Articles by Pranjal
Activity
-
We had an incredible close to 2024 at Disco. In December, we hit a $20M run rate for the first time, representing stellar 340% YoY growth. As we…
We had an incredible close to 2024 at Disco. In December, we hit a $20M run rate for the first time, representing stellar 340% YoY growth. As we…
Liked by Pranjal Daga
-
As we enter 2025, one of the things that I am most excited about is multi-agent AI systems. The space is still early, but we are exploring parts of…
As we enter 2025, one of the things that I am most excited about is multi-agent AI systems. The space is still early, but we are exploring parts of…
Liked by Pranjal Daga
-
Thank you AIM Research for highlighting Outset!!
Thank you AIM Research for highlighting Outset!!
Liked by Pranjal Daga
Experience & Education
-
Accend (YC S23)
View Pranjal’s full experience
See their title, tenure and more.
or
Publications
-
Adapting Event Embedding for Implicit Discourse Relation Recognition
CoNLL 2016 Shared Task
Predicting the sense of a discourse relation is particularly challenging when connective markers are missing. To address this challenge, we propose a simple deep neural network approach that replaces manual feature extraction by introducing event vectors as an alternative representation, which can be pre-trained using a very large corpus, without explicit annotation. We model discourse arguments as a combination of word and event vectors. Event information is…
Predicting the sense of a discourse relation is particularly challenging when connective markers are missing. To address this challenge, we propose a simple deep neural network approach that replaces manual feature extraction by introducing event vectors as an alternative representation, which can be pre-trained using a very large corpus, without explicit annotation. We model discourse arguments as a combination of word and event vectors. Event information is aggregated with word vectors and a Multi-Layer Neural Network
is used to classify discourse senses. This work was submitted as part of the CoNLL 2016 shared task on Discourse Parsing. We obtain competitive results, reaching an accuracy of 38%, 34% and 34% for the development, test and blind test datasets,
competitive with the best performing system on CoNLL 2015.Other authors -
-
Reducing Infrequent-token perplexity via variational corpora
Association of Computation Linguistics (ACL)
Recurrent neural network (RNN) is recognized as a powerful language model. We investigate deeper into its performance portfolio, which performs well on frequent grammatical patterns but much less so on less frequent terms. Such portfolio is expected and desirable in applications like autocomplete, but is less useful in social content analysis where many creative, unexpected usages occur (e.g., URL insertion). We adapt a generic RNN model and show that, with variational training corpora and…
Recurrent neural network (RNN) is recognized as a powerful language model. We investigate deeper into its performance portfolio, which performs well on frequent grammatical patterns but much less so on less frequent terms. Such portfolio is expected and desirable in applications like autocomplete, but is less useful in social content analysis where many creative, unexpected usages occur (e.g., URL insertion). We adapt a generic RNN model and show that, with variational training corpora and epoch unfolding, the model improves its performance for the task of URL insertion suggestions.
Other authors -
-
An Efficient Network Management and Power Saving Wake ON-LAN
Springer: Advances in Intelligent Systems and Computing Volume 259, 2014, pp 767-777
In distributed systems a computer generally process information of distributed application or provide service in distributed system. Therefore, computers connected in distributed system need to keep on all time. It leads to the concept of Wake-on-LAN (Local Area Network). However, keeping on during the idle period is the wastage of power. Therefore, it is essential to save the power while being efficient in network management. In this paper we propose an improved Wake-on-LAN device that…
In distributed systems a computer generally process information of distributed application or provide service in distributed system. Therefore, computers connected in distributed system need to keep on all time. It leads to the concept of Wake-on-LAN (Local Area Network). However, keeping on during the idle period is the wastage of power. Therefore, it is essential to save the power while being efficient in network management. In this paper we propose an improved Wake-on-LAN device that incorporates the usage of basic Wake-on-LAN technology into a network management and power saving product.
Other authors -
Courses
-
Algorithm Design and Analysis
-
-
Applied Probability, Statistics and Reliability
-
-
Artificial Intelligence
-
-
Computer Architecture and Organization
-
-
Computer Graphics
-
-
Computer Networks (+Lab)
-
-
Data Mining and Warehousing
-
-
Data Structures and Algorithms (+Lab)
-
-
Database Systems (+Lab)
-
-
Discrete Mathematical Structures
-
-
Embedded Systems (+Lab)
-
-
Linear Algebra
-
-
Machine Learning
EECS 349
-
Machine Learning: by Andrew Ng, Stanford University
-
-
Microprocessors and Interfacing (+Lab)
-
-
Object Oriented Paradigm (+Lab)
-
-
Operating Systems (+Lab)
-
-
Operations Research
-
-
Programming Language Translators
-
-
Soft Computing
-
-
Software Engineering (+Lab)
-
-
Theory of Computation
-
-
Write of Passage (David Perell)
-
Projects
-
Structured Extraction of Document Reading Order Text
- Present
- Worked on structured extraction of reading order text in complex documents.
- Proposed using and aligning open source Gutenberg corpus as dataset due to lack of annotated data.
- Deployed a probabilistic N-gram Language Model to evaluate the likelihood of the next reading order candidate.
- Suggested and used RNNs/LSTMs which surpassed the N-gram baseline results of 35000 files with just 2000 files. -
Detecting Direct Perception in Autonomous Driven Cars
- Present
Training a network for Automated Driver Assistance Systems (ADAS) which takes a version of the video collected and 'drive' commands.
Using Convolutional Neural Networks for making the trained network learn to apply proper acceleration and steering angle to the video input / scenery. -
Benchmarking Deep Learning Tools
- Present
Compared various deep learning software packages like Torch, Theano, Caffe on different metrics (accuracy, timing, memory usage, etc.) for different deep algorithms.
Other creators -
-
Understanding Climate Change: A Data Driven Approach
- Present
This Expeditions project aims to address key challenges in the science of climate change by developing methods that leverage the abundance of climate and ecological data available from satellite and ground-based sensors, the observational record for atmospheric, oceanic, and terrestrial processes, and physics-based climate model simulations.
Extended the work of Ghosh et al (http://www.nature.com/nclimate/journal/v2/n2/fig_tab/nclimate1327_F2.html) to global and multi-climate model…This Expeditions project aims to address key challenges in the science of climate change by developing methods that leverage the abundance of climate and ecological data available from satellite and ground-based sensors, the observational record for atmospheric, oceanic, and terrestrial processes, and physics-based climate model simulations.
Extended the work of Ghosh et al (http://www.nature.com/nclimate/journal/v2/n2/fig_tab/nclimate1327_F2.html) to global and multi-climate model data.
Downloaded the massive climate precipitation (multi-TB) NetCDF data from the servers.Other creators -
-
Exploring Deep Learning applications to fMRI
- Present
Used deepnet package to learn a 3-layer Boltzmann machine for pretraining.
Dataset to be used: pain, schizophrenia, cocaine addiction.Other creators -
-
Learning an Effective Individual Diabetes Management Policy
- Present
Developed a tool which automatically adjusts the amount of insulin to be injected.
Modified T1DMS Simulator to use it with 20 in-silico patients.
Demonstrated that a supervized learning approach is better than certain reinforcement learning
techniques, as optimal values were accessible.
Compared the performance of the learned method with an expert diabetologist.Other creators -
-
SnapBin!, a Windows 8 app
See projectAppreciated by Microsoft App Excellence Lab as a Rockstar app.
Felicitated with Editor’s Pick Award given to selected Best Windows 8 apps.
Quick way of taking quick notes in the midst of doing other activities and sharing them. -
Who's the Author (Wita)
-
Remember the times you just want to know the author of a particular article? Well now, you can, using Wita. Wita uses perceptrons (version 1), word vectors (version 2) and Recurrent Neural Networks (version 3) to find the author, their gender, age, demographic information, etc.
Other creators -
-
Reducing Infrequent-token perplexity via variational corpora
-
Published in proceedings of ACL 2015 Main Conference held at Beijing, China.
Recurrent neural network (RNN) is recognized as a powerful language model. We investigate deeper into its performance portfolio, which performs well on frequent grammatical patterns but much less so on less frequent terms. Such portfolio is expected and desirable in applications like autocomplete, but is less useful in social content analysis where many creative, unexpected usages occur (e.g., URL insertion). We…Published in proceedings of ACL 2015 Main Conference held at Beijing, China.
Recurrent neural network (RNN) is recognized as a powerful language model. We investigate deeper into its performance portfolio, which performs well on frequent grammatical patterns but much less so on less frequent terms. Such portfolio is expected and desirable in applications like autocomplete, but is less useful in social content analysis where many creative, unexpected usages occur (e.g., URL insertion). We adapt a generic RNN model and show that, with variational training corpora and epoch unfolding, the model improves its performance for the task of URL insertion suggestions.Other creators -
-
'Deep' Recommendation System
-
Proposed a ’collaborative’ deep learning model which tightly couples convolutional neural networks and probabilistic matrix factorization.
Better performance than Collaborative Filtering was shown using MovieLens dataset. -
Detection, Surveying and Supervision of Parkinson’s Disease
-
Developing an intelligent mobile application that collects patient input through tests (Gait, Handwriting, Speech, Tremors) and provides interactive feedback to patients and monitoring summaries for physicians.
Used K-means clustering to discover hidden relationships which may lead to identify different
Parkinson’s Disease variations. -
'Learning' Personality from Social Posts
-
Aiming to scrutinize linguistic content from a social platform, extracting keywords to analyze the personality.
Used a linear support vector machine (SVM) for classifying sex and ridge regression for predicting age and each factor of personality. -
MIT SANA Mobile: Protocol Builder
-
Built a database that contains rapid search, review options and a training set for (future) automated A.I. classification of images, audio and video.
Worked on integrating SANA with OpenMRS (a Medical Record System).Other creators -
-
Machine Learning based Brain controlled system
-
Obtained EEG signal records from PhysioNet dataset and used those EEG signals as a communication link between brain and the system.
Used Support Vector Machine (SVM) and Neural Networks (NN) algorithms to generate required decision rules from extracted features.
Used MATLAB NN toolbox and MySVM for all training and testing experiments.Other creators -
-
An Efficient Network Management and Power Saving Wake ON-LAN
-
Published in the AISC Series of Springer and oral presentation at 3rd International Conference on Soft Computing for Problem Solving (SocProS ’13) held on 26th – 28th December, 2013
In the sleep mode operation, the proposed system simultaneously caches data for its client while
waking it using the standard Wake-on-LAN implementation.
Designed the software capture packets from the internet using PCAP library programming.
Worked on Beagleboard XM, a single board ARM computer…Published in the AISC Series of Springer and oral presentation at 3rd International Conference on Soft Computing for Problem Solving (SocProS ’13) held on 26th – 28th December, 2013
In the sleep mode operation, the proposed system simultaneously caches data for its client while
waking it using the standard Wake-on-LAN implementation.
Designed the software capture packets from the internet using PCAP library programming.
Worked on Beagleboard XM, a single board ARM computer which has the ability to run LINUX.Other creators -
-
graVITas Registration Software
-
Developed events’ registration software for graVITas’13- An International TechnoManagement fest at VIT Vellore, India.
-
Adaptive GPS Algorithm using Djiktra’s Technique
-
Designed a hybrid algorithm using Fuzzy Logic allowing user to incorporate information such as
traffic levels, weather conditions to find an optimal shortest distance using Djikstra’s Algorithm.Other creators -
Honors & Awards
-
TechCrunch Disrupt SF Hackathon 2017
TechCrunch
Awarded the following for developing Drive Understand Improve (DUI) (https://devpost.com/software/drive-understand-improve-d-u-i):
- Best use of Arity API and SDK
- Judges Choice to get invited to attend TechCrunch Disrupt 2017
TechCrunch Stage Demo: https://techcrunch.com/video/drive-understand-improve-d-u-i/59bec9dd9efa89480f07a9f0/ -
HackPrinceton 2016
Princeton University
Awarded the following at HackPrinceton 2016 for developing EyePhone (http://devpost.com/software/eyephone):
- Best Mobile App
- Most Launchable Product
- Best Use of Data Visualization
- PrincetonPy / PICSciE Prize for Every Day Data for Tomorrow -
HackIllinois 2016 - Best Microsoft Hack
University of Illinois- Urbana Champaign
Awarded the Microsoft's Best Microsoft Hack award and Best use of Azure award at HackIllinois 2016 for developing NeuroDoc (http://devpost.com/software/neurodoc)
-
HackIllinois 2016 - First Prize (Best Software Hack)
University of Illinois- Urbana Champaign
Awarded the First Prize at HackIllinois 2016 for developing NeuroDoc (http://devpost.com/software/neurodoc)
-
PennApps XIII - Best Hack in Health Route
University of Pennsylvania
Awarded the Best Health Hack prize for developing DataDoc (http://devpost.com/software/datadoc)
-
Boston Hacks - First Place Prize
Major League Hacking | Boston University
Won the First Prize among 500+ participants for developing WhatsUpDoc (http://devpost.com/software/whatsupdoc).
-
IBM Winter School
IBM
Selected to attend IBM winter school on Big Data Analytics and Cognitive Computing.
Provided a Travel grant to attend the same. -
Mitacs Globalink Research Fellowship
Mitacs
Selected to pursue a fully funded summer research internship at University of Alberta, Edmonton, Canada.
Eligible for Mitacs Globalink Graduate Fellowships which provides financial support for graduate studies in Canada -
MIT- SANA mHealth Challenge
MIT
2nd runners up among 110 selected participants for pitching a mobile health solution to manage a patient suffering from Parkinson's Disease.
-
Kishore Vaigyanik Protsahan Yojana (KVPY) Fellowship
Department of Science and Technology, Government of India
Awarded to top 125 students of the country to pursue a research career.
Languages
-
English
Full professional proficiency
-
Hindi
Native or bilingual proficiency
-
French
Limited working proficiency
Organizations
-
Purdue Graduate Student Government
Senator, Department of Computer Science
- Present -
IEEE Computer Society
Joint Secretary
- -
Indian Society for Technical Education (ISTE)
Member, Tech League
-
Other similar profiles
Explore more posts
Explore collaborative articles
We’re unlocking community knowledge in a new way. Experts add insights directly into each article, started with the help of AI.
Explore MoreOthers named Pranjal Daga
-
Pranjal Daga
Associate @ZS | Ex-SDE Intern @Resilient-AI @IIT Bombay @Hill One Technologies
-
Pranjal Daga
🏗️ Project Management Executive in Heritage Sector | Spearheading Innovative Membership Programs & Online Engagement Strategies🏗️
-
Pranjal Daga
--
-
Pranjal Daga
Metal scrap/ E Vehicle/ Cement/ Gypsum/ Activated Carbon/ Anthracite Coal
11 others named Pranjal Daga are on LinkedIn
See others named Pranjal Daga