Pradipto Das - SUNY Buffalo, CSE Dept. Research

Publications

PhD Thesis

(+) Natural Language Summarization of Text and Videos using Topic Models - Pradipto Das, PhD Thesis, Computer Science and Engineering Dept., SUNY Buffalo, 2014 [UBCSE Technical Report: 2014-02] | [slides]

6) A few sentences are mightier than a thousand frames!

(+) A Thousand Frames in Just a Few Words: Lingual Description of Videos through Latent Topics and Sparse Object Stitching - Pradipto Das*, Chenliang Xu*, Richard F. Doell and Jason J. Corso, IEEE Computer Vision and Pattern Recognition (CVPR) Conference, Portland, Oregon, 2013. (Full paper + spotlight) [PDF] | [Data]

What is this about?

- Does predicting more keywords really improve relevancy of BoW summaries generated from videos?
- It turns out that selecting a few short sentences from the training set using topical weights of a few predicted keywords improves summary relevance dramatically vis-a-vis merely pulling in more keywords as measured by the ROUGE-R1 (recall) measure
- Sentence generation using more predicted keywords may not always be the best for summary generation although the best sentences from the training set might be able to automatically tune language models to generate sentences for the context at hand

5a) Summarizing an event through videos and text as well as individual videos through keywords

(+) Translating Related Words to Videos and Back through Latent Topics - Pradipto Das*, Rohini K. Srihari and Jason J. Corso, 6th ACM Web Search and Data Mining (WSDM) Conference, Rome, Italy, 2013. (Oral) [PDF]

Pradipto Das, Rohini K. Srihari, Jason J. Corso
WSDM '13 Proceedings of the sixth ACM international conference on Web search and data mining, 2013

5b) (Related) TRECVID 2012 GENIE: multimedia event detection and recounting - Amitha Perera, Sangmin Oh, Megha Pandey, Tianyang Ma, Anthony Hoogs [Kitware Inc.], Arash Vahdat, Kevin Cannons, Greg Mori [Simon Fraser University], Scott McCloskey, Ben Miller, Sharath Venkatesha, Pedro Davalos [Honeywell ACS Labs], Pradipto Das*, Chenliang Xu*, Jason Corso, Rohini Srihari [SUNY Buffalo], Ilseo Kim, You-Chi Cheng, Zhen Huang, Chin-Hui Lee [Georgia Tech], Kevin Tang, Li Fei-Fei, Daphne Koller [Stanford University], In NIST TRECVID Workshop, Gaithersburg, Maryland, 2012 [PDF]

What is this about?

- Model video documents with real and discrete features from videos and discrete features from text
- Find related words through joint topic modeling and label topics with video frames
- Summarize a test video with keywords from the training vocabulary

4) How do humans summarize? Addressing an unified information model of summarization

(+) Global and Local Models for Multidocument Summarization - Pradipto Das* and Rohini K. Srihari, Text Analysis Conference, Gaithersburg, Maryland, 2011. (Oral) [PDF | Prototype concept]

What is this about?

- Address the problem of summarization by catering as best as possible to a unified information model
- Summarize a set of documents capturing different relevant pieces of information on a major event
- Uses fine grained lingusitic analysis as well as exploratory faceted topic analysis.

3) Topic Models on Ubiquitous Document Structure - Documents Tagged from Two Perspectives

(+) Simultaneous Joint and Conditional Modeling of Documents Tagged from Two Perspectives - Pradipto Das*, Rohini K. Srihari and Yun Fu, 20th ACM Conference on Information Knowledge and Management (CIKM), Glasgow, Scotland, 2011. (Oral) [PDF]

Simultaneous joint and conditional modeling of documents tagged from two perspectives

Pradipto Das, Rohini Srihari, Yun Fu
CIKM '11 Proceedings of the 20th ACM international conference on Information and knowledge management, 2011

What is this about?

- Model documents tagged from two different perspectives - a document level tag perspective that is relevant to the document as a whole and a word level tag perspective pertaining to each word in the document
- Attribution of latent topics with word level tags (faceted topics) and labeling latent topics with images in case of multimedia documents
- Discovering the possible correspondence of the words to document level tags.

ERRATA: The typos in the expressions in the lines just above equations (17) and (20) have been corrected in this pdf - pi's have been replaced with beta's in the RHSs. Also the exponents of zeta's and log zeta's in Equs. (4) and (10) have corrected. These are also commented (see "Comments" tab/section) in the ACM DL repository for this paper

2a) Topic Model based Multi-document Summarization

(+) Learning to Summarize Using Coherence - Pradipto Das and Rohini K. Srihari, NIPS Workshop on Applications of Topic Models: Text and Beyond, Whistler, BC, 2009. (Short paper + poster) [PDF]

2b) (Related) Longer version of this paper appeared in TAC09 proceedings. [PDF(slightly modified in here)]

What is this about?

- Building a Hybrid pLSI/LDA style topic model based summarization system
- Select sentences directly as summaries given query words
- Uses coarse grained coherence properties across sentences in text

1) First Contact with Blogs

(+) Discovering Voter Preferences in Blogs using Mixtures of Topic Models - Pradipto Das, Rohini K. Srihari and Smruthi Mukund, AND'09, Barcelona, Spain, 2009 [Noisy Text Data Analytics Workshop in conjunction with ICDAR'09] (Oral) [PDF | PPT as PDF]

Discovering voter preferences in blogs using mixtures of topic models

Pradipto Das, Rohini Srihari, Smruthi Mukund
AND '09 Proceedings of The Third Workshop on Analytics for Noisy Unstructured Text Data, 2009

What is this about?

- Towards which direction would a blogger be oriented? The speaker or the contents of his/her speech? (A very tricky and hard problem - especially if generalized to data in any language)

* means significant contribution among co-authors

Major research events

Summer 2013

Portland, Oregon > CVPR Conference (Note: I am *not* a vision scientist)

Spring 2013

Rome, Italy > WSDM Conference 2013 (Could not attend but slides are uploaded to Slideshare. Comments welcome)

Spring 2012

Golden Valley, MN > Honeywell ACS Labs (RA meeting)

Fall 2011

Glasgow, Scotland > CIKM Conference 2011 (attended by my adviser - Rohini K. Srihari. Slides are uploaded to Slideshare. Comments welcome)
Gaithersburg, MD, USA > Text Analysis Conference 2011

Summer 2010

Intern at Janya Inc. (now ContentSavvy Inc.) Responsible for integrating an extensible topic modeling framework within the text analytics product called SEMANTEX.

Fall 2009

Whistler, BC, Canada > NIPS 2009 Workshop - Applications for Topic Models: Text and Beyond

Gaithersburg, MD, USA > Text Analysis Conference 2009

Professional Services

Program Committee Member > NAACL 2015, ACL 2014

Reviewing

Primary > IEEE-PAMI 2015, EMNLP 2015, ACL 2014, Elsevier-Image and Vision Computing 2014, NAACL 2013
Secondary > IEEE-International Conference on Semantic Computing 2013
Un-acknowledged > KDD 2013, KDD 2012, ACL 2012, AAAI Workshop 2011, KDD 2010

Previously

Previous Research > Visiting Research Fellow at the "Center for Soft Computing Research" at Indian Statistical Institute. Mostly worked on soft computing clustering algorithms.

Adhoc Datasets > Visit the datasets page.

n e x t

email: pdas3 at ["university domain" which is simply "buffalo dot edu"]