Year-2023 http://repository.iiitd.edu.in/xmlui/handle/123456789/1317 2026-04-11T15:40:17Z 2026-04-11T15:40:17Z Ananya Anand, Saket (Advisor) http://repository.iiitd.edu.in/xmlui/handle/123456789/1371 2023-12-20T22:00:25Z 2023-05-01T00:00:00Z

Visual monitoring for wildlife : detection and re-identification Ananya; Anand, Saket (Advisor) Visual wildlife monitoring of animals requires detection for species-level categorization and re-identification (Re-ID) for population estimation of an individual species. Traditionally, the monitoring is done via GPS collars which are invasive, but the advent of camera traps has given a convenient, non-invasive and inexpensive alternate method for monitoring of wild animals. This camera-trap image data can be used with AI-based algorithms for detecting animal presence, species-level categorization, as well as individual identification or animal biometrics for certain species. To this end, we have developed the Deep Learning (DL) based species categorization module for the Camera Trap Data Repository and Analysis Tool (CaTRAT), which was used by theWildlife Institute of India (WII) for processing the camera trap images during the All India Tiger Estimation 2022. Beyond species-level segregation, deep learning approaches have also shown good performance for re-identification tasks. However, these methods often fall short, when encountered with fine grained patterned species like tigers and leopards, both in terms of performance as well as interpretability. This limits their usability by conservation officials and practitioners. In this work, we propose an end-to-end network to learn feature representations, keypoints, and their descriptors. The keypoints enable the model to: a) learn better discriminative feature representations and b) focus on salient regions (patterns) of the image. It is important to note that while training, we don’t have groundtruth keypoint and descriptor annotations but only the label information. A pre-trained, DenseNet model is fine-tuned by a classification cross-entropy loss regularized by a pairwise Jensen-Shannon divergence. Further, feature map normalization regularizes the descriptor loss. The fusion of the keypoint attention feature map in the network helps focus on regions important for animal biometrics. We then evaluate the efficacy of our model on two datasets of patterned species, namely Amur Tigers and Leopards, under different biometric evaluation protocols: mAP, top-1, top-5, closed-set identification, open-set identification, and verification.

2023-05-01T00:00:00Z Jain, Tanisha Prasad, Ranjitha (Advisor) http://repository.iiitd.edu.in/xmlui/handle/123456789/1368 2023-12-20T22:00:23Z 2023-07-01T00:00:00Z

Optimizing medical costs and resource utilization using causal inference and explaining the model’s predictions Jain, Tanisha; Prasad, Ranjitha (Advisor) Machine learning (ML) models that accurately predict treatment effects and related healthcare costs can bring significant efficiencies in the healthcare industry. These models could help reduce fatalities resulting from incorrect treatment allocations and contribute to cost-effective healthcare delivery, which is crucial for both developed and developing nations. However, existing literature does not provide any comprehensive framework that effectively estimates both treatment effect and the overall medical expenditure while considering individual treatment effects. To address this gap, we propose CFMedNet, a pioneering counterfactual inference framework that jointly estimates treatment effect and medical costs. This novel framework not only predicts the potential impact and costs of a given treatment and its counterfactual but also provides individual treatment effects for both outcomes. However, a considerable challenge in the adoption of such ML models in healthcare is their perceived ’black box’ nature due to limited transparency in decision-making processes. Since medical professionals bear the responsibility for their decisions, it’s crucial to have Explainable AI (XAI) models, especially in sensitive domains like healthcare. As an innovative contribution, we introduce a post-hoc explainer, GMM-LIME, specifically designed for multi-output causal inference based counterfactual neural networks. This explainer offers crucial explanations and interpretations of our proposed model, thereby improving its transparency and applicability. This dual contribution of a comprehensive estimation framework and in-depth explanatory tools, holds great potential to significantly progress personalized healthcare, balancing economic efficiency with treatment efficacy. Our work represents an integration of Causal Inference, Deep Learning, and XAI, with results obtained from a semi-synthetic dataset.

2023-07-01T00:00:00Z Chatterjee, Sayantika Majumdar, Angshul (Advisor) http://repository.iiitd.edu.in/xmlui/handle/123456789/1366 2023-12-20T22:00:16Z 2023-05-01T00:00:00Z

A computational approach to assist healthcare professionals in selecting antibacterial drugs to treat bacterial infections Chatterjee, Sayantika; Majumdar, Angshul (Advisor) The reproducibility of experiments has been a long-standing obstruction for farther scientific evolution. Computational methods are being involved to accelerate and to economize drug discovery and the development process. In this work several computational models using matrix completion techniques including matrix factorization, deep matrix factorization, binary matrix completion and graph regularised techniques (graph regularised deep matrix factorisation, graph regularised matrix factorization, graph regularised binary matrix completion and graph regularised matrix completion) have been proposed to predict bacteria-drug association. Here drug-bacteria association matrix is formed. Along with it we gather similarity information using the chemical structure of drugs and genome-genome distance calculator Meier-Kolthoff et al. (2022) for bacteria. Using several matrix completion tools, the bacteria-drug association data and similarity data, the present study predicts the set of best possible drugs corresponding to each bacteria in the database. The graph regularised techniques consider the drugbacteria association matrix along with the similarity information for prediction. To evaluate robustness of the model, cross validation settings on different scenarios have been adopted on the training data. The AUC-AUPR metric is being reported corresponding these scenarios and association between drug-bacteria is being predicted with the help of various graph and non graph regularised methods. The result produced by graph regularised methods are better compared to non graph regularised methods. Hence it can be concluded that the graph regularised methods predicts the association data well. We anticipate that this work will provide opportunities to develop drugs for newly discovered bacteria and, conversely, enable the identification of potential bacteria targets for existing drugs

2023-05-01T00:00:00Z Arora, Kanika Prasad, Ranjitha (Advisor) http://repository.iiitd.edu.in/xmlui/handle/123456789/1364 2023-12-19T22:00:29Z 2023-07-01T00:00:00Z

Personalized unsupervised federated autoencoders Arora, Kanika; Prasad, Ranjitha (Advisor) An immense amount of data is generated daily usingmodern technologies in autonomous vehicles, IoT, smart grids, etc. But unfortunately, this data generated at the edge cannot be used for any machine learning model training due to privacy concerns or expensive computational costs. The data must be stored at the central server for any machine learning process to occur. To overcome the problem faced due to the traditional machine learning approach decentralized technique called Federated Learning started to gain popularity. Federated Learning allows multiple clients in the network to collaborate and learn a global machine learning model, which can be passed to all the edge devices for locally training a model while maintaining privacy since data is present at the edge device. The availability of annotated data is one of the challenges of supervised federated learning. Moreover, sometimes it is difficult for a global model to perform well for all the clients in the network due to the presence of heterogeneous data. In this thesis, a novel Personalized unsupervised Federated AutoEncoders,pFedAE, is proposed with the main motivation that local and global latent space representations of all the clients in the network. The optimisation framework of the autoencoder is divided into two parts global and per-client local optimisation frameworks. We have adopted two evaluation strategies to evaluate the latent space representation at both global and local levels. We demonstrated that pFedAE under both evaluation strategies performed better than the other baselines. pFedAE, most importantly, leads to faster convergence, is scalable for different numbers of clients, is effective in varying data distribution across clients and is robust to different numbers of local epochs. Using t-SNE projections and angle histogram plots, a comparison of the pFedAE with other baselines for latent space is also demonstrated in the later part of the thesis.

2023-07-01T00:00:00Z