corner of the sprawl­ing In­dian In­sti­tute of Science (IISc) cam­pus lies the seem­ingly quiet Ma­chine and Lan­guage Learn­ing (MALL) Lab.

It’s mak­ing a noise in the world of technology, with its grand mis­sion of giv­ing a “world­view” to ma­chines that will en­able them to make wider con­nec­tions be­tween var­i­ous data and draw the nec­es­sary in­fer­ences to make de­ci­sions. Make no mis­take, this is not a ran­dom aca­demic pur­suit. It is of ut­most im­por­tance in an era where au­to­ma­tion is the next big thing. Au­tomat­ing de­ci­sions could save companies huge amounts of money and re­duce man power, which can be in­stead used for more crit­i­cal tasks.

Recog­nis­ing this, global technology gi­ants such as IBM and oth­ers are fund­ing and col­lab­o­rat­ing with the lab.

“We, as hu­mans, ag­gre­gate a lot of knowl­edge about an en­tity and then use the knowl­edge to make de­ci­sions. In­creas­ingly, we are ask­ing ma­chines to make non-triv­ial de­ci­sions on our be­half. But ma­chines do not have the world­view that you and I have,” said Partha Taluk­dar who is lead­ing the team that works on this project.

“The the­sis is that, if you can make that kind of knowl­edge avail­able to these de­ci­sion-mak­ing agents — to chat­bots, ro­bots, per­sonal as­sis­tants — then their per­for­mance is go­ing to sig­nif­i­cantly in­crease,” said the 35-year-old pro­fes­sor.

The team con­sists of PhDs, mas­ter’s stu­dents, re­search as­sis­tants and in­terns, who work on var­i­ous as­pects of this huge task.

The work done at the lab could po­ten­tially help in cre­at­ing knowl­edge­able per­sonal as­sis­tants who can carry out tasks with in­creased ef­fi­ciency, come up with al­go­rithms to ful­fil taskspe­cific needs and build pro­grams that can scrounge through open fo­rums to track feed­back about prod­ucts, among oth­ers. “IBM Re­search is look­ing at task di­rected knowl­edge graph,” said Taluk­dar. The team uses core ar­ti­fi­cial in­tel­li­gence (AI) ar­eas such as nat­u­ral lan­guage pro­cess­ing, ma­chine learn­ing and large-scale data anal­y­sis meth­ods to build ‘knowl­edge graphs’. Knowl­edge graph, a term that Google made fa­mous in 2012, ba­si­cally tries to un­der­stand and learn facts about places, peo­ple and things and how these en­ti­ties are all re­lated. “You can think of that as a graph or a net­work where the nodes are of en­ti­ties of in­ter­est,” said Taluk­dar.

Sharmistha Jat, a PhD stu­dent who does re­la­tion ex­trac­tion at the lab, gave a peep into what they are do­ing. “There are dif­fer­ent re­la­tions, phrases like ‘located in’, ‘prime min­is­ter of ’, etc. We want to un­der­stand how en­ti­ties are con­nected. I look at pat­terns. For ex­am­ple, if there is a state­ment say­ing Naren­dra Modi was born in Gu­jarat. The ma­chine needs to know that Gu­jarat is in In­dia and so Naren­dra Modi was born in In­dia,” she said.

The knowl­edge to ‘teach’ ma­chines is gath­ered mostly through the in­ter­net. The in­ter­net, with its many so­cial net­work­ing sites, blogs and web­sites, is a repos­i­tory of in­for­ma­tion. Though there are cer­tainly labs work­ing on var­i­ous sub­sets of the same prob­lem, the ‘com­pre­hen­sive out­look’ is what makes MALL spe­cial, Taluk­dar said.

“The area that we are work­ing in is unique in the In­dian con­text. There are dif­fer­ent pieces of it that are stud­ied in other groups. But the col­lec­tive view is unique. I do not know any other lab in In­dia and a few glob­ally, that has this com­pre­hen­sive view,” he said.

Though these are still early days for the lab that be­gan in late 2014, it has al­ready seen vary­ing de­grees of suc­cess with its col­lab­o­ra­tions. “There are pro­to­types that have been de­vel­oped. Some of the al­go­rithms that we have de­vel­oped have been im­ple­mented on our part­ner’s data. They have seen dif­fer­ent de­grees of suc­cess,” said Taluk­dar.

Au­tomat­ing de­ci­sions could save cos huge amounts of money and re­duce man power, wh ich can be in­stead used for more crit­i­cal tasks

Newspapers in English

Newspapers from India

© PressReader. All rights reserved.