Hi, I'm Swyam Prakash Singh.

Framework for Set Matching and Bipartite Hypergraph(CATSETMAT)

M.Tech Project
Advisor: Prof. Narashimha Murty and Prof. V. Susheela Devi
Project description

• We are given a bipartite hypergraph and we aim at capturing relations between node pairs from the left and right hyperedges.
• It can be also seen as set matching problem.
• We have created a framework to predict the relation(link) between left and right hyperedges.

Link to Paper

Code

Graph Neural Networks for Text Classification(Text_GAT)

Aug 2020 – Nov2020
Advisor: Prof. Shirish Shevade
Project description

• Aim was to use graph attention networks to classify text documents.
• This was a course project and inspired from work TextGCN by Yao et al. (2019).
• Explored construction of graph from text, convolution over graph, attention mechanism.
• Accuracy: 97 % on R8 dataset, 76.45% on MR dataset, 58.42% on Ohsumed dataset.

Project Link

Text Classification using Tri-char grams

Sep 2020 – Oct 2020
Advisor: Prof. Shirish Shevade
Project description

• Aim was to use tri-char grams for sentence classification using CNN .
• explored different word embedding like Word2vec, Glove.
• technique like CNN over text.
• dataset:SST, accuracy :39.7%

Project Link

Cascading Graph representation Learning with Universal Deep embedding clustering

Mar 2019 – Apr 2019
Advisor: Prof. Ambedkar Dukkipati
Project description

• This was a course project for Machine Learning.
• main aim was to analyse the difference between results on node classification and link prediction obtained by two different methods node2vec and GCN.
• Other task was to use the representation obtain by GCN with Unsupervised deep Embedding Clustering.
• Explored Unsupervised deep embedding clustering, node classification and link prediction.

IISc

Apparent identification (Project AI) using CNN.

Mar 2019 – Apr 2019
Advisor: Prof. M. Narasimha Murty
Project description

• Designed a model which include CNN for classification.
• Dataset: CIFAR-10, Fashion-MNIST
• Accuracy: 66.86%, 90.31%

Project Link

Aggregating Graph Embeddings

May 2019 – June 2019
Advisor: Prof. M. Narasimha Murty
Project description

• Aim was to increase F1-scores for node classification tasks by aggregating embeddings from different mechanisms.
• algorithm used was node2vec, LINE , SDNE, and many more.
• The motive was to exploit the fact that different representation mechanisms capture different characteristics of a graph, so combining them might work better.
• explored various algorithms like Deepwalk, Node2vec, GCN, SDNE, LINE etc.
• Also Classification Algorithms like Random Forests, SVM, Ensemble methods.

IISc

Reconstruction and Classification of MNIST Dataset by K-NN Classifier

Nov 2018
Advisor: Prof. M. Narasimha Murty
Technologies: Python 3.0

Project description
MNIST is a handwritten dataset, originally has 60,000 digits with 784 (28x28) dimensions in its training set. In this assignment, a subset of MNIST dataset has been taken into account for reconstruction task using truncated SVD for different values of d and Reconstruction Error (RMSE) is calculated. Classification task for test points is conducted using K-NN algorithm for various K-values and classification accuracies are reported. This is done as a part of an assignment in Linear Algebra course at the Indian Institute of Science.

IISc

Unsupervised Learning Task of Clustering

Oct 2018
Advisor: Prof. M. Narasimha Murty
Technologies : Python 3.0

Project description
Design and implement unsupervised learning task of clustering similar data points using k-means and spectral clustering algorithms. This project deals with eigenvalues, eigenvectors and one of their numerous applications, namely clustering. K-means and Spectral Clustering have been applied on two different datasets and observed the differences. This was done as a part of assignment in Linear Algebra course at Indian Institute of Science.

IISc