September 25, 2014 | Written by: IBM Research Editorial Staff
Share this post:
New IBMer begins work to make Watson work smarter, having made HBO’s Silicon Valley look smarter
“We need to talk.” Generally not something you want to hear from your PhD advisor. But when it was followed by, “Have you heard of HBO?”, then-Stanford student, now-IBM Watson engineer Vinith Misra was intrigued (and as a film enthusiast, a little amused). His advisor, Dr. Tsachy Weissman, had been contacted by the technical advisor for HBO’s Silicon Valley in 2013 to help the show’s creators develop fictionalized compression algorithms, and he wanted to bring Vinith on board.
The series, about to begin its second season, follows the story of Richard Hendriks, a brilliant, young programmer and founder of Silicon Valley startup Pied Piper. He and his colleagues face a race against time – in the form of competition from a bigger tech company, reverse-engineering their work – to land funding from venture capitalists.
“The most memorable algorithm I’ve heard has to be the solution to the countably-infinite prisoners-with-hats problem. You really have to admire the ludicrousness of it (and the insanity of its claim).”
“I’m a big fan of the director Bong Joon-Ho, and his film Memories of Murder is probably my favorite. There is a new layer every time you watch it.”
For Vinith, who studied and now works in information theory and machine learning, the project held great promise. The challenge: create a lossless compressionalgorithm more powerful and efficient than anything that currently exists – or, rather, to make it look like they had created such an algorithm. They had to come up with something that isn’t currently possible, but isn’t immediately identifiable as such.
Vinith approached the challenge of writing the algorithm like he would any research problem. “You have a few things to keep in mind when formulating a problem. It needs to be impactful to people, it needs the potential to work, and the ideas should be elegant, compelling, and provocative,” he said.
With script in hand, Vinith knew that the algorithms were in essence their own character in the show, so he needed to make sure his work could play the part. He combined elements of lossy algorithms for their visual aesthetics on a whiteboard, with the lossless algorithms the show called for, and even created the Weissman Score — a fictitious compression benchmark that could fool even the biggest fanboys.
Vinith and Weissman think it’s reasonable to imagine radically innovative compression algorithms of this sort emerging in the future – which explains why these algorithms would also survive a cursory glance from highly trained engineers. “If you gave these ideas to a first-year grad student, they could run with them,” Vinith said. It would require a more detailed analysis to uncover the algorithms’ fundamental unfeasibility.
Hard at work finishing his doctoral studies in electrical engineering, Vinith didn’t spend much time on set during filming. Instead, he would conduct what he called “firefighting calls” with cast and crew, adjusting technical dialogue, finding material for the whiteboards and sets, and helping producers and actors deal with unexpected technical elements.
Vinith knew that the finale to the first season would involve a drastic alteration to Pied Piper’s compression algorithm – an alteration that was the product of a mathematical joke from a pivotal scene. Developing and coding even as his team pitches the compression during the final episode, Richard comes up with a groundbreaking development after a humorous debate about how to influence the potential funders to choose their work. This breakthrough, which set the stage for the second season, stemmed from the phrase “middle out.” Vinith even published a 12-page analysis of the scenario that gives rise to the breakthrough, and while it is highly detailed and mathematically rigorous, it is explicit; reader discretion is advised.
Vinith will continue to work on the show, but this month he also began a new chapter with IBM’s Watson Group. “Watson is transitioning from the conceptual to working on real problems…and I’m glad to be a part of it,” he said.
Vinith will work on Watson out of the IBM Research – Almaden lab, where he will be developing algorithms to reason about concepts across a variety of domains, including subjects ranging from food to baseball cards — all in an effort to make Watson more agile.
Regarding the future of the show, Vinith can’t reveal much. He continues to work on Pied Piper’s breakthroughs and is looking forward to seeing the finished product of his work on screen. “I hope people treat the show like science fiction,” he said, “but the good kind.” He says it has been a fantastic and unique experience, and one from which he has gained a greater appreciation for the aesthetic side of what he does. “Algorithms and systems are designed to be used, but the ideas behind them can often be compelling, even to non-technical people.”