About
Activity
-
We call them "pests." Nature calls them "architects." 🐿️🏗️ We often see squirrels as nothing more than a backyard nuisance or a hazard on our…
We call them "pests." Nature calls them "architects." 🐿️🏗️ We often see squirrels as nothing more than a backyard nuisance or a hazard on our…
Liked by Anusha Prakash
-
One of the biggest challenges in AI is context. We're bridging that gap with the launch of Personal Intelligence in Gemini. I’m proud to have worked…
One of the biggest challenges in AI is context. We're bridging that gap with the launch of Personal Intelligence in Gemini. I’m proud to have worked…
Liked by Anusha Prakash
-
Debugging My Sleep: How Sudoku Broke My Doom-Scrolling Loop I realized I was running on adrenaline. Recently, I pressed pause and took a sabbatical…
Debugging My Sleep: How Sudoku Broke My Doom-Scrolling Loop I realized I was running on adrenaline. Recently, I pressed pause and took a sabbatical…
Liked by Anusha Prakash
Experience & Education
Volunteer Experience
-
Volunteer
ACM, Association for Computing Machinery
- 9 months
Education
Volunteered at Shiksha program, aimed to teach computers to high school students
-
Subject Matter Expert
JPMorgan Chase & Co.
- Present 11 years 2 months
Science and Technology
Served as Subject Matter Expert for Code for Good Challenge to support NGOs
-
Volunteer
Good Work India
- Present 11 years 1 month
Education
Volunteered in School Kit Collection drive for young school going children
-
Volunteer
Global Cancer Concern India
- Present 18 years 10 months
Social Services
Volunteered by spreading awareness about cancer and helping with relief efforts for cancer sufferers
-
Mentor
Girls Who Code
- 2 months
Education
Provided a young female student participating in the Girls Who Code summer immersion program, with insight and guidance on how they can succeed in the workplace.
Publications
-
Multi-Relational Question Answering from Narratives: Machine Reading and Reasoning in Simulated Worlds
ACL 2018
See publication• Question Answering (QA), as a research field, has primarily focused on either knowledge bases (KBs) or free text as a source of knowledge. These two sources have historically shaped the kinds of questions that are asked over these sources, and the methods developed to answer them.
• In this work, we look towards a practical use-case of QA over user-instructed knowledge that uniquely combines elements of both structured QA over knowledge bases, and unstructured QA over narrative…• Question Answering (QA), as a research field, has primarily focused on either knowledge bases (KBs) or free text as a source of knowledge. These two sources have historically shaped the kinds of questions that are asked over these sources, and the methods developed to answer them.
• In this work, we look towards a practical use-case of QA over user-instructed knowledge that uniquely combines elements of both structured QA over knowledge bases, and unstructured QA over narrative, introducing the task of multi- relational QA over personal narrative. As a first step towards this goal, we make three key contributions:
(i) we generate and release TextWorldsQA, a set of five diverse datasets, where each dataset contains dynamic narrative that describes entities and relations in a simulated world, paired with variably compositional questions over that knowledge
(ii) we perform a thorough evaluation and analysis of several state-of-the-art QA models and their variants at this task, and
(iii) we release a lightweight Python-based framework we call TEXTWORLDS for easily generating arbitrary additional worlds and narrative, with the goal of allowing the community to create and share a growing collection of diverse worlds as a test-bed for this task. -
Kinect based Real Time Gesture Recognition Tool for Air Marshallers and Traffic Policemen
The 8th IEEE International Conference on Technology for Education (T4E 2016), IIT Bombay, India
See publication• The Microsoft Kinect which is a motion sensing input device presents a very straightforward and affordable approach to facilitate real-time user interaction. Although a lot of research has been conducted on the application of Kinect to gaming and virtual reality environments, its relevance to real-world scenarios has not been explored much.
• The features provided by the driver platforms such as OpenNI and Microsoft Kinect Software Development Kit (SDK) for development using Kinect…• The Microsoft Kinect which is a motion sensing input device presents a very straightforward and affordable approach to facilitate real-time user interaction. Although a lot of research has been conducted on the application of Kinect to gaming and virtual reality environments, its relevance to real-world scenarios has not been explored much.
• The features provided by the driver platforms such as OpenNI and Microsoft Kinect Software Development Kit (SDK) for development using Kinect coupled with the motion sensing ability of Kinect, presents a unique opportunity for extending the scope of the Kinect sensor.
• This paper proposes a system for automatically recognizing the road traffic control gestures of police officers and air marshalling commands by ground personnels. This system is aimed for self-learning, training and testing these officers to equip them with the skills to tackle real-world situations.
• Since these applications are very crucial and performing accurate gestures are of at most importance, this system will prove to be very essential. Experimental results also demonstrate that our system is robust and effective and is suitable for real-time application. -
iKAN - A Kannada Transliteration Tool for Assisted Linguistic Learning
The 4th International Conference on Emerging Research in Computing, Information, Communication and Applications (ERCICA 2016), India
See publication• Kannada is the official language of the state of Karnataka and is a prominent South Indian language. Despite being one among the forty most spoken languages of the world, the rich heritage of Kannada literature along with its rich culture has so far remained limited only to those who know how to read Kannada.
• In order to bring this to the masses, there has been minimal research work in automatic translation and transliteration of Kannada works. In this paper, we present iKAN, a…• Kannada is the official language of the state of Karnataka and is a prominent South Indian language. Despite being one among the forty most spoken languages of the world, the rich heritage of Kannada literature along with its rich culture has so far remained limited only to those who know how to read Kannada.
• In order to bring this to the masses, there has been minimal research work in automatic translation and transliteration of Kannada works. In this paper, we present iKAN, a platform that aims to automatically transliterate and later, translate the Kannada script into English language, thus providing access to Kannada Literature.
• A phonetic mapping scheme was developed to enable readability and Kannada alphabets were represented in UTF-8 format, for transliteration into English. The Unicode approach used for the representation of characters resulted in high accuracy.
• We demonstrate the effectiveness of the proposed approach towards building a user-friendly Transliterator for Kannada that can be used as an e-learning tool. -
Data Synchronization on Android Clients
The 7th IEEE International Conference on Communication Software and Networks (ICCSN 2015), China
• Past decade has witnessed meteoric advances in the field of mobile computing owing to the development of affordable hardware technologies as well as user-friendly software platforms. Android, the platform marketed by Google has boomed in sales over the past few years making it one of the major mobile platforms in the market.
• The steady growth of wireless information and communication technology in convergence with rise in the penetration of Internet has led to the evolution of a…• Past decade has witnessed meteoric advances in the field of mobile computing owing to the development of affordable hardware technologies as well as user-friendly software platforms. Android, the platform marketed by Google has boomed in sales over the past few years making it one of the major mobile platforms in the market.
• The steady growth of wireless information and communication technology in convergence with rise in the penetration of Internet has led to the evolution of a wide range of mobile applications like news, multi-player games, social networking, messaging, etc. that need to access remote data.
• For the optimal functioning of all these applications an efficient synchronization mechanism is vital. However smart-phones have limited computational resources, power restrictions and intermittent Internet connections which pose a challenge for smooth synchronization.
• This paper proposes a two-way data synchronization mechanism between multiple Android clients and a central server to address these challenges. We employ a batching logic to ensure efficient data transfer in poor network environments and a server-side conflict resolution mechanism to reduce overhead on the clients, which ensures optimal processing and battery power consumption by the clients.Other authorsSee publication
Patents
-
System and Method for Prediction of Sellability of Fashion Products
Issued 11276100 15/979,641
See patentDeep Learning ensemble model to predict the sellability of fashion products
Courses
-
Advance Computer Networks
IT307
-
Automata Theory and Compiler Design
IT303
-
Cloud Computing
IT409
-
Computer Communication and Networking
IT251
-
Computer Graphics
IT254
-
Computer Programming
CO110
-
Data Structures and Algorithms
IT200
-
Database System
IT301
-
Deep Learning
11-785
-
Design and Analysis of Algorithms
IT252
-
Digital Design and Computer Organization
IT201
-
Directed Research
11-690
-
Distributed Computing
IT360
-
Human Computer Interaction
IT351
-
Information Assurance and Security
IT352
-
Language and Statistics
11-761
-
Large Scale Multimedia Analysis
11-775
-
Linear Algebra and Matrices
MA204
-
Machine Learning
10-601
-
Machine Learning for Text Mining
11-641
-
Mathematical Foundation of Information Technology
MA200
-
Mobile Adhoc Networks
IT411
-
Number Theory and Application
EC400
-
Object Oriented Analysis and Design
IT306
-
Operating System
IT250
-
Paradigms of Programming - I
IT206
-
Paradigms of Programming - II
IT253
-
Parallel Computing
IT300
-
Principles of Management
HU302
-
Search Engines
11-642
-
Soft Computing
IT355
-
Software Engineering
IT350
-
Unix Programming and Practice
IT202
-
Web Services
IT450
-
Web Technologies and Application
IT302
-
Wireless Sensor Networks
IT410
Projects
-
Contextual Code Generation with Interspersed Natural Language Descriptions
-
• Creating a new benchmark data set by mining open-source Jupyter notebooks on GitHub
• Proposing an approach for code generation which leverages both previous code context and Natural Language descriptions through multi-source attention with a copy mechanism
df -
Predicting Patients’ Conversation Transitions in Online Health Support Groups
-
-
Zero-Shot Content Learning for Email Classification
-
• Implemented an Unsupervised approach by applying Bigram / Unigram Sent2Vec embeddings to the emails and concept descriptions and classified the emails based on cosine similarity in Python
• Implemented a Regression based approach where the input was modified to learn a weight based on the concept descriptions and was then used for classification in PyTorch -
Localizing Video Moments based on Natural Language Descriptions
-
• Implemented a Long Short-Term Memory Network (LSTM) architecture to extract textual features and extracted ResNet RGB and OpticalFlow (global and local) video features to retrieve specific temporal moments
• Experimented with language embedding, language model, video and audio features, and fusion techniques (late, early and double fusion) to improve the Rank@1 and Rank@5 metrics
• Implemented BiDNN multimodal fusion technique along with activity recognition and image captioning…• Implemented a Long Short-Term Memory Network (LSTM) architecture to extract textual features and extracted ResNet RGB and OpticalFlow (global and local) video features to retrieve specific temporal moments
• Experimented with language embedding, language model, video and audio features, and fusion techniques (late, early and double fusion) to improve the Rank@1 and Rank@5 metrics
• Implemented BiDNN multimodal fusion technique along with activity recognition and image captioning features to enhance the retrieval performance -
Text Generation from Structured Data
-
• Proposed and implemented a Hierarchical Attention based Pointer Generator model to generate long summaries from the tables in the RotoWire dataset
• Experimented with Long Short-Term Memory Network (LSTM), Recurrent Neural Network (RNN) and BiLSTM based encoder-decoder models to minimize the negative log-likelihood of the words in gold text
• Visualized the global and local attention to understand the attention weights learnt by the trained model -
End-to-End Speech recognition System using Deep Learning
-
• Implemented speech recognition system by using Multilayer Perceptrons (MLP), Long Short-Term Memory Networks (LSTM), Convolutional Neural Networks (CNN) and attention to obtain transcriptions from the Mel-frequency cepstral coefficient (MFCC) utterance data
• Performed Beam search and used Sphinx and Connectionist Temporal Classification (CTC) loss to build the end-to-end system -
Large-Scale Multimedia Event Detection
-
• Extracted audio features like SoundNet, Mel-frequency cepstral coefficients (MFCC), Automatic Speech Recognition (ASR) Transcriptions and video features like Convolutional Neural Network (CNN), Speeded Up Robust Features (SURF) and Improved Dense Trajectories (IDT) to perform multimedia event detection
• Implemented early fusion, late fusion and double fusion techniques using these features to obtain a high Mean Average Precision (MAP) value -
Movie Recommendation System using Collaborative Filtering
-
• Implemented memory-based and item-based collaborative filtering methods in Python to predict the ratings of movies on the Netflix Prize dataset
• Employed Pearson’s Correlation Coefficient to account for user and movie bias in the ratings
• Implemented Probabilistic Matrix Factorization as an alternative to the nearest neighbor based approaches for ratings prediction -
Deep Question Answering for Personal Agents
-
• Worked on the End-to-End Memory Networks approach to perform Machine Reading and Reasoning in Simulated Worlds using TensorFlow
• Enhanced the architecture to account for multiple answers and experimented with word embeddings to improve the performance
-
News Article Classifier to Detect Real and Fake Articles
-
• Extracted various Semantic, Syntactic, Statistical and Language Model based features to capture the difference between real articles and Trigram generated fake articles
• Experimented with various classifiers and performed feature selection and GridSearch to improve model accuracy and log perplexity -
QryEval - Text Based Information Retrieval System
-
• Developed a text-based search engine which includes exact match (Ranked and Unranked Boolean) and best match (BM25, Indri) retrieval models using the Lucene in Java
• Implemented query expansion, search diversification (xQuAD, PM2) and pseudo-relevance feedback to improve the quality of search results on the Gov2 dataset
• Implemented LearningtoRank(LeToR) and employed an SVM classifier to re-rank the retrieved documents
-
Network based attack detection technique for Open vSwitch based Virtual Networks
-
Proposed approaches to a novel network based intrusion detection system for Open vSwitch using SnortAI and implemented the same in C and Python to strengthen the virtual network security
-
Aetrial - A training tool for Air Marshallers and Traffic Policemen
-
Developed a system for automatic gesture recognition using Microsoft Kinect SDK, Visual Studio and C# to identify the road traffic control gestures and air marshalling commands
-
Clustering of Web Services to Bootstrap its Discovery by Search Engines
-
• Performed feature extraction on the WSDL documents, followed by feature integration and clustering using Python NLTK to address the problem of searching non-semantic web services
• Computed Similarity Factor based on Normalized Google Distance to cluster functionally similar services
-
iKAN - A Transliteration Tool for Assisted Linguistic Learning
-
• Proposed an algorithm to perform transliteration based on Consonant*Vowel approach and a phonetic mapping scheme ensuring high accuracy
• Developed an effective framework for transliteration from Kannada to English using C and MATLAB to assist people who understand Kannada but lack the knowledge of its script -
Demonstration of OpenGL Functionalities
-
"The Solar System" project was implemented using OpenGL which included features such as 2D and 3D animations, lighting, shadow, transformations, clipping, keyboard and mouse I/O methods, etc.
-
Mini TCP/IP Stack Implementation
-
Implemented all the four layers of the TCP/IP model using Socket Programming in C ensuring an end- to-end connectivity with features including CRC error check, TCP Handshaking and flow control
-
LibKart - Library management System
-
An android application which acts as a Library Information System was developed using GUI, File Handling, XML, Java and Android API.
Honors & Awards
-
2018 WomenSCS DataVisor Scholar
DataVisor
Awarded Scholarship by DataVisor to attend Grace Hopper Celebration of Women in Computing 2018
-
IAS Summer Research Fellow
Indian Academy of Science
Awarded a fellowship by the Indian Academy of Sciences' to pursue a summer research internship at Indian Institute of Science (IISc)
-
Student Welfare Fund Award
VVS Sardar Patel PU College
Received a Scholarship for Academic Excellence
-
Prathibha Puraskara
LIC and Karnataka Secondary Teachers' Association
Awarded for Outstanding Performance in academics
-
Vidhyaratna Award
Karnataka Pratibha Vardhaka Academy
Awarded for Outstanding Performance in academics
Languages
-
English
-
-
Hindi
-
-
German
-
-
Kannada
-
-
Sanskrit
-
More activity by Anusha
-
Ending 2025 stepping into a leadership role at Adobe as Senior Manager, Machine Learning 🎉 I’m truly thrilled for the opportunity to bring my voice…
Ending 2025 stepping into a leadership role at Adobe as Senior Manager, Machine Learning 🎉 I’m truly thrilled for the opportunity to bring my voice…
Posted by Anusha Prakash
Other similar profiles
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top contentOthers named Anusha Prakash in United States
11 others named Anusha Prakash in United States are on LinkedIn
See others named Anusha Prakash