project / abstract / code

Adversarial attacks during training can strongly influence the performance of multi-agent reinforcement learning algorithms. It is, thus, highly desirable to augment existing algorithms such that the impact of adversarial attacks on cooperative networks is at least bounded. We consider a fully decentralized network, where each agent receives a local re- ward and observes the global state and action. We propose a resilient consensus-based actor-critic algorithm, whereby each agent estimates the team-average reward and value function, and communicates the associated parameter vectors to its immediate neighbors. We show that in the presence of Byzantine agents, whose estimation and communication strategies are completely arbitrary, the estimates of the cooperative agents converge to a bounded consensus value with probability one, provided that there are at most H Byzantine agents in the network that is (2H + 1)-robust. Furthermore, we prove that the policy of the cooperative agents converges with probability one to a bounded neighborhood around a stationary point of their team-average objective function under the assumption that the policies of the adversarial agents asymptotically become stationary.

project / paper / abstract / bibtex

Deep learning has extensively helped us to analyze complicated distributions of data and extract meaningful information from the same. But the fundamental problem still remains—what if there is no or rare occurrence of any instance which needs to be predicted? This is a significant problem in the health-care sector. The primary motivation of the work is to detect the region of anomaly in the context of breast cancer detection. The novelty here lies in designing a zero-shot learning induced Generative Adversarial Network (GAN) based architecture which has the efficacy to detect anomaly and thereby to discriminate the healthy and anomalous images even if the network is trained with the healthy instances only. In addition, this work has also been extended to segmentation of the anomalous region. To be precise, the novelty of this approach lies in the fact that the GAN takes only healthy samples in the training phase, though it is capable of segmenting tumors in the “unseen” anomalous samples explaining each pixel’s contribution to the model’s prediction. Experiments have been conducted on the mini MIAS breast cancer dataset and significantly appreciable results have been obtained.

@Inbook{Mukherjee2021,
author="Mukherjee, Preeti and Pal, Mainak and Ghosh, Lidia and Konar, Amit",
editor="Pedrycz, Witold and Chen, Shyi-Ming",
title="A Generative Model Based Approach for Zero-Shot Breast Cancer Segmentation Explaining Pixels' Contribution to the Model's Prediction",
bookTitle="Interpretable Artificial Intelligence: A Perspective of Granular Computing",
year="2021",
publisher="Springer International Publishing",
address="Cham",
pages="401--425",
isbn="978-3-030-64949-4",
doi="10.1007/978-3-030-64949-4_13",
url="https://doi.org/10.1007/978-3-030-64949-4_13"
}

project / code / abstract / bibtex / presentation

Clustering aims at grouping of objects or data-points based on a certain measure of similarity. Existing clustering algorithms estimate the measure of similarity of expected data-points to fall in a cluster with respect to presumed or computed cluster centroids. Such approach of distance measure between cluster centroid and possible data points to lie in cluster often results in misclustering, particularly for points equidistant to multiple cluster centroids. This paper offers an interesting solution to this problem by quantization of the attributes of the preferred cluster centroids and then checking the existence of the respective attributes of data-points within the quantized intervals for possible inclusion of data-point in the cluster. This approach, referred to as vector quantization offers additional merits of clustering at different user- defined resolutions of data-points of varying local density. Experiments undertaken confirm the superior performance of the proposed clustering over the state-of-art algorithms with respect to Jaccard coefficient on breast cancer dataset.

@Inbook{Pal2020,
author="Pal, Mainak and Mukherjee, Preeti and Konar, Amit",
editor="Gunjan, Vinit Kumar and Senatore, Sabrina and Kumar, Amit and Gao, Xiao-Zhi and Merugu, Suresh",
title="Multi-resolution Hierarchical Clustering by Vector Quantization",
bookTitle="Advances in Cybernetics, Cognition, and Machine Learning for Communication Technologies", year="2020",
publisher="Springer Singapore",
address="Singapore",
pages="243--249",
isbn="978-981-15-3125-5",
doi="10.1007/978-981-15-3125-5_26",
url="https://doi.org/10.1007/978-981-15-3125-5_26"
}

project / code / abstract / bibtex

This paper describes our system submissions as part of our participation (team name: JU_ETCE_17_21) in the SemEval 2019 shared task 6: “OffensEval: Identifying and Catego- rizing Offensive Language in Social Media”. We participated in all the three sub-tasks: i) Sub-task A: offensive language identification, ii) Sub-task B: automatic categorization of of- fense types, and iii) Sub-task C: offense target identification. We employed machine learn- ing as well as deep learning approaches for the sub-tasks. We employed Convolutional Neural Network (CNN) and Recursive Neu- ral Network (RNN) Long Short-Term Memory (LSTM) with pre-trained word embeddings. We used both word2vec and Glove pre-trained word embeddings. We obtained the best F1- score using CNN based model for sub-task A, LSTM based model for sub-task B and Lo- gistic Regression based model for sub-task C. Our best submissions achieved 0.7844, 0.5459 and 0.48 F1-scores for sub-task A, sub-task B and sub-task C respectively.

@inproceedings{mukherjee-etal-2019-ju,
title = "{JU}{\_}{ETCE}{\_}17{\_}21 at {S}em{E}val-2019 Task 6: Efficient Machine Learning and Neural Network Approaches for Identifying and Categorizing Offensive Language in Tweets",
author = "Mukherjee, Preeti and Pal, Mainak and Banerjee, Somnath and Naskar, Sudip Kumar",
booktitle = "Proceedings of the 13th International Workshop on Semantic Evaluation",
month = jun,
year = "2019",
address = "Minneapolis, Minnesota, USA",
publisher = "Association for Computational Linguistics",
url = "https://www.aclweb.org/anthology/S19-2118",
doi = "10.18653/v1/S19-2118",
pages = "662--667",
}

project / preprint / abstract / bibtex

Zero-shot Learning (ZSL) is a transfer learning technique which aims at transferring knowledge from seen classes to unseen classes. This knowledge transfer is possible because of underlying semantic space which is common to seen and unseen classes. Most existing approaches learn a projection function using labelled seen class data which maps visual data to semantic data. In this work, we propose a shallow but effective neural network-based model for learning such a projection function which aligns the visual and semantic data in the latent space while simultaneously making the latent space embeddings discriminative. As the above projection function is learned using the seen class data, the so-called projection domain shift exists. We propose a transductive approach to reduce the effect of domain shift, where we utilize unlabeled visual data from unseen classes to generate corresponding semantic features for unseen class visual samples. While these semantic features are initially generated using a conditional variational auto-encoder, they are used along with the seen class data to improve the projection function. We experiment on both inductive and transductive setting of ZSL and generalized ZSL and show superior performance on standard benchmark datasets AWA1, AWA2, CUB, SUN, FLO, and APY. We also show the efficacy of our model in the case of extremely less labelled data regime on different datasets in the context of ZSL.

@misc{gune2020generative,
title={Generative Model-driven Structure Aligning Discriminative Embeddings for Transductive Zero-shot Learning},
author={Omkar Gune and Mainak Pal and Preeti Mukherjee and Biplab Banerjee and Subhasis Chaudhuri},
year={2020},
eprint={2005.04492},
archivePrefix={arXiv},
primaryClass={cs.CV}
}