Skip to main content
main-content

Über dieses Buch

This book presents reports and methods that demonstrate the ease with which cognitive applications can be built using IBM Watson application program interfaces (APIs). It includes application reports from two IBM Watson API-based competitions – Hackathon (24 hours) and a Challenge task (~3 months). It also features a selection of papers presented at I-CARE 2016, the IBM Collaborative Academia Research Exchange event, from the areas of “Theory and Cognitive Computing”, “Data Platforms and Systems,” and “Societal Applications.” IBM has a long tradition of research collaboration with colleagues in academia, and I-CARE is an annual event initiated in 2009 to promote collaborative innovation and learning, and explore new ways of fostering a culture of innovation. I-CARE’s main goal is to “amalgamate” the thought leadership in Indian academia with that in industry, and foster a symbiotic environment for establishing a rich research culture in India.


The 8th edition of I-CARE presents a collection of thought-provoking ideas and novel Indian research projects related to three crucial areas: cognitive computing, systems and platforms that support large-scale data processing and practical systems that are designed for the public good.

Inhaltsverzeichnis

Frontmatter

Introduction

Abstract
IBM has a long-running tradition of research collaboration with our colleagues in the academia. IBM Collaborative Academia Research Exchange (I-CARE), an annual event initiated in 2009, aims to promote collaborative innovation and learning, and explores new ways of fostering a culture of innovation. I-CARE’s main goal is to “amalgamate” the thought leadership in Indian academia with that in the Industry, and foster a symbiotic environment for establishing a rich research culture in India.
Danish Contractor, Aaditya Telang

Research Papers

Frontmatter

Using Trusted Execution Environments to Enable Integrity of Offline Test Taking

Abstract
Today, automated assessment in online courses (MOOCs, SPOCS, etc.) is done in a client–server fashion. The only trusted component is the server as it is run by the test creator. The user accesses the automated assessment module by logging into the server using their mobile device. The server sends questions to the user’s device one by one. For each question, the user types their answer into their device, which is then transmitted to the server for evaluation and scoring. This process requires constant internet connectivity on the user’s device, which may be a difficult requirement in developing nations. For evaluating complex assignments, it also places a significant load on the server. This paper aims to address these issues by providing a secure offline solution for the automated assessment problem. An offline approach performs the evaluation and scoring on the user’s mobile device and we must protect the evaluation and scoring process from being tampered by the user. To achieve this, we divide the assessment module into two parts: (1) an untrusted user-facing GUI component, and (2) a trusted enclave component that handles the evaluation and scoring, and use a hardware-rooted isolated execution technology (ARM TrustZone) to create an isolated enclave for the execution of the enclave component. The execution of the enclave component remains untampered due to the isolation provided by ARM TrustZone. With this approach, we can successfully encourage the adoption of online courses in developing nations with limited internet connectivity and also reduce the load on the server during the assessment.
Rahul Shivu Mahadev, Arvind Seshadri, Sriram Rajamani, Viraj Kumar

ME : Movie Review Exploration Engine

Abstract
To address the needs and aspirations of movie fans living in the world of big data, we propose a Movie review Exploration Engine, ME\(^2\). Different from most of the previous work with paragraph-level or sentence-level summarization methods, we formulate the movie review summarization task as a keyphrase mining problem. We propose a novel framework for exploration and enrichment of movie reviews using keyphrases extracted from news articles and enriching associated \(\langle \)element, role\(\rangle \) pairs. Our user case studies have demonstrated that ME\(^2\) keyphrases [machine selected] are comparable to or even better than online Internet Movie Database (IMDb) and The Movie Database (TMDb) Plot keywords [human selected].
Nikita Jain, Achuth Kandikuntta, Deepak Jannarapu, Nallagatla Manikanta, Tella Tarun Kumar, Dhaval Patel

Convex Model Databases—Solving Real-World OR Problems

Abstract
Many real-world optimization problems deal with uncertain data and several modelling approaches including robust optimization (convex uncertainty sets) and stochastic programming (probabilistic uncertainty sets) are used to handle the uncertainty. These approaches lead to more complex models, however, advances in the field of convex optimization Boyd and Vandenberghe (Convex Optimization, Cambridge University Press, [1]) have made many complex problems quite tractable and applicable in many branches of information processing, including operations research, data science, signal processing, optimal control, and more. Though much research has been undertaken to improve the problem solving mechanisms, relatively little has been done in the area of data representation—design of appropriate databases for storing and querying these models and using these models for data analysis, and this is the focus of our work.
Anushka Chandrababu, Abhilasha Aswal, G. N. Srinivasa Prasanna

Vision System for Medicinal Plant Leaf Acquisition and Analysis

Abstract
Medicinal plant identification is a challenging but very useful task in computer vision (CV). Deep convolutional neural network (CNN) is promisingly used in plant identification as experimentally proved in this paper. It presents a new setup to capture efficiently plant leaves and are used for classification. Secondly, \(l\alpha \beta \) color space is used to improve the performance of CNN in plant species recognition. For this experiment, two different types of datasets are used showing the robustness of our approach.
Shitala Prasad, Pankaj P. Singh

On the Relevance of Very Deep Networks for Diabetic Retinopathy Diagnostics

Abstract
Detection of Diabetic Retinopathy (DR) has been worked on for a long time, but no commercially viable solutions that work for different populations exist yet. In this work, we investigate the performance of Very Deep Networks for the binary classification of fundus images provided by EyePACS as part of Kaggle’s DR detection challenge.
B. Akilesh, Tanya Marwah, Vineeth N Balasubramanian, Kumar Rajamani

EffGenPerm: An Efficient and Fast Generalized Community Detection for Massive Complex Networks

Abstract
The advances in community detection (CD) algorithms resulted in a study of analyzing the massive complex networks for resilience to perturbations. To address this issue, recently, few researchers had proposed CD algorithms by proposing new metrics like permanence and neighborhood connectivity. In this manuscript, we are proposing a new metric called “Effective Pull,” based on that an efficient CD algorithm has been developed, which will identify the underlying communities by maximizing effective permanence of a community by maximizing the effective permanence of each node in that community. As a peripheral output, our proposed algorithm fixes the drawbacks found in the recent advanced CD algorithms. The proposed is evaluated with real-time datasets and its efficiency is found better compared to recent literature.
Mrudula Sarvabhatla, Chandra Sekhar Vorugunti

Hackathon Applications

Frontmatter

DearDiary—An IBM Watson Powered Mental Healthcare Tracker

Abstract
Depression is a common illness worldwide, with an estimated 350 million people affected by it. Barriers to effective care include lack of resources and trained healthcare providers, social stigma associated with mental disorder and inaccurate assessment. Our project, Dear Diary, is aimed at tackling the problem of diagnosis and monitoring of patients suffering from depression. In addition to psychotherapy and medication as treatment, patients can maintain a daily log using the app. The app performs sentiment analysis on the entries made by users using IBM Watson’s® Alchemy API. A weekly report of the user’s emotions is provided to help with self-monitoring of one’s emotional well-being. Doctors can view and manage reports of their patients generated by the app through a web application. This helps doctors with diagnosis and assessment of patients mental health. The app also allows patients to receive feedback from doctors.
Abhilash Kishore, Amisha Agarwal, Anisha Mascarenhas, Arjun Rao

K9, for Things You Care About

Abstract
We love our pets, and when they are separated from us, it is like our whole world comes crashing down on us. We cannot afford to leave any stone unturned in the process of finding them.
Ravi Vats, Amogh Mannekote, Deepak D. Rao

Moksha (An App to Stop Child Labour)

Abstract
Moksha is an app developed on android with IBM Bluemix Visual Recognition API as the back end. It is an app intended to stop child labour by reporting it to the concerned agencies. It was developed during the IBM hackathon held as part of the IBM ICARE event for the year of 2016.
Elson D’Souza, M. S. Eshwar, Aditya Kalyani, Sumit Khaitan

Watson Cognitive Challenge Applications

Frontmatter

CognitiveCam: A Visual Question Answering Application

Abstract
CognitiveCam is a Visual Question Answering application designed for helping the visually impaired. A user can ask any question pointing towards the object, the app will then attempt to provide the relevant answer. Visual Question Answering is a hot-research area and CognitiveCam is an attempt to put the current state-of-the-art research to practise using the IBM Watson Bluemix APIs. The application is successful enough to identify everyday objects, read text and identify age and gender of a human. The application uses Visual Recognition, Natural Language Classifier, Text to Speech convertor and Speech to Text convertor of the Watson APIs.
S. K. Kolluru, Shreyans Shrimal, Sudharsan Krishnaswamy

News Buddy

Abstract
News Buddy is a web-based chatbot application with three tier architecture using Waston Services. The solution is hosted on an IBM Bluemix platform. This is an ML-based interactive chat application which converses with the user and provides users various news as per user’s interactions.
Chandana Kotta, Shashank Motepalli, Tadepalli Sandeep

Conclusion

Frontmatter

Concluding Remarks

Abstract
Working with computer scientists, application developers are bringing state-of-the-art Artificial Intelligence (AI) into our day-to-day lives, changing the way we work and live. Finance, health care, education, industrial automation, security are some of the areas seeing tremendous growth in AI applications. Such applications typically rely on a rich layer of cognitive AI services which are continuously developed and improved by researchers.
Danish Contractor, Aaditya Telang
Weitere Informationen