Publications
Explaining OOD detection using two-step counterfactual generation
An approach for generating counterfactuals for OOD data points to better understand why a data point is categorized as OOD.
Nawid Keshtmand, Raul Santos-Rodriguez, Jonathan Lawry
FACCT 2023 In Review [Paper] [Code]
Typicality-based point OOD detection
A typicality-based approach for point OOD detection which utilizes the nearest neighbours of test data points.
Nawid Keshtmand, Raul Santos-Rodriguez, Jonathan Lawry
[Paper] [Code]
Understanding the properties and limitations of contrastive learning for Out-of-Distribution detection
Examining the effect of supervision on the properties and OOD detection performance of contrastive learning algorithms.
Nawid Keshtmand, Raul Santos-Rodriguez, Jonathan Lawry
ICPR 2022 Workshop: SSL - Theories, Applications, and Cross Modality for Self-Supervised Learning Models [Paper] [Code]
Projects
Literature notes
Ongoing notes on the various papers and textbooks I read.
[Link]
Contrastive Learning with autoregressive decoder
Implemented a contrastive learning algorithm (AMDIN) which uses a pixelCNN decoder.
[Code]
Multimodal VQ-VAE
Implemented a multi-modal VQ-VAE which takes in RGB and infrared image data into a joint latent space which can then be used to decode the differnet modalites.[Code]
Adversarial Learning in a Predator-Prey game
Training a predator and prey in a simultaenosuly using reinforcement learning and examining the learning dynamics of the agents by fitting decision trees to the policies learnt by the different agents.
Nawid Keshtmand, Raul Santos-Rodriguez, Jonathan Lawry
MSc Disseration: University of Bristol and University of the West of England [Paper]
Development of a motion and muscle sensing human-machine interface
Training Linear Discriminant analysis and Support Vector Machine classifiers to classify Mechanomyography signals from different hand gestures. Using the trained classifiers to develop a game to aid stroke victims to regain functionality in their hands.
Nawid Keshtmand, Ravi Vaidyanathan
MSc Thesis: Imperial College London [Paper]