Projects

Hateful Memes Classification

Implemented a multi-modal deep learning model to classify memes as Hateful and Non-Hateful. Used an ensemble of VisualBert, ERNIE-VIL, DeVLBERT, OSCAR, UNITER and LXMERT to make the prediction using Mixture of Experts. The proposed method ranked 12th on the challenge hosted by Facebook AI. [ Github ] [ Slides ]

Hybrid Machine Translation

A Machine Translation model that utilizes the strengths of both Statistical and Neural MT. The neural sequence-to-sequence archtitecture is enhanced by the phrase-table constructed using Phrase-Based SMT. The generated unknown tokens are replaced with the translations of the aligned source phrases. [ Github ]

Learning Bilingual Word Embeddings with Minimal Bilingual Data

An unsupervised method to learn bilingual word embeddings involving two languages. Transformed the word embeddings in individual languages to the bilingual embedding space, using neural networks. Optimized the network parameters using self training. Visualising the embedding space shows that the embeddings of words in different languages with same meaning were close. [ Github ]

Deep Crypt

Designed Bayesian and Hyperplane classifiers for encrypted data. The ML model on the server runs inference on the encrypted data provided by the client. The classifier is built using the DGK, Paillier, and Goldwasser Macali cryptosystems, which help to implement the argmax, comparison, and dot product operators for encrypted inputs.[ Github ]