Tech Briefings

DARVIZ: Visual IDE for Deep Learning

Deep learning is one of the most exciting and active research area in the machine learning community. As deep learning continues to pave way into diverse and widespread applications, many research problem are choosing to adopt deep learning as a solution approach. There is a need for simple intuitive tools to cater the needs of the diverse communities and enable them to implement and prototype deep learning models in a quicker fashion. The existing libraries such as Tensorflow, Theano, Torch, Caffe, Keras etc. requires an initial large learning curve and also provides minimum communication across them. To overcome these challenges, we have created an abstractive tool called DARVIZ which provides an easy, intuitive drag-and-drop interface for deep learning modelling. Using the abstract DARVIZ model, the execution ready source code can be generated in any language and any library of preference, making a developer's life easy. The current version of DARVIZ is made freely available for research community to use: https://darviz.mybluemix.net/

Presenter Information

Shreya Khare is software engineer working in the area of deep learning and automation with IBM Research AI. She is currently working on DARVIZ, a deep learning IDE for implementing, visualizing, and validating deep learning models. Her research interests include Machine Learning, Speech Processing, Computer Vision and Natural Language Processing. She has written many peer reviewed conference papers. She has completed his MS by Research from IIT Madras where she had worked extensively on multimodal multimedia content analysis and its various application.

Anush Sankaran is a researcher with IBM Research AI. His research interests include deep learning, image processing, human cognition, and their applications. He is now primarily leading efforts in two projects - DARVIZ and Machine Learning for Creativity. He completed his Ph.D. degree with the IIIT Delhi in the area of biometrics and machine learning. He was a recipient of the TCS Ph.D. Research Fellowship from 2010 to 2015. He has written many peer-reviewed conferences and journals and also has the Best Poster Awards in the IEEE BTAS 2013 and the IEEE IJCB 2014. Anush received the B.Tech. degree (Gold Medal) in computer science from the Coimbatore Institute of Technology, Coimbatore, India, in 2010.

Senthil Mani is an Senior Technical Staff Member with IBM Research AI, leading the cognitive application support team. He his passionate about solving business, IT and data management problems by developing innovative solutions. Over the past 12 years in Research and Development, he has gained expertise in the following areas of Computer Science and Service Science: Web Services, Text Analysis, Recommender System, Expertise Mining, Business analytics, IT service management, Business process management. Over the years, he has initiated multiple research projects, led the development of software solutions and successfully deployed solutions. He has actively published over 30+ papers at reputed conferences in the areas of Computer Science and is an active Member of the IBM Academy of Technology since 2016. In recent two years, he has focused on bringing in software engineering principles in building cognitive systems specifically focused on deep learning technology. Given the heterogeneous nature of the DL development libraries and lack of standards in the development process, there is a huge learning curve among developers, silos of DL models with minima reuse. DARVIZ help in addressing these issues and much more in the DL development lifecycle.

Type
Tech briefing

Requirement
The audience might want to use a laptop with internet connection to try our tool hands on. No additional softwares would be required.


Understanding Digital accessibility issues and engineering opportunities – A practitioner’s approach

Objective

  • To understand the difference between ‘accessibility and inaccessibility’ in digital world
  • To learn the importance, theories and current trends in accessibility
  • To know the methods and tools to measure non-conformance to accessibility designs
  • To be aware of world wide web consortium and government stand to create a more digital accessible society
  • To demonstrate the open source tools that address the engineering solution

Motivation

This decade is a digital era wherein about 3.5 Billion people (50% of the world population) are using internet through desktop and mobile devices. Though these percentages do not directly indicate the details within each country, the fact remains that China and India are the number 1 and 2 countries among the world for digital acceptance. The young ‘millennial’ users spend about 180 minutes every day on the internet as per reports. It is also reported that ‘English’ is ranked the number 1 language on the internet. The main activities on the internet includes working, researching, emailing, social networking, online searching for information and knowledge, online shopping/ trading, online transactions and gaming.

There are about 4.9 billion unique mobile numbers in 2017 and has bought in the world to hands of the digital user. The digital disruption has shaken the industry as a whole wherein many banks have led to closure, and many of the large retail chains that ruled the world have shifted to Amazon’s and Flipkarts of the world. The list of industries that have benefitted with digitization is endless. Even in education system, there is more acceptance of publishing content online in various forums.

In the year 1991, the first website was created and it took 2 years to reach 100 websites. Today these numbers have reached over 1 billion in 2014, and continues to grow. The rate of digital data growth has also been steadily increasing, and it is expected to add 1.7 Mbytes of information every second by 2020. The estimated digital data by 2020 is in the range of 44 trillion gigabytes.

The big question that needs to addressed is whether these 3.5 billion users (or potential 6 billion users by 2020) are physically apt for accessing the digital world.

There are broadly four types of common disability that the focus is required – visually challenged, hearing challenges, cognitively challenge and physically challenges of limbs. It has been reported that 1 in every 100 have some degree of challenge. Hence, it is very important to address the question on ‘How accessible are the websites’.

Target Audience

  • All digital end-users who use www, read online documents, publish papers through online submission or emails
  • Developers who design applications on mobile and online channels
  • Testing professional
  • Business & Industry leaders

Outline

Interactive Tutorial outline
  • Problem statement, overview and expectation from the audiences
  • Call out on key take-away from the tutorial
  • Disability facts and definition
  • Current problems and challenges with accessibility – from physical to digital era
  • Laws and Acts (regulation) on accessibility
  • Principles of Web accessibility
  • Guidelines for designing better websites
  • Open source tools for Accessibility – Demo, Pros and Con
  • Detailed analysis of a sample web-page: Issues and recommendations
  • Checklist & Best practices for software engineers & QA professionals, and people developing websites
  • Q & A

About the Speaker

Dr. Kiran Marri is an engineer, consultant, educator and an avid researcher with 18 years of software and systems experience and 4 years of research experience. He is currently a Vice President and Head of solutions in Engineering & Testing business unit at CSS Corp Limited, Bangalore. His role is to develop innovative solutions, create sustainable business models for growth, and launch new services in the practices.

Kiran received a B.E degree from University of Madras, a M.S by research and Ph.D. from IIT Madras. His current research interest is to apply machine Learning techniques and optimization algorithms to solve complex engineering & software problems, build solutions for early validation strategies, develop user experience measurement tools and software metrics analysis. He has published over 70 papers at various international and national conferences in Biomedical engineering, Software testing, Signal Processing and Machine learning algorithms. He has won 4 Best paper awards in 2004, 2011, 2014 and 2015 at International conferences. Dr. Kiran has conducted several workshops in India and abroad, and can be reached at Kiran.Marri@csscorp.com

Type
Tech briefing


Architecture, Method and Tool Support for Automated Regulatory Compliance

In spite of the proliferation of the business process and data compliance checking approaches, in practice, regulatory compliance management still demands considerable manual intervention. Previous research in the field of compliance has established that the manual specification/tagging of the regulations not only fails to ensure their proper coverage but also negatively affects the turnaround time both in proving and maintaining the compliance. To overcome these challenges, this tech-briefing presents an architecture that supports semi-automated checking of regulations in a non-intrusive way. In particular, we describe a method along with supporting architecture and tool support to transform regulatory legal text present in natural english language to a more formal specification via a series of steps using machine learning, domain model creation and model authoring using SBVR Structured English (SE). The key benefit of our approach is the direct involvement of the domain experts to specify regulations using SE, which is close to English, rather than a formal specification language. We use OMG’s SBVR metamodel as our intermediate model representation which the domain expert is oblivious of, but required internally by the underlying transformation toolsets. We also describe how data required for compliance checking and stored in silos throughout the enterprise can be integrated using an intermediate conceptual model that is generated from the SBVR model and mapped using an in-house enterprise data integration (EDI) tool. We substantiate the approach using examples from industry regulations in banking and financial services domain.

Presenter Information

Dr. Suman Roychoudhury is a Senior Scientist at Tata Consultancy Services Research (TCSR). His research interests include model-driven software engineering, domain-specific languages and program / model transformation systems. He is currently leading regulatory compliance research in TCSR and his recent work in software language engineering for compliance specification has been recognized in acedemia as well as industry.

Dr. Sagar Sunkle is a Senior Scientist at at Tata Consultancy Services Research (TCSR). His current research interests include using natural language processing and machine learning techniques for information extraction and legal text comprehension for a more rigorous formal compliance checking in industry setting. His recent work on high profile information extraction project has won prestigious awards and he was nominated by TCS for the INAE young engineer award.

Deepali Kholkar is a Senior Scientist at TCS Research. Her research interests are applying model-based techniques and logic programming to complex business problem scenarios, currently formal regulatory compliance checking. Her prior work includes development of a business process model-based test framework with formal techniques for model-checking and data generation in order to optimize testing of large processes in an industry setting.

Vinay Kulkarni is Chief Scientist at Tata Consultancy Services Research (TCSR). His research interests include model-driven software engineering, self-adaptive systems, and enterprise modeling. His work in model-driven software engineering has led to a toolset that has been used to deliver several large business-critical systems over the past 15 years. Much of this work has found a way into OMG standards. Vinay also serves as Visiting Professor at Middlesex University London.

Type
Tech briefing


MC/DC Testing – A Cost Effective White Box Testing Technique

MC/DC (Modified Condition/Decision Coverage) testing is a method of ensuring adequate testing for safety-critical software. MC/DC is a white box testing criterion to measure source code coverage. MC/DC combines basic condition coverage (C), branch coverage (DC), plus one additional condition (M), i.e. every condition must independently affect the decision’s output. MC/DC was designed to take the advantages of Multiple Condition testing when retaining the linear growth of the test cases. MC/DC testing should satisfy the following properties (i) each and every condition in a decision statement affects the outcome of the predicate, (ii) every point of entry and exit in the program has been invoked at least once (iii) every condition in a decision in the program has taken on all possible outcomes at least once and (iv) each condition has been shown to independently affect the decision’s outcome. MC/DC is subsumed by compound conditions and subsumes all other criteria. This coverage maintains a good balance between thoroughness and test size, and therefore widely used. The MC/DC criterion is much stronger than the condition/decision coverage criterion, but the number of test cases to achieve the MC/DC criterion still varies linearly with the number of conditions in the decisions. It is a much more complete coverage than condition/decision coverage, but at the same time it is not terribly costly in terms of number of test cases.

It has become a mandatory requirement for testing of avionics software as per the DO-178B standard. DO-178B standard is used to assure safety of avionics software. For level A software (that is, software whose anomalous behavior could have catastrophic consequences), DO-178B requires that testing achieves modified condition/decision coverage (MC/DC) of the software structure. MC/DC is a structural coverage measure consisting of four criteria mostly concerned with exercising Boolean logic. The MC/DC criteria were developed to provide many of the benefits of exhaustive testing of Boolean expressions without requiring exhaustive testing. MC/DC percentage can be increased by using preprocessing techniques such as code transformation techniques.

Presenter Information

Durga Prasad Mohapatra received his Ph. D. from Indian Institute of Technology Kharagpur and M. E. from Regional Engineering College (now NIT), Rourkela, in Computer Science and Engineering. He joined the faculty of the Department of Computer Science and Engineering at the National Institute of Technology, Rourkela in 1996, where he is now Associate Professor & Head. His research interests include software engineering, real-time systems, discrete mathematics and distributed computing. He has published more than hundred research papers in these fields in various international Journals and conferences. He has guided more than 15 Ph. D. Theses in these areas. Dr. Mohapatra has been teaching software engineering to UG and PG students at NIT Rourkela for the past twenty years. He has received Young Scientist Award for the year 2006 by Orissa Bigyan Academy. He has also received Prof. K. Arumugam National Award and Maharashtra State National Award for outstanding research work in Software Engineering for the years 2009 and 2010 respectively by Indian Society for Technical Education (ISTE), NewDelhi. He has received three research projects amounting Rs. 36 Lakhs from DST and UGC, Govt. of India. Dr. Mohapatra has co-authored the book Elements of Discrete Mathematics: A computer Oriented Approach published by Mc-Graw Hill. Currently, he is a member of CSI, Fellow of Institution of Engineers (I) and Sr. member of IEEE.

Prateeva Mahali received the degree in Master of Technology in Software Engineering from KIIT University, Bhubaneswar, Odisha, India, in 2014. Now, she is continuing her Ph. D in National Institute of Technology, Rourkela. Her research interests include Software Engineering, Software Testing, Object-Oriented Systems and Data Mining. She can be reached at prateevamahali@gmail.com.

Swadhin Kumar Barisal received his M. Tech. degree in Computer Science and Engineering from IIT Kharagpur, West Bengal, India in the year 2011. He is working as Assistant Professor in the department of Computer science and engineering at Institute of technical education and Research of Siksha ‘O’ Anusandhan University, Odisha. Now he is continuing his Ph.D. in computer science department at NIT Rourkela as a full time sponsored candidate. His Research interests are in Software Engineering MachineLearning, Object-Oriented System, and Real-Ttime systems. He can be reached at swadhinbarisal@gmail.com.

Kulamala Vinod Kumar received his M.Tech degree in Computer science from University of Hyderabad, Hyderabad, Telangana, India, in 2008. He served as Assistant Professor in Department of Computer Science and Engineering, Siksha 'O' Anusandhan Deemed University, Bhubaneswar, Odisha, India, for the past 8 years. Now, he is continuing his PhD in National Institute of Technology, Rourkela. His research interests include Software Engineering, Machine Learning, Object-Oriented System, and Data Mining. He can be reached at kvinod2208@gmail.com.

Type
Tech briefing