KIRAN PUROHIT

Trying to learn about deep learning faster than deep learning can learn about me.

EDUCATION

PhD student in CSE at IIT Kharagpur.

RESEARCH INTERESTS

Machine Learning, Deep Learning, Computer Vision, Image Processing

Scopus Publications

A Data-Driven Defense Against Edge-Case Model Poisoning Attacks on Federated Learning
Kiran Purohit, Soumi Das, Sourangshu Bhattacharya, and Santu Rana
IOS Press
Federated Learning systems are increasingly subjected to a multitude of model poisoning attacks from clients. Among these, edge-case attacks that target a small fraction of the input space are nearly impossible to detect using existing defenses, leading to a high attack success rate. We propose an effective defense using an external defense dataset, which provides information about the attack target. The defense dataset contains a mix of poisoned and clean examples, with only a few known to be clean. The proposed method, DataDefense, uses this dataset to learn a poisoned data detector model which marks each example in the defense dataset as poisoned or clean. It also learns a client importance model that estimates the probability of a client update being malicious. The global model is then updated as a weighted average of the client models’ updates. The poisoned data detector and the client importance model parameters are updated using an alternating minimization strategy over the Federated Learning rounds. Extensive experiments on standard attack scenarios demonstrate that DataDefense can defend against model poisoning attacks where other state-of-the-art defenses fail. In particular, DataDefense is able to reduce the attack success rate by at least ∼ 40% on standard attack setups and by more than 80% on some setups. Furthermore, DataDefense requires very few defense examples (as few as five) to achieve a near-optimal reduction in attack success rate.


A Greedy Hierarchical Approach to Whole-Network Filter-Pruning in CNNs




EXPLORA: Efficient Exemplar Subset Selection for Complex Reasoning




Accurate and Efficient Channel pruning via Orthogonal Matching Pursuit
Kiran Purohit, Anurag Parvathgari, Soumi Das, and Sourangshu Bhattacharya
ACM
The deeper and wider architectures of recent convolutional neural networks (CNN) are responsible for superior performance in computer vision tasks. However, they also come with an enormous model size and heavy computational cost. Filter pruning (FP) is one of the methods applied to CNNs for compression and acceleration. Various techniques have been recently proposed for filter pruning. We address the limitation of the existing state-of-the-art method and motivate our setup. We develop a novel method for filter selection using sparse approximation of filter weights. We propose an orthogonal matching pursuit (OMP) based algorithm for filter pruning (called FP-OMP). We also propose FP-OMP Search, which address the problem of removal of uniform number of filters from all the layers of a network. FP-OMP Search performs a search over all the layers with a given batch size of filter removal. We evaluate both FP-OMP and FP-OMP Search on benchmark datasets using standard ResNet architectures. Experimental results indicate that FP-OMP Search consistently outperforms the baseline method (LRF) by nearly . We demonstrate both empirically and visually, that FP-OMP Search prunes different number of filters from different layers. Further, timing profile experiments show that FP-OMP improves over the running time of LRF.


Measuring the Degree of Suitability of Edge Detection Operators Prior to an Application
Abhishek Kesarwani, Kiran Purohit, Mamata Dalui, and Dakshina Ranjan Kisku
IEEE
Unlike image restoration, image enhancement techniques are found to be subjective in nature as the appearance of an output image depends upon human perception. Hence, it is very difficult to determine the appropriateness of image enhancement techniques including edge detection operators prior to an application. This paper makes use of regression models to determine the suitability of edge detection operators before operators to be executed. With the existing operators, a novel Hybrid technique is used in the evaluation. The Hybrid detector is designed by combining Canny and Sobel operators with the gradient of texton image. This approach estimates a model as an objective function to determine the degree of proximity or suitability of edge detection operators under regression constraints on two publicly available databases, viz. the BSDS300 and the Multi-cue. The experimental results exhibit that the Hybrid edge detector outperforms other operators for measuring the proximity for appropriateness.

RECENT SCHOLAR PUBLICATIONS

EXPLORA: Efficient Exemplar Subset Selection for Complex Reasoning
K Purohit, R Devalla, KM Yerragorla, S Bhattacharya, A Anand
Empirical Methods in Natural Language Processing (EMNLP-main track) 2024



A Greedy Hierarchical Approach to Whole-Network Filter-Pruning in CNNs
K Purohit, AR Parvathgari, S Bhattacharya
Transactions on Machine Learning Research (TMLR) 2024



A Data-Driven Defense against Edge-case Model Poisoning Attacks on Federated Learning
K Purohit, S Das, S Bhattacharya, S Rana
27th European Conference on Artificial Intelligence (ECAI) 2024



Accurate and Efficient Channel pruning via Orthogonal Matching Pursuit
K Purohit, A Parvathgari, S Das, S Bhattacharya
Proceedings of the Second International Conference on AI-ML Systems, 1-8 2022



Covid-19 detection on chest x-ray and ct scan images using multi-image augmented deep learning model
K Purohit, A Kesarwani, D Ranjan Kisku, M Dalui
International Conference on Mathematics and Computing (ICMC), 395-413 2022



Reproducibility Report: Contextualizing Hate Speech Classifiers with Post-hoc Explanation
K Purohit, O Iqbal, A Mullick
arXiv preprint arXiv:2105.11412 2021



Measuring the Degree of Suitability of Edge Detection Operators Prior to an Application
A Kesarwani, K Purohit, M Dalui, DR Kisku
2020 IEEE Applied Signal Processing Conference (ASPCON), 128-133 2020

MOST CITED SCHOLAR PUBLICATIONS

Covid-19 detection on chest x-ray and ct scan images using multi-image augmented deep learning model
K Purohit, A Kesarwani, D Ranjan Kisku, M Dalui
International Conference on Mathematics and Computing (ICMC), 395-413 2022
Citations: 84



Measuring the Degree of Suitability of Edge Detection Operators Prior to an Application
A Kesarwani, K Purohit, M Dalui, DR Kisku
2020 IEEE Applied Signal Processing Conference (ASPCON), 128-133 2020
Citations: 6



A Data-Driven Defense against Edge-case Model Poisoning Attacks on Federated Learning
K Purohit, S Das, S Bhattacharya, S Rana
27th European Conference on Artificial Intelligence (ECAI) 2024
Citations: 4



Accurate and Efficient Channel pruning via Orthogonal Matching Pursuit
K Purohit, A Parvathgari, S Das, S Bhattacharya
Proceedings of the Second International Conference on AI-ML Systems, 1-8 2022
Citations: 3



A Greedy Hierarchical Approach to Whole-Network Filter-Pruning in CNNs
K Purohit, AR Parvathgari, S Bhattacharya
Transactions on Machine Learning Research (TMLR) 2024
Citations: 1