ICS Video Player

Faculty Mini-Talks - Verma Rakesh's Computer Science Faculty Mini Talks 2022

Transcript ×

Auto highlight

Off

Font-size

00:03	Hello, I am professor rakesh verma today I'm going to give you an

00:11	of the readers lab research. So a stands for um the reasoning and

00:22	analytics for security laboratory and a cool about the University of Houston is that

00:31	are a Department of Homeland Security, Security Agency Certified Center of Academic Excellence

00:40	cyber defense research as well as which is a rare honor granted to

00:46	about 50 or so universities and colleges the US. Some of the recent

00:53	of readers are listed here, several students, master student and a couple

01:00	bachelor's students also. Uh these are of the current members of the readers

01:07	and all the members are listed here on this slide and the readers lab

01:17	built up over the years a lot expertise in algorithms, design and

01:24	cybersecurity, natural language processing and symbolic and logical reasoning. And I will

01:35	a few of our current projects as of the kind of work that we

01:40	doing. So. The first one is on automatic deception detection. The

01:47	is the natural language processing work And I will also talk a little

01:52	about data quality and augmentation and analitico cybersecurity project that we're working on.

02:02	we started with phishing emails and we both texting analysis as well as link

02:10	and so the link analysis was published 2017 and the text analysis in

02:17	And then from there on, we started looking at fishing website detection automatically

02:24	a phishing website. And most recently have worked on a broader class of

02:32	attacks such as fake news, spear um and misinformation, disinformation and so

02:42	. And the latest work on automatic detection appeared in ACM kord sp conference

02:49	this year and the natural language processing , we started as consumers of basic

02:59	language processing operations such as part of tagging, named entity tagging and so

03:05	. So we our very first project on authorship detection When we looked at

03:13	set of articles and books by which supposed to have been authored by Daniel

03:20	, an 18th century British author. then a couple of professors from Oxford

03:25	removed several of his works. About of his works from the list of

03:30	attributed to Daniel Defoe and thus began project on whether those works were de

03:36	correctly or not. So whether they authored by Daniel defoe or not.

03:40	from authorship detection, we moved on automatic summarization given a set of documents

03:46	a single document, construct a summary of the information. The useful information

03:52	the document. We also looked at answering and we have systems for both

03:58	and question answering in question answering. are given a set of documents and

04:02	list of questions and then you're supposed answer the questions based on the information

04:08	the documents. We started looking at scores in the competitions organized by National

04:17	of Standards and Technology and I. . D. As it's called?

04:21	we found that we are lagging behind human experts by almost a point on

04:27	question answering task. And so we looking at why this was the

04:32	And we found out that the fundamental operations such as as part of speech

04:37	and named entity recognition etcetera were not correctly. And that's when we moved

04:43	the producer side and we did some on idioms detection and location detection and

04:50	and colorations are special phrases they have meanings when two words two or more

04:56	come together the meaning changes in the things like high school for example.

05:01	so we also received an award for work best paper award for our work

05:07	ka location detection at the cycling 2016 . So that's a little bit about

05:12	natural language processing Workbench and the motivation the work bench is somewhat obvious because

05:19	have huge amounts of unstructured data on World Wide Web moving on to our

05:25	on data collection quality and augmentation. even though there is lots of data

05:32	learning models require a lot of data training. And also data augmentation is

05:40	for determining whether the machine learning models robust or not. And so we

05:46	been working on several projects in this which are organized around quality and augmentation

05:55	the last example project that I will is adapting data science for security and

06:01	may say why do we need to data science for security And there are

06:06	reasons for that. Some of them listed on the right and the main

06:10	is of course the active attacker who constantly trying to defeat the machine learning

06:16	. And so we really need to data science, not just apply it

06:21	security. And if you're interested in exploring some of the work that I

06:28	today, you can take a look our books, cybersecurity analytics co authored

06:34	dr David Marshak. The work that are doing in the reader's lab is

06:42	recognition. So I mentioned the best award for provocation detection research. Our

06:48	also got an outstanding paper award from American Educational Research Association and I mentioned

06:56	couple of the awards that went to us students best PhD student award to

07:02	Baki and the Computing Research Association honorable to Bhutan Faridi who did research with

07:11	as an undergraduate at your patch and was recognized with the mentoring award for

07:20	research and if you need more information or if you have questions please feel

07:27	to reach out to me um at office or we are email and I

07:34	given you some links to to take look at and dig deeper into the

07:39	that we are doing. And I be happy to discuss more with

07:43	Thank you for

Previous Next

00 : 01