SAiDL Spring Assignment 2022

This repository is the Spring Induction Assignment of Society for Artificial Intelligence and Deep Learning for the year 2022.

The attempted sections are:

1. Natural Language Processing
2. Core ML
3. Literature Review

Natural Language Processing

Aim

To implement Mixup data augmentation technique on the TREC dataset.

Implementation

Created a baseline BiLSTM model, with the following results:

Epochs: 15 Train Loss: 0.065 | Train Acc: 97.71% Test. Loss: 0.528 | Test. Acc: 88.59%

Then, created a BiLSTM with MixText algorithm for the forward function, giving the following results:

Epochs: 15 Train Loss: 1.002 Test. Loss: 0.614 | Test. Acc: 84.81%

Core ML

To train a model using Markov Chain Monte Carlo (MCMC) sampling on noisy XOR dataset

Implementation

Created a BNN sampled with Metropolis-Hastings algorithm.

Epochs: 10000 Accuracy Score: 0.911

Also, created a BNN sampled with random sampler for baseline.

Epochs: 1000 Accuracy Score: 0.692

Literature Review

Writing a review of the paper "Induction of Decision Trees".