“Did the two shoe prints come from the same source? Solving crime with Deep Learning”

PROFESSOR SARGUR N SRIHARI AS INTERVIEWED BY GAUTHAM SANTHOSH AND FAHEEM ZUNJANI

Professor Sargur N Srihari is a SUNY Distinguished Professor in the School of Engineering and Applied Sciences at the University at Buffalo, Buffalo, New York, USA. His work at Center of Excellence for Document Analysis and Recognition(CEDAR) led to the first handwritten address interpretation system in the world, versions of which were deployed by the Internal Revenue Service, US Postal Service, Australia Post and UK Royal Mail. He also visited IIIT Allahabad and delivered a short course on Deep Learning from Aug 3, 2017 to Aug 4, 2017.

Instead of starting off with the more insightful questions, as an ice-breaker we would like to ask you a few short and fun questions and you can answer them in a few words or a sentence.

What is something you miss about the place where you were born?

Ahh the weather!

What website you love to spend your time on?

I’m a news junkie. So, I love The New York Times.

What Machine Learning papers are you most interested in?

Well, I’ve been in this field for many years. There have been many excellent papers. Well, there also have been many books, I should say. Papers are small little contributions. So text books are there. Chris Bishop’s books are excellent.

What do you like to do in your free time?

I like to read.

Fiction or non-fiction?

I like history, so non-fiction. I’m a history buff.

Indian history or anything else?

Well I love Indian history but I also read American history.

What’s more interesting to you — Physics or neuroscience?

Well, Neuroscience, I should say.

Why do you think Deep Learning has become so popular now?

Due to the advancements and the kind of result that have been obtained which are astounding.

What’s your favourite operating system?

Well I’m an Apple guy, you know. So I’m quite happy with the it.

What website do you refer to stay updated about your field?

Surprisingly, it is Facebook. There are a lot of good professional contacts. Other than the family, I subscribe to a lot of professional friends who are active at posting on Facebook.

So they probably share their own blog posts..

Yeah, yeah! Facebook itself is excellent. The leaders in A.I., their favourite place to post is on Facebook. Some of my friends are on Facebook and I follow them everyday.

What is your programming language of choice for machine learning?

It used to be MATLAB till recently but my students convinced me to switch to Python. So it’s Python now.

Who’s your favourite professor?

Now I have to go back to my student days. My favourite professors are from India, from Bangalore, in IISc and the national college where I studied.

What’s your favourite comfort food?

You know, I’m getting old, so I have to be careful about what I eat. I enjoy a glass of wine.

So, Italian wines?

Yeah. I like European red wine. And Californian red wine too.

Who is the one person you admire?

Well there are many historical figures. Perhaps, one I admire the most is Abraham Lincoln.

What is an open-source library that you really like?

Today in Machine Learning, things like ‘Tensorflow’ and all are good open source libraries.
* What’s the best advice you have ever received?*

Best advice that I have ever received? Well, there are so many things. I’ve received good advice from my mother on what are the ingredients that go into cooking, you know.

So, you like cooking?

Yeah. I enjoy cooking. The advices are general rules, like you know, put equal amounts of salt and spices and for sourness, there are tamarinds and other things. Yeah, I put up a few dishes.

Do you think A.I. will ever take over humans?

The science looks like it is going to happen. Fortunately, not during my lifetime, but I think it’s going to be 50 years before it happens.

If you had to start Machine Learning from scratch right now, what resources would you use?

Machine Learning has got the theory algorithms implementations and all that specs. But I think there is a tremendous amount to be learned in theory and algorithms. What I’m creating on the web are meant to be aids for my students who want to get into it. I’m not sure this is completely it, but I highly recommend my own slides, other than the various books on Machine Learning. Each of them covers some aspect very well. None of them is absolutely complete. There are books by Chris Bishop, Daphne Koller and by Goodfellow. These are excellent resources.

Sir, you’ve studied in India and the U.S. What are the differences that you found in both education systems?

I was lucky to have had a wonderful education in India, in some way, even better than that in the U.S. In Bangalore, I went to a place called The National College and the Indian Institute of Science, which during the days, was an absolutely amazing place. It set the background for what I’ve been doing for 50 years. That was probably the best education for me.

So, did you find the Indian Education system to be better than the American?

Yes, the dedication of teachers in India is quite extraordinary. The national college was a renowned institution and the teachers were extremely committed to students.They put a lot of efforts into it. The environment in IISc was of the highest calibre. I went to good places like The Ohio State University, but the extraordinary education I received was here in India.

How did you come up with the idea of automating the US Postal Service using handwriting recognition?

You know, I was very early into this thing — Machine Learning and Pattern recognition. And the U.S. Postal Service had decided that they needed more automation. So they wanted groups to do some research. In the beginning, it was primarily to find out what technologies were available to use. I started helping them like that. I decided that I could help them by developing some technology. Such technology had kind of disappeared from America, it was largely there in Japan and Germany. But America didn’t have anything in the area of recognizing text, OCR, handwriting and so on. So although it originally started in America, it kind of disappeared. So I was kind of instrumental in bringing it back to America with my research and others followed suit.

What are some of the things that you’re most proud of?

Our work with the U.S. Postal Service is one the things that I’m very happy about. And the other one is the the work that I’ve done in the area of forensics, to try and say whether two things came from the same source or not. Even when I worked in the Postal Service, to compare handwritings and to help the justice system. I served in the United States National Panel on what kind of work should be done so that the forensic system is fairer towards people who are accused and things like that. So I’m very happy that that work has tremendously impacted my life.

Sir, what difference do you find in the research that is done in the academia and the research that is done in the industry?

It is very complimentary. In industry, there are tremendous resources like computing power and data, a lot of it is private data that they cannot give out to anybody, like Facebook and all that, you cannot give out private information. But if you’re working in Facebook, you get to access all of the data. Same with Google, you want to use all of that to develop good technology. But they are all oriented towards specific needs that they fulfil. I’ve done that in developing things for the U.S. Postal Service which is also directed towards solving a problem that they are facing. But in academia I’ve got the freedom to pursue all kinds of topics at the same time so you’re not kind of focussed on one specific problem. You can go wherever you want. We pursue things that are, you know, open ended. But also one disadvantage is that you don’t have the resources the industries have. But this kind of open-endedness can eventually help industry. They never thought that something like this would be promising. So in a sense, it’s complimentary. Today there is a tremendous interaction between academia and industries.

What are some of the projects that you are working on?

I am working all the time just to keep up in this area — Artificial Intelligence and Deep learning, because the whole field is exploding. I still am working on forensic type of problems — how do I apply Deep Learning in problems where you are comparing things or for two inputs, you’re asking the question — Did it come from the same source? Did these two handwritings come from the same source? Did these two shoe prints come from the same source? Did these two fingerprints come from the same source? These are all special types of problems. And I’m applying those things. But I’m spending 90% of my time just to keep up with what’s going on today ,what’s being generated. And how can we make sense of all of this? So that is where most of my work today lies — the algorithms that are being developed in machine and deep learning.

So, how is applying Deep Learning in Forensics different from its other applications?

Forensic has specific problems. In the crime scene, there might be a particular evidence there and then there will be a suspect who has a certain evidence associated with him. So the kind of questions we are asking is — could these two have come from the same source? So that’s not the kind of problem that’s mostly encountered in normal objective definition type problems. Here you are given two inputs, rather than one input and usually in your typical problems, you’re given more than 48 hours to say what it is. But here you’re given 2 inputs and you ask — ’Did this person commit that crime?’

What advice would you give to an undergraduate students who is interested in pursuing research?

I would say that — seriously consider getting into research. A lot of youngsters today are not interested. They want to move on to industry. There are exciting things to be done here. It’s a bit of a dilemma.If you’re in academia, you’re establishing yourself for a life-long career. In industry, the opportunities might be less, but academia allows you this freedom throughout the life. So that’s the attraction of being in academia. But I can see the attraction of industry as well.
GS: So that’s it.

Would you like to say anything to the students reading this?

I have been tremendously impressed by the students here at IIIT Allahabad, to my surprise — how well informed the students are, and the great questions that I’m getting from you in the class that I’ve been teaching in, and the interest and enthusiasm shown by you all is great. Also, I’m going away with a very positive impression of the student body as well as the faculty. You all seem to be working on very state of the art things. I haven’t seen all of allahabad — a few places that I saw seemed like they weren’t very well developed, like when I was coming from the airport, I saw the road and I was like ‘This is really, like, from the ancient age’. I still have to see the city and all. But to see the students and faculty here in this background which seems somewhat under-developed is kind of surprising. I hope to see more parts of Allahabad tomorrow. So that’s one of the surprising things that I’ve seen.
But I’m going away with a very positive impression of IIIT, Allahabad. I hope to go with a positive impression of Allahabad as well.

(Special thanks to Arpit Misraw for transcribing the audio recording of the interview)

By Gautham Santhosh

Source: Deep Learning on Medium