Projects


Model-based Offline Multi-Agent Dialogue Policy Learning

The conventional problem of selecting an action when provided with a state, environment and history (context) falls in the domain of self-play reinforcement learning. Numerous such algorithms learn a dialog policy with the reward function requiring an elaborate design of a comprehensive user simulator and pre-specified user goals. Here, in the implemented method, it is regarded that the user agent can learn with the system agent in a joint/shared fashion. The method involves the concept of role-aware reward decomposition using Hybrid Value Network with the integration of actor-critic framework to maximize the global reward for the policy learner.

Few-shot Unsupervised Discrete Sentence Representation Learning based Dialogue Generation

An unsupervised discrete sentence representation learning method that can integrate with any existing encoder-decoder dialogue model, for interpretable response generation using a minimal amount of data that is not annotated.

A Hybrid Classification Approach using Topic Modeling and Graph Convolutional Networks


A novel multi-class text classification technique that harvests features from two distinct feature extraction methods. Firstly, a structured heterogeneous text graph built based on document-word relations and word co-occurrences is leveraged using a Graph Convolution Network (GCN). Secondly, the documents are topic modeled to use the document-topic score as features into the classification model. The concerned graph is constructed using Point-Wise Mutual Information (PMI) between pair of word co-occurrences and Term Frequency-Inverse Document Frequency (TF-IDF) score for words in the documents for word co-occurrences.

Debunking Fake News by Leveraging Speaker Credibility and BERT Based Model


A novel intuitive approach to exploit data from multiple sources to segregate news into real and fake using contextual embeddings, sequence models with a credibility score for speaker.

The Paperless


PaperLess, an automation PWA, is responsible for auto-filling the required form and forwarding the filled application form to various concerned authorities. The only manual thing in PaperLess is selection of form by the student. After the selection of form, everything is automated including forwarding the application to the concerned authorities. At the end of the process, PaperLess produces the final approved application.

Unsupervised Domain Adaptation for Fingerprint Recognition

The concept of domain adaptation in the absence of labelled training data for a deep learning architecture was implemented by augmenting the given deep neural network with the proposed new gradient reversal layer.

Pore-based Fingerprint Recognition System

A customized deep learning based fingerprint recognition system has been developed using the multitask residual learning based convolutional neural network architecture to extract the fixed length feature representations from a high resolution pore latches.