Invented & Patented - A Method for Acquiring QoS and QoE Parameter of PaaS Cloud Renderfarm Services

• It is innovative to acquire the Quality of Service (QoS) and Quality of Experience (QoE) parameters of PaaS Cloud Renderfarm Services for evaluating the services on multiple criteria and recommending the service.

Intent Recognition of Musicians for Music Generation using LLM's for NER

Role:  Lead Research Coordinator - Text to Sound Team

NLP Packages used: spaCy NER component with "Bloom'' embeddings, RoBERTa Transformer Model for NER, Word2Vec for Topic Modeling.

Led "Text to Sound" team and Built two modules for the intent recognition namely: a) Named Entity Recognition (NER) module and b) Word to Words Matcher (WWM) module.

a) Named Entity Recognition (NER) module:

The Transformer-based Model was implemented using the pre-trained model of spaCy's RoBERTa base model provided by HuggingFace for NER. This model is then tuned according to our musical instrument dataset for learning to recognize musical instruments and qualities.

b) Word to Words Matcher (WWM) for Standardized Intent Recognition:

A word-to-word matching component was devised to convert the free text sound quality intent recognized by the NER model to a standardized intent of 18 key value pairs of data relating to the sound quality are returned which are the instrument names mentioned and a subset of 18 standardized sound qualities.

YouTube Github - Text to Sound

Active Learning for automatic labelling of keywords to generate labelled data

Role: Lead Research Coordinator - Dataset Team

Tools / Techniques used:   Bidirectional LSTM, Active learning, Doccanno - Data Annotation Tool, Data Augmentation techniques

• As the "Dataset" Team Lead, led the team, coordinated with the music team members to identify the right dataset from "Reddit".

• Led the process of Scrapping and annotating the “Reddit” data that contain “Guitar Timbre” keywords.

• Applied Active Learning for automated Labeling of “Guitar Timbre” keywords.

Github - Dataset Team

A Tool for Automatic Data Annotation and Labeling through Active Learning

• Applied Active Learning techniques to annotate the music domain specific keywords and label them programmatically without human labelers.

Text Generation Tool for Data Augmentation using Bidirectional LST

•   Applied Bidirectional LSTM to generate augmented data that involves music domain specific keywords in a natural speech form for data annotation.

Knowledge Graph for Domain Specific Recommender System using Ontology & Taxonomy

Project Tools Used: Neo4j, NetworkX

• • Developed a cloud renderfarm services domain specific knowledge graph based on the ontology & tree structured taxonomy from the websites of eleven popular real time cloud renderfarm services.

• • Applied GraphML to filter the services based on the functional and non functional offerings of services and client requirements.

Consulting Project - Cost & Service Offerings Modelling of PaaS Cloud Renderfarm services for Comparison & Selection based on Demand Forecasting

• Machine Learning for forecasting the future demand and the future cost of the animation jobs rendering & recommend on On-Prem rendering setup or migration to Cloud Renderfarm services for cost benefit / efficiency.

Personalized Recommender System for Domain Specific Cloud Services Recommendation

• Combining content based and collaborative filtering Machine Learning based techniques with re-ranking modules for personalizing recommendation methodology.

Ranking of Filtered Cloud Services using Multi Criteria Decision Making Algorithms

• Applied Multi Criteria Decision Making (MCDM) methods to the QoS data collected from the real time experiments and ranked the filtered cloud services based on the multiple criteria requirements of the users. 

Development of an Indicative Biomedical System for Detecting Diseases using captured IoT and Medical Images

•  The Indicative Biomedical System aids in Telemedicine by facilitating emergency medicine delivery and patient care using targeted test and treatment using the captured IoT & medical images data.