Portfolio item number 1
Short description of portfolio item number 1
Posts by Collection
portfolio
Portfolio item number 2
Short description of portfolio item number 2
projects
Spam Classifier
Published:
This project tries to classify an incoming email as spam or ham using a Naive Bayes Classifier. The algorithm that we have used gives a strong baseline for the task. Infact many email clients use some variation of this algorithm for detecting spam emails.
Text Summarization Tool
Published:
It is an automatic summarization tool that generates summary of a given text using linguistic, statistical and extractive methods. The algorithm for generating summary was implemened in C++ and the user interface was implemented in Visual Basic. It was a group project submitted by a group of 4 studennts in partial fulfillment of the requirement for the award of the degree of Bachelor of Technology in Software Engineering, to the Delhi Technological University in 2013.
Reimplementation of InfoGAN
Published:
InfoGAN is almost similar to GAN except for the fact that it is also able to learn disentangled(interpretable) representations in a completely unsupervised manner. These interpretable features can be used by many downstream tasks such as classification, regression, visualization, and policy learning in reinforcement learning. The code in this repository was reimplemented on top the official repository of the paper - InfoGAN: Interpretable Representation Learning by Information Maximizing Generative Adversarial Nets.
Dynamic Models with IAF
Published:
Through this work, we propose the incorporation of Inverse Autoregressive Flows for determining the state space (latents) in a dynamical system model. This reduces the number of samples that need to be obtained in order to approximate the posterior distribution (and thus the underlying states/latents for a set of observations and controls) from one per time step to one per sequence of observations. Our experiments with pendulum-v01, an environment from openai gym confirmed that the accuracy with which the observations are generated are close to the state of the art for sequence models.
Predicting depth image from a single RGB image
Published:
This work implements the architecture laid out in the paper by Li, Jun, Reinhard Klein, and Angela Yao. “Learning fine-scaled depth maps from single RGB images.” arXiv preprint (2016).
A tool for generating RGB and depth images of human poses
Published:
This project helps you create random human poses with the help of Makehuman and then generates their rgb and depth data using Blender. The main motivation for this project was scarcity of existing datasets for human body segmentation from depth images. The data generation pipeline is capable of generating different body types and poses and thus creating a varied and large database of depth and segmented images.
Analysis of Iterative Closest Point Algorithm
Published:
This project evaluates several algorithms for iterative closest point algorithm. Iterative closest Point (ICP) is an algorithm employed to minimize the difference between two point clouds given an initial estimate of the relative pose. It is often used to reconstruct 2D or 3D surfaces from different scans, to localize robots and achieve optimal path planning and to register medical scans. ICP has several steps and each step may be implemented in various ways which give rise to a multitude of ICP variants. In our project, we implement and analyze several variants of ICP, comparing them on the basis of execution speed and quality of the result.
Neural Network Visualization using Guided Backprop
Published:
In this project we implemented a web interface for visualizing the working of a protein structure predicting Deep Neural Network using guided backpropagation technique. The goal of this technique is to visualize parts of the input that most activates a given neuron. An effective way of doing this is to visualize gradients of activations(output) with respect to the input.
Development of a system that allows registration of segmented point cloud to patient CT data and provide augmentations
Published:
In this project we are trying to develop a system that will augment the volumetric medical images onto a segmented body part. In order to correctly augment medical information onto the patient it is essential to accurately determine the transformation between the current pose of the patient, the viewing point of the observer, and the associated medical image data. The required transformations can be estimated by registering the surface (skin) of the patient with a surface extracted from the associated volumetric scans.
Natural Language Generation (NLG)
Published:
The goal of this seminar was to explore the classical three stage architecture along with it’s several tasks in detail. A data driven news generation system for automated journalism was also explored.
Aspect-based Sentiment Detection using Deep Neural Networks and Transfer Learning
Published:
A novel approach to jointly detect aspect and the corresponding sentiment in a given sentence using DNN and transfer learning.
Approximate Nearest Neighbor(ANN) Libraries
Published:
A Comparison and benchmarking of different Approximate Nearest Neighbor(ANN) Libraries.
Multilingual Sentence Embeddings
Published:
Undertook a solo research project to implement a multilingual sentence similarity model (de,en,fr). This model performs 3% points better and has 74% smaller embeddings than the current best ebot7 model.
publications
An Evaluation of Progressive Neural Networks for Transfer Learning in Natural Language Processing
Published in Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), 2020
A major challenge in modern neural networks is the utilization of previous knowledge for new tasks in an effective manner, otherwise known as transfer learning. Fine-tuning, the most widely used method for achieving this, suffers from catastrophic forgetting. The problem is often exacerbated in natural language processing (NLP). In this work, we assess progressive neural networks (PNNs) as an alternative to fine-tuning. The evaluation is based on common NLP tasks such as sequence labeling and text classification. By gauging PNNs across a range of architectures, datasets, and tasks, we observe improvements over the baselines throughout all experiments.
Recommended citation: Gerhard Hagerer1†, Abdul Moeed1†, Sumit Dugar1†, Sarthak Gupta1†, Mainak Ghosh1†, Hannah Danner, Oliver Mitevski, Andreas Nawroth, Georg Groh, LREC 2020) https://aclanthology.org/2020.lrec-1.172.pdf
talks
An Introduction to Automatic Speech Recognition
Published:
The goal of this presentation was to give a short introduction about automatic speech recognition. We are mainly focused on the classical ASR pipeline and in that we discussed what audio signals are? We also discussed a common feature extraction technique that is used with audio data and we also talked about other components of the pipeline like Acoustic Model, Language Model and Hypothesis Search. Towards the end we discussed an evaluation metric that is used to compare the performance of various ASR models.
Multi-Domain Dialogue State Tracking
Published:
Last summer our team of data scientists attended the 57th Annual Meeting of the Association for Computational Linguistics (ACL) at Florence. It was a great learning and team bonding experience for all of us. During the conference we learned about this really interesting paper - Transferable Multi-Domain State Generator for Task-Oriented Dialogue Systems. It received two awards in the conference, an outstanding paper award the best paper award at NLP for Conversational AI Workshop.
Multiprocessing vs Multithreading vs Async IO
Published:
In this free Friday talk we discussed about what are the differences between multiprocessing and multithreading in Python. We also talked about what is AsyncIO and how to implement it in Python.
An overview of NLP evaluation metrics and distance/similarity measures
Published:
Evaluation in NLP is an open problem especially when it comes to the evaluation of generated text. It’s a really hard problem and therefore there is no single go to evaluation metric available. What makes this problem even more hard is that metrics need to evaluate competing goals i.e Correctness (quality)/Specicity (diversity).
teaching
Teaching experience 1
Undergraduate course, University 1, Department, 2014
This is a description of a teaching experience. You can use markdown like any other post.
Teaching experience 2
Workshop, University 1, Department, 2015
This is a description of a teaching experience. You can use markdown like any other post.