Natarajan Subramanyam | PES University (original) (raw)

Papers by Natarajan Subramanyam

2021 Smart Technologies, Communication and Robotics (STCR), Oct 9, 2021

Traditional image super resolution centered around purely mathematical models are capable of crea... more Traditional image super resolution centered around purely mathematical models are capable of creating gradient based textures, but fail to render the specific lineaments that would be expected in a realistically upscaled image. This is especially problematic in scenarios involving images of subjects whose recognition is reliant on the presence of specific characteristics, for example, faces. In this paper, we describe a deep learning model that is capable of generating an 8x upscaled photo from a low resolution image of a face. The underlying model is based on the SRGAN architecture that deviates from the conventional GAN approach of the adversarial back and forth between generator and discriminator by incorporating an added content loss whose value is dependent on the detection of the natural features in the generated image by a pre-trained VGG model. The model is trained on the Celeb Faces Attributes dataset with over 1,00,000 data points and can produce upscaled images that are realistic with a coherent presence of the natural attributes of a face.

Communications in computer and information science, 2020

State-of-the-art techniques that perform reasoning over large knowledge graphs incorporate a path... more State-of-the-art techniques that perform reasoning over large knowledge graphs incorporate a path-finder and a path-reasoner. The path-reasoner is a learning agent that performs the inference task by producing a relation. The path-finder is usually the knowledge graph environment that moves the agent to the next-hop entity. While the path-reasoner can work on a continuous state space, the knowledge graph environment on which it is being trained, operates on a discrete state space. This restricts the agent from deducing implicit paths. In this paper, a novel path-finder called Stochastic Dynamic Environment for Knowledge Graphs (SDE-KG) has been proposed. SDE-KG is a meta-framework that can be combined with many embedding methods to create a continuous function of the knowledge graph environment which may facilitate smarter multi-hop reasoning.

Discovering interesting, implicit knowledge and general relationships in geographic information d... more Discovering interesting, implicit knowledge and general relationships in geographic information databases is very important to understand and to use the spatial data. Spatial clustering has been recognized as a primary data mining method for knowledge discovery in spatial databases. In this paper, we have analyzed an efficient method for the fusion of the outputs of the various clusterers, with less

Low-dimensional embeddings for knowledge graph entities and relations help preserve their latent ... more Low-dimensional embeddings for knowledge graph entities and relations help preserve their latent semantics while enabling computation efficiency. These embeddings are often used to perform tasks such as knowledge graph completion, question answering and inference. Knowledge graph embedding methods aid in the representation of entities and relationships of a knowledge graph in continuous vector spaces. However, most existing techniques ignore the inherent hierarchical structure of entities present in the knowledge graph, defined by ontological relationships between entity types. This paper introduces a novel score function called GrCluster that helps fill that gap. GrCluster is a simple, intuitive and efficient scoring function that incorporates the entity hierarchical correlation into existing knowledge graph embeddings. The effectiveness of GrCluster is demonstrated by integrating it into several well known embedding models. The experimental results show consistent improvements across metrics and embedding models for the tasks of entity prediction and triplet classification.

Advances in intelligent systems and computing, Dec 14, 2018

Question answering (QA) is a field of Natural Language Processing that deals with generating answ... more Question answering (QA) is a field of Natural Language Processing that deals with generating answers automatically to questions asked to a system. It can be categorized into two types—open-domain and closed-domain QA. Open-domain QA can deal with questions about anything, whereas closed-domain QA deals with questions in a specific domain. In our work, we use the architectures of LSTM and memory networks to perform closed-domain question answering and compare the performances of the two. LSTMs are specialized RNNs that can remember necessary data and forget the irrelevant bits. Since data in QA consist of stories and questions based on them, this model seems appropriate, with the ability to handle long sequences. On the other hand, memory networks provide an architecture where there is a provision to store the information learnt by the system in an explicit memory component, rather than just as weight matrices. This also seems like an architecture well-suited to question answering. We implement each model and train it on the Facebook bAbi dataset. This dataset is specifically generated for the purpose of evaluating QA systems on the twenty prerequisite toy bAbi tasks. Each dataset corresponds to one task and checks whether the model is able to perform chaining, counting, answer with single and multiple supporting facts, understand relations, directions, etc. Based on the performances of each model on the bAbi tasks, we perform a comparative study of the two.

arXiv (Cornell University), May 6, 2022

In India, post demonetization exercise in 2016, digital payments have become extremely popular. A... more In India, post demonetization exercise in 2016, digital payments have become extremely popular. Among them, the volume of transactions using Paytm wallets and UPI (Unified Payment Interface) have grown manyfold. The lockdowns due to COVID-19 Pandemic have furthered this trend. Side by side, crypto-currencies such as bitcoin are also gaining traction. Many countries are considering issuing a Digital Currency via their Central Banks. In this paper, we propose a novel Decentralized Digital Currency System (DDCS) that makes use of Merkle Hash-Trees as Authenticated Data Structures. DDCS uses a Ledger-less, distributed, peer-to-peer architecture. We name the proposed currency δ-Money. δ-Money is intended as a replacement for physical currency and has in-built security features that rival crypto-currencies. Transactions using δ-Money happen in a disintermediated manner but with post-facto reconciliation. In place of Central Bank-issued Digital Currency (CBDC), we envisage a scenario where multiple Payment Banks issue digital currencies that have stable valuations without being subject to either volatility or perennial devaluation.

arXiv (Cornell University), Jan 16, 2021

Generalization is the ability of a model to predict on unseen domains and is a fundamental task i... more Generalization is the ability of a model to predict on unseen domains and is a fundamental task in machine learning. Several generalization bounds, both theoretical and empirical have been proposed but they do not provide tight bounds. In this work, we propose a simple yet effective method to predict the generalization performance of a model by using the concept that models that are robust to augmentations are more generalizable than those which are not. We experiment with several augmentations and composition of augmentations to check the generalization capacity of a model. We also provide a detailed motivation behind the proposed method. The proposed generalization metric is calculated based on the change in the model's output after augmenting the input. The proposed method was the first runner up solution for the competition "Predicting Generalization in Deep Learning".

arXiv (Cornell University), Nov 24, 2022

Image inpainting is the process of regenerating lost parts of the image. Supervised algorithm-bas... more Image inpainting is the process of regenerating lost parts of the image. Supervised algorithm-based methods have shown excellent results but have two significant drawbacks. They do not perform well when tested with unseen data. They fail to capture the global context of the image, resulting in a visually unappealing result. We propose a novel self-supervised learning framework for image-inpainting: Weighted Self-Supervised Learning (WSSL) to tackle these problems. We designed WSSL to learn features from multiple weighted pretext tasks. These features are then utilized for the downstream task, image-inpainting. To improve the performance of our framework and produce more visually appealing images, we also present a novel loss function for image inpainting. The loss function takes advantage of both reconstruction loss and perceptual loss functions to regenerate the image. Our experimentation shows WSSL outperforms previous methods, and our loss function helps produce better results.

Large scale video retrieval is a field of study with a lot of ongoing research. Most of the work ... more Large scale video retrieval is a field of study with a lot of ongoing research. Most of the work in the field is on video retrieval through text queries using techniques such as VSE++. However, there is little research done on video retrieval through image queries, and the work that has been done in this field either uses image queries from within the video dataset or iterates through videos frame by frame. These approaches are not generalized for queries from outside the dataset and do not scale well for large video datasets. To overcome these issues, we propose a new approach for video retrieval through image queries where an undirected graph is constructed from the combined set of frames from all videos to be searched. The node features of this graph are used in the task of video retrieval. Experimentation is done on the MSR-VTT dataset by using query images from outside the dataset. To evaluate this novel approach P@5, P@10 and P@20 metrics are calculated. Two different ResNet models namely, ResNet-152 and ResNet-50 are used in this study.

Springer eBooks, Oct 2, 2021

CSI Transactions on ICT

The information systems have been extremely useful in managing businesses, enterprises, and publi... more The information systems have been extremely useful in managing businesses, enterprises, and public institutions such as government departments. But today's challenges are increasingly about managing ecosystems. Ecosystem is a useful paradigm to better understand a variety of domains such as biology, business, industry, agriculture, and society. In this paper, we look at the Indian Agricultural ecosystem. It is a mammoth task to assimilate the information for the whole ecosystem consisting of consumers, producers, workers, traders, transporters, industry, and Government. There are myriad interventions by the state and the central Governments, whose efficacy is difficult to track and the outcomes hard to assess. A policy intervention that helps one part of the ecosystem can harm the other. In addition, sustainability and ecological considerations are also extremely important. In this paper, we make use of the Knowledge-based Tantra Social Information Management Framework to analyze the Indian Agricultural Ecosystem and build related Knowledge Graphs. Our analysis spans descriptive, normative, and transformative viewpoints. Tantra Framework makes use of concepts from Zachman Framework to manage aspects of social information through different perspectives and concepts from Unified Foundational Ontology (UFO) to represent interrelationships between aspects.

SRELS Journal of Information Management, 2019

The idea of Balanced Development is to ensure equitable distribution of necessities of life and c... more The idea of Balanced Development is to ensure equitable distribution of necessities of life and creation of productive remunerative employment across society. Balanced development is a prerequisite for sustainable economic growth and shared prosperity. To facilitate this challenging task, we have proposed an Ontology-based Tantra Social Information Management Framework that can help set goals, design and evaluate interventions, define and monitor relevant metrics in an ongoing basis. Tantra Framework represents social information using ideas from Zachman Framework and concepts from Unified Foundational Ontology. Tantra Framework inter-operates with well-regarded models. Balanced Scorecard is used to define development objectives. Bartels’ Theory of Separations is used to identify barriers to access, consumption and profitable participation. Theory of Change process is used to arrive at right intervention. Distributed Ledgers are proposed to handle Change Management. A Social Informa...

Procedia Computer Science, 2016

We are in the era of wearable technologies, biometrics and multi-factor authentication, where one... more We are in the era of wearable technologies, biometrics and multi-factor authentication, where one's face is increasingly becoming a digital identifier for access control and authentication. Compared to the other biometrics such as Fingerprint, Iris and Palm print, Face Recognition (FR) has the distinctness of being non-intrusive, and has hence garnered substantial mainstream attention. As the devices that incorporate FR are evolving into miniaturization, there is a need to develop more robust algorithms that are computationally less expensive. Hence, in an effort to provide a computationally effective FR methodology, we extend the costeffective GIST descriptor that was designed primarily for object recognition, to be commensurate with FR. This paper proposes a double filtered GIST based descriptor for FR that embodies certain inventive preprocessing steps such as edge detection via Prewitt descriptor, DCT and IDCT transformation to reduce noise prior to feature description with GIST. We will demonstrate by performing extensive experimentations on the ORL and IIT-K face databases that the proposed methodology is capable of effectively performing FR, even in the presence of sharp variations in a number of crucial FR parameters among the faces being compared.

Proceedings of the International Conferences on WWW/Internet 2021 and Applied Computing 2021

Large scale video retrieval is a field of study with a lot of ongoing research. Most of the work ... more Large scale video retrieval is a field of study with a lot of ongoing research. Most of the work in the field is on video retrieval through text queries using techniques such as VSE++. However, there is little research done on video retrieval through image queries, and the work that has been done in this field either uses image queries from within the video dataset or iterates through videos frame by frame. These approaches are not generalized for queries from outside the dataset and do not scale well for large video datasets. To overcome these issues, we propose a new approach for video retrieval through image queries where an undirected graph is constructed from the combined set of frames from all videos to be searched. The node features of this graph are used in the task of video retrieval. Experimentation is done on the MSR-VTT dataset by using query images from outside the dataset. To evaluate this novel approach P@5, P@10 and P@20 metrics are calculated. Two different ResNet models namely, ResNet-152 and ResNet-50 are used in this study.

arXiv (Cornell University), May 6, 2022

In India, post demonetization exercise in 2016, digital payments have become extremely popular. A... more In India, post demonetization exercise in 2016, digital payments have become extremely popular. Among them, the volume of transactions using Paytm wallets and UPI (Unified Payment Interface) have grown manyfold. The lockdowns due to COVID-19 Pandemic have furthered this trend. Side by side, crypto-currencies such as bitcoin are also gaining traction. Many countries are considering issuing a Digital Currency via their Central Banks. In this paper, we propose a novel Decentralized Digital Currency System (DDCS) that makes use of Merkle Hash-Trees as Authenticated Data Structures. DDCS uses a Ledger-less, distributed, peer-to-peer architecture. We name the proposed currency δ-Money. δ-Money is intended as a replacement for physical currency and has in-built security features that rival crypto-currencies. Transactions using δ-Money happen in a disintermediated manner but with post-facto reconciliation. In place of Central Bank-issued Digital Currency (CBDC), we envisage a scenario where multiple Payment Banks issue digital currencies that have stable valuations without being subject to either volatility or perennial devaluation.

Algorithms for Intelligent Systems, 2021

Machine Learning and Knowledge Discovery in Databases, 2020

State-of-the-art techniques that perform reasoning over large knowledge graphs incorporate a path... more State-of-the-art techniques that perform reasoning over large knowledge graphs incorporate a path-finder and a path-reasoner. The path-reasoner is a learning agent that performs the inference task by producing a relation. The path-finder is usually the knowledge graph environment that moves the agent to the next-hop entity. While the path-reasoner can work on a continuous state space, the knowledge graph environment on which it is being trained, operates on a discrete state space. This restricts the agent from deducing implicit paths. In this paper, a novel path-finder called Stochastic Dynamic Environment for Knowledge Graphs (SDE-KG) has been proposed. SDE-KG is a meta-framework that can be combined with many embedding methods to create a continuous function of the knowledge graph environment which may facilitate smarter multi-hop reasoning.

Deep learning has been recently successfully applied to an ever larger number of problems, rangin... more Deep learning has been recently successfully applied to an ever larger number of problems, ranging from pattern recognition to complex decision making. However, several concerns have been raised, including guarantees of good generalization, which is of foremost importance. Despite numerous attempts, conventional statistical learning approaches fall short of providing a satisfactory explanation on why deep learning works. In a competition hosted at the Thirty-Fourth Conference on Neural Information Processing Systems (NeurIPS 2020), we invited the community to design robust and general complexity measures that can accurately predict the generalization of models. In this paper, we describe † Members of the top three teams ∗ Partly done while at Google ‡ Partly done while at IMT Atlantique § Now at Apple © 2021 Y. Jiang et al. The PGDL Competition the competition design, the protocols, and the solutions of the top-three teams at the competition in details. In addition, we discuss the o...

2021 Smart Technologies, Communication and Robotics (STCR), Oct 9, 2021

Traditional image super resolution centered around purely mathematical models are capable of crea... more Traditional image super resolution centered around purely mathematical models are capable of creating gradient based textures, but fail to render the specific lineaments that would be expected in a realistically upscaled image. This is especially problematic in scenarios involving images of subjects whose recognition is reliant on the presence of specific characteristics, for example, faces. In this paper, we describe a deep learning model that is capable of generating an 8x upscaled photo from a low resolution image of a face. The underlying model is based on the SRGAN architecture that deviates from the conventional GAN approach of the adversarial back and forth between generator and discriminator by incorporating an added content loss whose value is dependent on the detection of the natural features in the generated image by a pre-trained VGG model. The model is trained on the Celeb Faces Attributes dataset with over 1,00,000 data points and can produce upscaled images that are realistic with a coherent presence of the natural attributes of a face.

Communications in computer and information science, 2020

State-of-the-art techniques that perform reasoning over large knowledge graphs incorporate a path... more State-of-the-art techniques that perform reasoning over large knowledge graphs incorporate a path-finder and a path-reasoner. The path-reasoner is a learning agent that performs the inference task by producing a relation. The path-finder is usually the knowledge graph environment that moves the agent to the next-hop entity. While the path-reasoner can work on a continuous state space, the knowledge graph environment on which it is being trained, operates on a discrete state space. This restricts the agent from deducing implicit paths. In this paper, a novel path-finder called Stochastic Dynamic Environment for Knowledge Graphs (SDE-KG) has been proposed. SDE-KG is a meta-framework that can be combined with many embedding methods to create a continuous function of the knowledge graph environment which may facilitate smarter multi-hop reasoning.

Discovering interesting, implicit knowledge and general relationships in geographic information d... more Discovering interesting, implicit knowledge and general relationships in geographic information databases is very important to understand and to use the spatial data. Spatial clustering has been recognized as a primary data mining method for knowledge discovery in spatial databases. In this paper, we have analyzed an efficient method for the fusion of the outputs of the various clusterers, with less

Low-dimensional embeddings for knowledge graph entities and relations help preserve their latent ... more Low-dimensional embeddings for knowledge graph entities and relations help preserve their latent semantics while enabling computation efficiency. These embeddings are often used to perform tasks such as knowledge graph completion, question answering and inference. Knowledge graph embedding methods aid in the representation of entities and relationships of a knowledge graph in continuous vector spaces. However, most existing techniques ignore the inherent hierarchical structure of entities present in the knowledge graph, defined by ontological relationships between entity types. This paper introduces a novel score function called GrCluster that helps fill that gap. GrCluster is a simple, intuitive and efficient scoring function that incorporates the entity hierarchical correlation into existing knowledge graph embeddings. The effectiveness of GrCluster is demonstrated by integrating it into several well known embedding models. The experimental results show consistent improvements across metrics and embedding models for the tasks of entity prediction and triplet classification.

Advances in intelligent systems and computing, Dec 14, 2018

Question answering (QA) is a field of Natural Language Processing that deals with generating answ... more Question answering (QA) is a field of Natural Language Processing that deals with generating answers automatically to questions asked to a system. It can be categorized into two types—open-domain and closed-domain QA. Open-domain QA can deal with questions about anything, whereas closed-domain QA deals with questions in a specific domain. In our work, we use the architectures of LSTM and memory networks to perform closed-domain question answering and compare the performances of the two. LSTMs are specialized RNNs that can remember necessary data and forget the irrelevant bits. Since data in QA consist of stories and questions based on them, this model seems appropriate, with the ability to handle long sequences. On the other hand, memory networks provide an architecture where there is a provision to store the information learnt by the system in an explicit memory component, rather than just as weight matrices. This also seems like an architecture well-suited to question answering. We implement each model and train it on the Facebook bAbi dataset. This dataset is specifically generated for the purpose of evaluating QA systems on the twenty prerequisite toy bAbi tasks. Each dataset corresponds to one task and checks whether the model is able to perform chaining, counting, answer with single and multiple supporting facts, understand relations, directions, etc. Based on the performances of each model on the bAbi tasks, we perform a comparative study of the two.

arXiv (Cornell University), May 6, 2022

In India, post demonetization exercise in 2016, digital payments have become extremely popular. A... more In India, post demonetization exercise in 2016, digital payments have become extremely popular. Among them, the volume of transactions using Paytm wallets and UPI (Unified Payment Interface) have grown manyfold. The lockdowns due to COVID-19 Pandemic have furthered this trend. Side by side, crypto-currencies such as bitcoin are also gaining traction. Many countries are considering issuing a Digital Currency via their Central Banks. In this paper, we propose a novel Decentralized Digital Currency System (DDCS) that makes use of Merkle Hash-Trees as Authenticated Data Structures. DDCS uses a Ledger-less, distributed, peer-to-peer architecture. We name the proposed currency δ-Money. δ-Money is intended as a replacement for physical currency and has in-built security features that rival crypto-currencies. Transactions using δ-Money happen in a disintermediated manner but with post-facto reconciliation. In place of Central Bank-issued Digital Currency (CBDC), we envisage a scenario where multiple Payment Banks issue digital currencies that have stable valuations without being subject to either volatility or perennial devaluation.

arXiv (Cornell University), Jan 16, 2021

Generalization is the ability of a model to predict on unseen domains and is a fundamental task i... more Generalization is the ability of a model to predict on unseen domains and is a fundamental task in machine learning. Several generalization bounds, both theoretical and empirical have been proposed but they do not provide tight bounds. In this work, we propose a simple yet effective method to predict the generalization performance of a model by using the concept that models that are robust to augmentations are more generalizable than those which are not. We experiment with several augmentations and composition of augmentations to check the generalization capacity of a model. We also provide a detailed motivation behind the proposed method. The proposed generalization metric is calculated based on the change in the model's output after augmenting the input. The proposed method was the first runner up solution for the competition "Predicting Generalization in Deep Learning".

arXiv (Cornell University), Nov 24, 2022

Image inpainting is the process of regenerating lost parts of the image. Supervised algorithm-bas... more Image inpainting is the process of regenerating lost parts of the image. Supervised algorithm-based methods have shown excellent results but have two significant drawbacks. They do not perform well when tested with unseen data. They fail to capture the global context of the image, resulting in a visually unappealing result. We propose a novel self-supervised learning framework for image-inpainting: Weighted Self-Supervised Learning (WSSL) to tackle these problems. We designed WSSL to learn features from multiple weighted pretext tasks. These features are then utilized for the downstream task, image-inpainting. To improve the performance of our framework and produce more visually appealing images, we also present a novel loss function for image inpainting. The loss function takes advantage of both reconstruction loss and perceptual loss functions to regenerate the image. Our experimentation shows WSSL outperforms previous methods, and our loss function helps produce better results.

Large scale video retrieval is a field of study with a lot of ongoing research. Most of the work ... more Large scale video retrieval is a field of study with a lot of ongoing research. Most of the work in the field is on video retrieval through text queries using techniques such as VSE++. However, there is little research done on video retrieval through image queries, and the work that has been done in this field either uses image queries from within the video dataset or iterates through videos frame by frame. These approaches are not generalized for queries from outside the dataset and do not scale well for large video datasets. To overcome these issues, we propose a new approach for video retrieval through image queries where an undirected graph is constructed from the combined set of frames from all videos to be searched. The node features of this graph are used in the task of video retrieval. Experimentation is done on the MSR-VTT dataset by using query images from outside the dataset. To evaluate this novel approach P@5, P@10 and P@20 metrics are calculated. Two different ResNet models namely, ResNet-152 and ResNet-50 are used in this study.

Springer eBooks, Oct 2, 2021

CSI Transactions on ICT

The information systems have been extremely useful in managing businesses, enterprises, and publi... more The information systems have been extremely useful in managing businesses, enterprises, and public institutions such as government departments. But today's challenges are increasingly about managing ecosystems. Ecosystem is a useful paradigm to better understand a variety of domains such as biology, business, industry, agriculture, and society. In this paper, we look at the Indian Agricultural ecosystem. It is a mammoth task to assimilate the information for the whole ecosystem consisting of consumers, producers, workers, traders, transporters, industry, and Government. There are myriad interventions by the state and the central Governments, whose efficacy is difficult to track and the outcomes hard to assess. A policy intervention that helps one part of the ecosystem can harm the other. In addition, sustainability and ecological considerations are also extremely important. In this paper, we make use of the Knowledge-based Tantra Social Information Management Framework to analyze the Indian Agricultural Ecosystem and build related Knowledge Graphs. Our analysis spans descriptive, normative, and transformative viewpoints. Tantra Framework makes use of concepts from Zachman Framework to manage aspects of social information through different perspectives and concepts from Unified Foundational Ontology (UFO) to represent interrelationships between aspects.

SRELS Journal of Information Management, 2019

The idea of Balanced Development is to ensure equitable distribution of necessities of life and c... more The idea of Balanced Development is to ensure equitable distribution of necessities of life and creation of productive remunerative employment across society. Balanced development is a prerequisite for sustainable economic growth and shared prosperity. To facilitate this challenging task, we have proposed an Ontology-based Tantra Social Information Management Framework that can help set goals, design and evaluate interventions, define and monitor relevant metrics in an ongoing basis. Tantra Framework represents social information using ideas from Zachman Framework and concepts from Unified Foundational Ontology. Tantra Framework inter-operates with well-regarded models. Balanced Scorecard is used to define development objectives. Bartels’ Theory of Separations is used to identify barriers to access, consumption and profitable participation. Theory of Change process is used to arrive at right intervention. Distributed Ledgers are proposed to handle Change Management. A Social Informa...

Procedia Computer Science, 2016

We are in the era of wearable technologies, biometrics and multi-factor authentication, where one... more We are in the era of wearable technologies, biometrics and multi-factor authentication, where one's face is increasingly becoming a digital identifier for access control and authentication. Compared to the other biometrics such as Fingerprint, Iris and Palm print, Face Recognition (FR) has the distinctness of being non-intrusive, and has hence garnered substantial mainstream attention. As the devices that incorporate FR are evolving into miniaturization, there is a need to develop more robust algorithms that are computationally less expensive. Hence, in an effort to provide a computationally effective FR methodology, we extend the costeffective GIST descriptor that was designed primarily for object recognition, to be commensurate with FR. This paper proposes a double filtered GIST based descriptor for FR that embodies certain inventive preprocessing steps such as edge detection via Prewitt descriptor, DCT and IDCT transformation to reduce noise prior to feature description with GIST. We will demonstrate by performing extensive experimentations on the ORL and IIT-K face databases that the proposed methodology is capable of effectively performing FR, even in the presence of sharp variations in a number of crucial FR parameters among the faces being compared.

Proceedings of the International Conferences on WWW/Internet 2021 and Applied Computing 2021

Large scale video retrieval is a field of study with a lot of ongoing research. Most of the work ... more Large scale video retrieval is a field of study with a lot of ongoing research. Most of the work in the field is on video retrieval through text queries using techniques such as VSE++. However, there is little research done on video retrieval through image queries, and the work that has been done in this field either uses image queries from within the video dataset or iterates through videos frame by frame. These approaches are not generalized for queries from outside the dataset and do not scale well for large video datasets. To overcome these issues, we propose a new approach for video retrieval through image queries where an undirected graph is constructed from the combined set of frames from all videos to be searched. The node features of this graph are used in the task of video retrieval. Experimentation is done on the MSR-VTT dataset by using query images from outside the dataset. To evaluate this novel approach P@5, P@10 and P@20 metrics are calculated. Two different ResNet models namely, ResNet-152 and ResNet-50 are used in this study.

arXiv (Cornell University), May 6, 2022

In India, post demonetization exercise in 2016, digital payments have become extremely popular. A... more In India, post demonetization exercise in 2016, digital payments have become extremely popular. Among them, the volume of transactions using Paytm wallets and UPI (Unified Payment Interface) have grown manyfold. The lockdowns due to COVID-19 Pandemic have furthered this trend. Side by side, crypto-currencies such as bitcoin are also gaining traction. Many countries are considering issuing a Digital Currency via their Central Banks. In this paper, we propose a novel Decentralized Digital Currency System (DDCS) that makes use of Merkle Hash-Trees as Authenticated Data Structures. DDCS uses a Ledger-less, distributed, peer-to-peer architecture. We name the proposed currency δ-Money. δ-Money is intended as a replacement for physical currency and has in-built security features that rival crypto-currencies. Transactions using δ-Money happen in a disintermediated manner but with post-facto reconciliation. In place of Central Bank-issued Digital Currency (CBDC), we envisage a scenario where multiple Payment Banks issue digital currencies that have stable valuations without being subject to either volatility or perennial devaluation.

Algorithms for Intelligent Systems, 2021

Machine Learning and Knowledge Discovery in Databases, 2020

State-of-the-art techniques that perform reasoning over large knowledge graphs incorporate a path... more State-of-the-art techniques that perform reasoning over large knowledge graphs incorporate a path-finder and a path-reasoner. The path-reasoner is a learning agent that performs the inference task by producing a relation. The path-finder is usually the knowledge graph environment that moves the agent to the next-hop entity. While the path-reasoner can work on a continuous state space, the knowledge graph environment on which it is being trained, operates on a discrete state space. This restricts the agent from deducing implicit paths. In this paper, a novel path-finder called Stochastic Dynamic Environment for Knowledge Graphs (SDE-KG) has been proposed. SDE-KG is a meta-framework that can be combined with many embedding methods to create a continuous function of the knowledge graph environment which may facilitate smarter multi-hop reasoning.

Deep learning has been recently successfully applied to an ever larger number of problems, rangin... more Deep learning has been recently successfully applied to an ever larger number of problems, ranging from pattern recognition to complex decision making. However, several concerns have been raised, including guarantees of good generalization, which is of foremost importance. Despite numerous attempts, conventional statistical learning approaches fall short of providing a satisfactory explanation on why deep learning works. In a competition hosted at the Thirty-Fourth Conference on Neural Information Processing Systems (NeurIPS 2020), we invited the community to design robust and general complexity measures that can accurately predict the generalization of models. In this paper, we describe † Members of the top three teams ∗ Partly done while at Google ‡ Partly done while at IMT Atlantique § Now at Apple © 2021 Y. Jiang et al. The PGDL Competition the competition design, the protocols, and the solutions of the top-three teams at the competition in details. In addition, we discuss the o...

https://arxiv.org/abs/2110.09297v1, 2021

The information systems have been extremely useful in managing businesses, enterprises, and publi... more The information systems have been extremely useful in managing businesses, enterprises, and public institutions such as government departments. But today's challenges are increasingly about managing ecosystems. Ecosystem is a useful paradigm to better understand a variety of domains such as biology, business, industry, agriculture, and society. In this paper, we look at the Indian Agricultural ecosystem. It is a mammoth task to assimilate the information for the whole ecosystem consisting of consumers, producers, workers, traders, transporters, industry, and Government. There are myriad interventions by the state and the central Governments, whose efficacy is difficult to track and the outcomes hard to assess. A policy intervention that helps one part of the ecosystem can harm the other. In addition, sustainability and ecological considerations are also extremely important. In this paper, we make use of the Knowledge-based Tantra Social Information Management Framework to analyze the Indian Agricultural Ecosystem and build related Knowledge Graphs. Our analysis spans descriptive, normative, and transformative viewpoints. Tantra Framework makes use of concepts from Zachman Framework to manage aspects of social information through different perspectives and concepts from Unified Foundational Ontology (UFO) to represent interrelationships between aspects.

This paper proposes ontology-based Tantra Framework that accommodates Societal Information in an ... more This paper proposes ontology-based Tantra Framework that accommodates Societal Information in an orderly, compact and unified manner to achieve Good Governance. In this work, Zachman Framework for Enterprise Architecture is used as reference and it is extended to address information operating at societal scale. The complexity of social information gets accentuated due to myriad possibilities of relationships, say compared to information pertaining to an Enterprise. In light of that, two additional columns namely relationships and relators have been added to Zachman Framework in Tantra Framework. The existing six columns of Zachman Framework namely who, what, how, when, where and why are interpreted as People, Assets/Attributes, Process, Event, Location and Objectives respectively. In addition to the six interrogatives of Zachman Framework that are ontology based, the concept of relationships and relators has been derived from The Unified Foundational Ontology. Even though Zachman considered his framework as Ontology, some researchers regarded it as taxonomy. Tantra Framework has addressed that perceived gap by extending the Zachman Framework by adding two additional columns. The utility of Tantra Framework is analyzed by applying it to a set of Application Scenarios pertaining to People and Locations information, Revenue Capture, Social Benefit Coverage Analysis, Financial Inclusion and Metrics analysis. Tanta Framework interoperates with Balanced Score Card approach to set objectives, Theory of Change to lay out a process of change and Bartels' theory of Market Separations to assess the access barriers within Society. This research study is focused on Indian context. Tantra Framework relies on eliciting the inherent order in social information to make it complete, correct and current. For example, in Chemistry, discovery of periodic table by Mendeleev led to a unifying scheme which not only captured the present but allowed for place-holders for future discoveries. A similar focus on eliciting inherent order in social information, can potentially reveal latent 'facts', 'truths' and 'relationships.