Department of Computer Science & Engineering

&ldquo;IIT Jodhpur is committed to impart quality education to facilitate the creation of qualified, competent and responsible human resources to meet the emerging technological challenges of the world. It is working towards creating an equitable learning platform that is positive, accessible and effective in the face of the modern societal and scientific advancements.

Flexibility in curriculum, in all its academic programs, encompassing elements of teaching, research, creativity, innovation and entrepreneurship is the most unique feature of IIT Jodhpur. The Department of Science and Technology (DST), under the National Mission on Cyber Physical Systems, has identified IIT Jodhpur as a Technology &amp; Innovation Hub in the area of Augmented Reality and Virtual Reality. The institute has also been nominated as the coordinator for the Jodhpur City and Knowledge Cluster by the Office of Principal Scientific Adviser, Government of India.

The sprawling 852-acre IIT Jodhpur campus is a majestic tribute to the deep-rooted diverse cultural heritage of the region. The campus is an architectural masterpiece, which is slated to become an international exemplar of sustainability with net-zero energy, and advanced water and waste management strategies.&rdquo;

IIT Jodhpur

The growth and success of deep learning approaches can be attributed to two major factors: availability of hardware resources and availability of large number of training samples. For problems with large training databases, deep learning models have achieved superlative performances. However, there are a lot of small sample size or S3 problems for which it is not feasible to collect large training databases. It has been observed that deep learning models do not generalize well on S3 problems and specialized solutions are required. In this paper, we first present a review of deep learning algorithms for small sample size problems in which the algorithms are segregated according to the space in which they operate, i.e. input space, model space, and feature space. Secondly, we present Dynamic Attention Pooling approach which focuses on extracting global information from the most discriminative sub-part of the feature map. The performance of the proposed dynamic attention pooling is analyzed with state-of-the-art ResNet model on relatively small publicly available datasets such as SVHN, C10, C100, and TinyImageNet. © 2020 IEEE.

World University root node

Proceedings - 2020 IEEE 6th International Conference on Multimedia Big Data, BigMM 2020

Unravelling Small Sample Size Problems in the Deep Learning World

We propose a fast few-shot learning framework that uses transfer learning to identify different lung and chest diseases and conditions from chest x-rays. Our model can be trained with as few as five training examples, making it potentially applicable for diagnosis of rare diseases. In this work, we divide different chest diseases into two disjoint categories: (i) base classes (with large training set) and (ii) novel classes (with a few training examples per class). Our method consists of two steps, namely feature extraction and classification. For the feature extraction, we employ a deep convolutional neural network, customized for chest x-rays. We train the feature extractor with data only from base classes. So the novel classes are unseen to the feature extractor during training. However, we use the feature extractor for extracting features from the data of novel classes resulting in transfer learning. Our classifier, on the other hand, uses the data only from the novel classes for training. We introduce the idea of autoencoder ensemble to design the classifier. Only a few feature vectors from each of the novel classes are used for training the classifier making it a few-shot learner. Incorporating new novel classes require training only the classifier which makes the entire process extremely fast. The performance of the classifier is evaluated on the test data from the novel classes. Experiments show - 18% improvement in the F1 score compared to the baseline on identifying the novel diseases from publicly available chest x-ray dataset. © 2020 SPIE.

Progress in Biomedical Optics and Imaging - Proceedings of SPIE

Fast few-shot transfer learning for disease identification from chest x-ray images using autoencoder ensemble

Diabetic Retinopathy (DR) is a polygenic disorder issue that affects human eyes. Bruise to the blood vessels of the photosensitive tissue of the retina causes this complication. It's most frequent in patients who had diabetes for more than ten years. This downside is going on in several individuals worldwide. However, the number of medical practitioners and also the tools needed for the detection of DR are very less for serving the mass population. In this paper, we have proposed DRDNet (Diabetic Retinopathy Diagnosis Network), a neural network framework based on capsule networks (CapsNets) for DR diagnosis. Experiments on a dataset with 1,265 images demonstrate that CapsNet shows better accuracy and convergence behavior for the complex data than the state-ofthe-art techniques. The proposed DRDNet performs with an overall accuracy of 80.59% for five class, as compared to the closest competitor with an accuracy of 75.83%. We performed a study on a mixed dataset for two class and found that testing accuracy was 80.59%. We have also done training on a two class model and testing on other unseen datasets. Moreover, we observed that DRDNet has much higher confidence for the predicted probabilities as compared to other state-of-the-art techniques. © 2020 IEEE.

Drdnet: Diagnosis of Diabetic Retinopathy Using Capsule Network (Workshop Paper)

A graph operation that contracts edges is one of the fundamental operations in the theory of graph minors. Parameterized Complexity of editing to a family of graphs by contracting k edges has recently gained substantial scientific attention, and several new results have been obtained. Some important families of graphs, namely the subfamilies of chordal graphs, in the context of edge contractions, have proven to be significantly difficult than one might expect. In this paper, we study the F-Contraction problem, where F is a subfamily of chordal graphs, in the realm of parameterized approximation. Formally, given a graph G and an integer k, F-Contraction asks whether there exists X ⊆ E(G) such that G/X ∈ F and |X| ≤ k. Here, G/X is the graph obtained from G by contracting edges in X. We obtain the following results for the F-Contraction problem. Clique Contraction is known to be FPT. However, unless NP ⊆ coNP/poly, it does not admit a polynomial kernel. We show that it admits a polynomial-size approximate kernelization scheme (PSAKS). That is, it admits a (1 + ε)-approximate kernel with O(kf(ε)) vertices for every ε &gt; 0. Split Contraction is known to be W[1]-Hard. We deconstruct this intractability result in two ways. Firstly, we give a (2+ε)-approximate polynomial kernel for Split Contraction (which also implies a factor (2 + ε)-FPT-approximation algorithm for Split Contraction). Furthermore, we show that, assuming Gap-ETH, there is no (5/4 − δ)-FPT-approximation algorithm for Split Contraction. Here, ε, δ &gt; 0 are fixed constants. Chordal Contraction is known to be W[2]-Hard. We complement this result by observing that the existing W[2]-hardness reduction can be adapted to show that, assuming FPT6= W[1], there is no F(k)-FPT-approximation algorithm for Chordal Contraction. Here, F(k) is an arbitrary function depending on k alone. We say that an algorithm is an h(k)-FPT-approximation algorithm for the F-Contraction problem, if it runs in FPT time, and on any input (G, k) such that there exists X ⊆ E(G) satisfying G/X ∈ F and |X| ≤ k, it outputs an edge set Y of size at most h(k) · k for which G/Y is in F. We find it extremely interesting that three closely related problems have different behavior with respect to FPT-approximation. © 2020 Schloss Dagstuhl- Leibniz-Zentrum fur Informatik GmbH, Dagstuhl Publishing. All rights reserved.

Leibniz International Proceedings in Informatics, LIPIcs

On the parameterized approximability of contraction to classes of chordal graphs

Participatory budgeting systems allow city residents to jointly decide on projects they wish to fund using public money, by letting residents vote on such projects. While participatory budgeting is gaining popularity, existing aggregation methods do not take into account the natural possibility of project interactions, such as substitution and complementarity effects. Here we take a step towards fixing this issue: First, we augment the standard model of participatory budgeting by introducing a partition over the projects and model the type and extent of project interactions within each part using certain functions. We study the computational complexity of finding bundles that maximize voter utility, as defined with respect to such functions. Motivated by the desire to incorporate project interactions in real-world participatory budgeting systems, we identify certain cases that admit efficient aggregation in the presence of such project interactions. © 2020 Inst. Sci. inf., Univ. Defence in Belgrade. All rights reserved.

IJCAI International Joint Conference on Artificial Intelligence

Participatory budgeting with project interactions

We study committee selection with multimodal preferences: Assuming a set of candidates A, a set of voters V, and layers, where each voter v V has ordinal preferences over the alternatives for each layer separately, the task is to select a committee S A of size k. We discuss applications of our model and study the computational complexity of several generalizations of known committee scoring rules (specifically, k-Borda and Chamberlin-Courant) to our setting, as well as discuss domain restrictions for our model. While most problems we encounter are computationally intractable in general, we nevertheless design efficient algorithms for certain cases. © 2020 The authors and IOS Press.

Frontiers in Artificial Intelligence and Applications

Committee selection with multimodal preferences

Vertex deletion problems on graphs involve finding a set of minimum number of vertices whose deletion results in a graph with some specific property. In a recent study on vertex deletion problems on split graphs, it was shown that transforming a split graph to a block graph or a threshold graph using minimum number of vertex deletions is NP-hard. We call the decision version of these problems as SPLIT TO BLOCK VERTEX DELETION (SBVD) and SPLIT TO THRESHOLD VERTEX DELETION (STVD), respectively. In this paper, we study the parameterized complexity of these problems with the number of vertex deletions, k, as a parameter. These problems are implicit 4-HITTING SET and thus admit an algorithm with running time O⋆(3.0755k), a kernel with O(k3) vertices, and a 4-approximation algorithm. In this paper, we exploit the structural properties of the concerned graph classes and obtain a kernel for SBVD with O(k2) vertices and FPT algorithms for SBVD and STVD with running times O⋆(2.3028k) and O⋆(2.7913k), respectively. We further give improved approximation algorithms for SBVD and STVD. © 2020 Elsevier B.V.

Theoretical Computer Science

Vertex deletion on split graphs: Beyond 4-hitting set

The paper defines a new theoretical measure Total Influence, of a node as well as a set of nodes in the social network. Total influence uses probabilistic theory to obtain the expected size of the information spreading in the social network under the independent cascade model of diffusion. In order to quantify the size of the spreading practically, the paper proposes a new hybrid simulation methodology for the independent cascade model. The hybrid method uses rough set theory and defines rough knowledge agents around all the seed nodes from which the information is propagating. The lower approximation is calculated using the probabilistic approach, while the size of influence in the boundary region is quantified by Monte-Carlo simulation on a reduced network. The reduce network is formed by compacting all the nodes in the lower approximate region as a super-node. Experimental results on two synthetically generated directed network show that the hybrid method runs magnitude faster than its counterpart with a similar accuracy of the spreading size. © 2020 IEEE.

4th International Conference on Computational Intelligence and Networks, CINE 2020

Total Influence and Hybrid Simulation of Independent Cascade Model using Rough Knowledge Granules

With the advancements in technology, smartphones' capabilities have increased immensely. For instance, the smartphone cameras are being used for face and ocular biometric-based authentication. This research proposes finger-selfie based authentication mechanism, which uses a smartphone camera to acquire a selfie of a finger. In addition to personal device-level authentication, finger-selfies may also be matched with livescan fingerprints present in the legacy/national ID databases for remote or touchless authentication. We propose an algorithm which comprises segmentation, enhancement, Deep Scattering Network based feature extraction and Random Decision Forest to authenticate finger-selfies. This paper also presents one of the largest finger-selfie database with over 19, 400 images. The images in the IIIT-D Smartphone Finger-selfie Database v2 are captured using multiple smartphones and include variations due to background, illumination, resolution and sensors. Results and comparison with existing algorithms show the efficacy of the proposed algorithm which yields equal error rates in the range of 2.1-5.2% for different experimental protocols.

IEEE Transactions on Biometrics, Behavior and Identity Science

On Matching Finger-Selfies Using Deep Scattering Networks

One of the most challenging scenarios of face recognition is matching images in presence of multiple covariates such as cross-spectrum and cross-resolution. In this paper, we propose a Subclass Heterogeneity Aware Loss (SHEAL) to train a deep convolutional neural network model such that it produces embeddings suitable for heterogeneous face recognition, both single and multiple heterogeneities. The performance of the proposed SHEAL function is evaluated on four databases in terms of the recognition performance as well as convergence in time and epochs. We observe that SHEAL not only yields state-of-the-art results for the most challenging case of Cross-Spectral Cross-Resolution face recognition, it also achieves excellent performance on homogeneous face recognition.