Sep 2018 – Aug 2021 · 2 yrs 11 mos · Bangalore, India

Aug 2017 – Feb 2018 · 6 mos · Heverlee

• Deep learning based face recognition for indexing digital archives
• Processed terabytes of noisy multi-modal data and developed semi-automatic methods to curate a training dataset by integrating a state-of-the-art deep learning based face detection system on the vision side and a named entity recognition system on the language side.
• Fine-tuned a cutting edge convolutional neural network on procured data and built a retrieval model with 97% accuracy on the top 50% of the results.
• Collaborated with a leading national media group and a media asset management company to deploy the software and models into a search-and-retrieval system that is widely adopted by thousands of journalists on multiple platforms including desktop and mobile apps.

Sep 2011 – Jun 2017 · 5 yrs 9 mos · Heverlee

• 1. Enhancing Object Recognition in Computer Vision using Natural Language
• Designed a novel weakly supervised framework to perform multi-label classification using domain adaptation on multi-modal data.
• Proposed a feature transformation for deep learning features and demonstrated its benefit in classification problems.
• Applied the machine learning pipeline to perform wildlife object recognition, achieving a gain of 4x in F1 score over several state-of-the-art methods.
• 2. Entity Linking across Vision and Language
• Designed a probabilistic graphical model to jointly link entities across vision and language.
• Demonstrated the use of the model in a video with subtitles to resolve textual entities using visual cues and to recognise visual objects using textual inputs.
• Proved that the joint modelling of textual and visual entities yields significantly better performance in both vision and language over strong independent baselines.
• 3. Building Search Tools for Networked Media Production and Archiving
• Developed weakly supervised algorithms for indexing, retrieval and summarization of videos using the accompanying closed captions and speech data from an Automatic Speech Recognition (ASR) system.
• Extracted key events automatically from wedding documentaries by learning from external corpora.
• Collaborated with top European media companies and research institutions leading to product realisation.
• 4. Customizing Web Search for Children
• Designed a machine learning pipeline directed at improving the presentation of search results for children.
• Proposed an unsupervised method to estimate the breadth of a web search query and built models to summarize or expand query facets based on their scope.
• Achieved high quality facets in terms of relevance and diversity, as validated by users.

Sep 2006 – Aug 2010 · 3 yrs 11 mos · Chennai Area, India

• Key initiatives:
• 1. World Class Closure: Customized the standard SAP system in Financial Accounting (FI) and Controlling (CO) areas to shorten the number of days for closing the accounting books, generating productivity savings of $4M.
• 2. SAP Upgrade: Upgraded the existing enterprise platform for the client and set the stage for many other business improvements ahead. Performed the business critical task of ABAP Remediation throughout the project.
• 3. Transport Management System: Reengineered transportation processes to align with industry best practices. Developed custom programs in the areas of Material Management (MM) and Sales & Distribution (SD) and increased the efficiency of handling invoices and freight by 25%.
• 4. SAP-ABAP Mentoring: Developed training modules with theory and hands-on sessions for fresh graduates. The coaching included advanced topics such as business add-ins, user-exits and object-oriented programming in SAP.