SSLAM is a self-supervised deep learning framework designed to generate labels while minimizing the overhead associated with tabular data annotation. SSLAM learns valuable representations from unlabeled data that are applied to the downstream task of label generation generation by utilizing two pretext tasks with a novel log-cosh loss function.

We aim to implement a robust and efficient machine learning-based solution for restoring motor functions in patients with spinal cord injury using Epidural Electrical Stimulation. This thesis examines the problem of neural drift and the possibility of using meta-learning to build an adaptive algorithm to tackle this drift.

We propose to measure the data drift that takes place when new data kicks in so that one can adaptively re-train the models whenever re-training is required. In addition to that, we generate various sentence and dataset-level explanations to capture why a given payload text has drifted.

The Abstraction and Reasoning Corpus (ARC) was introduced by François Chollet as a benchmark to measure AI skill acquisition on unknown tasks, with the constraint that only a handful of demonstrations are shown to learn a complex task. We show how modern neural networks fail ARC and test the idea of using meta-learning to learn priors useful to solve ARC.

This work explores the contexts associated with errors in the automatic transcription of spontaneous speech and its effects on current State-of-the-art methods for dementia detection. We attempt to build a purely acoustic solution based on Vision Transformers for dementia detection.

We develop a model miming the Visual Hierarchy observed in animals for decoding fMRI to classify images. We show that hierarchical processing of fMRI data from different parts of the animal visual system aids classification.

In this work, we study the problems dealing with OOD data for skin cancer classification. We compare the performance of Virtual Outlier Synthesis, a recent work published at ICLR 2022, with other State-of-the-art OOD detection methods. We further show how using another inference method help achieve better results than VOS. We also study the effects of different types of OOD data on our method.

An SVM-based model which can recognize COVID coughs using short audio cough recordings. An attempt at the INTERSPEECH 2021 Computational Paralinguistics Sub-Challenge .

A ResNet model which detects important facial key-points in images. Facial key-point detection plays a pivotal role in tasks such as Emotion Analysis.

An LSTM-based model which classifies Reddit posts to their appropriate flairs. Used GloVe for word embedding.