Published in IEEE International Symposium on Devices Circuits and Systems (ISDCS), 2018, 2018
The objective of this project is to utilize a LASER INDUCED BREAKDOWN SPECTROGRAPHY (LIBS) instrument for soil mineral composition analysis in agriculture, facilitated by an Android-controlled platform.
The objective of this work is to prevent the generation of outputs containing undesired features from a pre-trained GAN where the underlying training data set is inaccessible.
The primary goal of this study is twofold: first, to elucidate the relationship between filtering and unlearning processes, and second, to formulate a methodology aimed at mitigating the display of undesirable outputs generated from models characterized as black-box systems.
The principal aim of this study is to formulate a methodology tailored for the purposeful elimination of information linked to a specific class of data from a pre-trained classification network.
Published in NeurIPS 2024 Workshop SafeGenAi, 2024
The principal objective of this work is to propose a machine unlearning methodology that can prevent the generation of outputs containing undesired features from a pre-trained diffusion model
Over the past century, the field of artificial intelligence (AI) has experienced remarkable progress, primarily driven by the aspiration to enable machines to learn in a manner akin to humans. This pursuit has been rooted in a fundamental question: “how” can machines be effectively trained to perform specific tasks? It seems like the answer is the optimization of some specific loss functions, typically relying on carefully curated datasets.
Generative Models output unwanted and even harmful images. To tackle this problem recent line of works explores how to unlearn so that it stops generate these output. On the other hand when these models are used as Block-Box(i.e. the models architechture, parameters and underlying dataset are unknown) in several other downstream task. In this situation unlearning the model becomes harder and only blocking gives pratical way to tackle the problem of stop showing undesired outputs. In this work we devise a way to block the generative models outputs in latent space.
Due to data bias and undesired data present in the dataset, generative models output unintended output which are harmful. To tackle this problem, we unlearn the generative models with the help of the user feedback. In our work using the feedback from the user we try to adapt our GAN so that it only produces the unintened images(negetive images). From there we try to purturb out initial model in such a way that it remain far from adapted model in parameter space.
Published in IEEE International Symposium on Devices Circuits and Systems (ISDCS), 2018, 2018
The objective of this project is to utilize a LASER INDUCED BREAKDOWN SPECTROGRAPHY (LIBS) instrument for soil mineral composition analysis in agriculture, facilitated by an Android-controlled platform.
The objective of this work is to prevent the generation of outputs containing undesired features from a pre-trained GAN where the underlying training data set is inaccessible.
The primary goal of this study is twofold: first, to elucidate the relationship between filtering and unlearning processes, and second, to formulate a methodology aimed at mitigating the display of undesirable outputs generated from models characterized as black-box systems.
The principal aim of this study is to formulate a methodology tailored for the purposeful elimination of information linked to a specific class of data from a pre-trained classification network.
Published in NeurIPS 2024 Workshop SafeGenAi, 2024
The principal objective of this work is to propose a machine unlearning methodology that can prevent the generation of outputs containing undesired features from a pre-trained diffusion model
Pattern Recognition and Neural Network(E2-233): This is an introductory course for machine learning and deep learning. The course contents can be found here.
Advanced Deep Representation Learning(E2-333): This course is an advance course in representation learning. This course broadly covers recent research works in generative modeling, continual learning and meta learning. The course contents can be found here.