arXiv subjects

Ankit Sharma

Publications and source records attributed to Ankit Sharma.

At least 37 records · Page 2Linked to original sources

arXiv2021

Towards Consistent Predictive Confidence through Fitted Ensembles

Deep neural networks are behind many of the recent successes in machine learning applications. However, these models can produce overconfident decisions while encountering out-of-distribution (OOD) examples or making a wrong prediction. This inconsistent predictive confidence limits the integration of independently-trained learning models into a larger system. This paper introduces separable concept learning framework to realistically measure the performance of classifiers in presence of OOD examples. In this setup, several instances of a classifier are trained on different parts of a partition of the set of classes. Later, the performance of the combination of these models is evaluated on a separate test set. Unlike current OOD detection techniques, this framework does not require auxiliary OOD datasets and does not separate classification from detection performance. Furthermore, we present a new strong baseline for more consistent predictive confidence in deep models, called fitted ensembles, where overconfident predictions are rectified by transformed versions of the original classification task. Fitted ensembles can naturally detect OOD examples without requiring auxiliary data by observing contradicting predictions among its components. Experiments on MNIST, SVHN, CIFAR-10/100, and ImageNet show fitted ensemble significantly outperform conventional ensembles on OOD examples and are possible to scale.

cs.LG↗

arXiv2021

Hi-CI: Deep Causal Inference in High Dimensions

We address the problem of counterfactual regression using causal inference (CI) in observational studies consisting of high dimensional covariates and high cardinality treatments. Confounding bias, which leads to inaccurate treatment effect estimation, is attributed to covariates that affect both treatments and outcome. The presence of high-dimensional co-variates exacerbates the impact of bias as it is harder to isolate and measure the impact of these confounders. In the presence of high-cardinality treatment variables, CI is rendered ill-posed due to the increase in the number of counterfactual outcomes to be predicted. We propose Hi-CI, a deep neural network (DNN) based framework for estimating causal effects in the presence of large number of covariates, and high-cardinal and continuous treatment variables. The proposed architecture comprises of a decorrelation network and an outcome prediction network. In the decorrelation network, we learn a data representation in lower dimensions as compared to the original covariates and addresses confounding bias alongside. Subsequently, in the outcome prediction network, we learn an embedding of high-cardinality and continuous treatments, jointly with the data representation. We demonstrate the efficacy of causal effect prediction of the proposed Hi-CI network using synthetic and real-world NEWS datasets.

stat.ME↗

arXiv2021

Origin of the quasi-quantized Hall effect in ZrTe5

The quantum Hall effect (QHE) is traditionally considered a purely two-dimensional (2D) phenomenon. Recently, a three-dimensional (3D) version of the QHE has been reported in the Dirac semimetal ZrTe5. It was proposed to arise from a magnetic-field-driven Fermi surface instability, transforming the original 3D electron system into a stack of 2D sheets. Here, we report thermodynamic, thermoelectric and charge transport measurements on ZrTe5 in the quantum Hall regime. The measured thermodynamic properties: magnetization and ultrasound propagation, show no signatures of a Fermi surface instability, consistent with in-field single crystal X-ray diffraction. Instead, a direct comparison of the experimental data with linear response calculations based on an effective 3D Dirac Hamiltonian suggests that the quasi-quantization of the observed Hall response is an intrinsic property of the 3D electronic structure. Our findings render the Hall effect in ZrTe5 a truly 3D counterpart of the QHE in 2D systems.

cond-mat.str-el↗

arXiv2021

MetaCI: Meta-Learning for Causal Inference in a Heterogeneous Population

Performing inference on data obtained through observational studies is becoming extremely relevant due to the widespread availability of data in fields such as healthcare, education, retail, etc. Furthermore, this data is accrued from multiple homogeneous subgroups of a heterogeneous population, and hence, generalizing the inference mechanism over such data is essential. We propose the MetaCI framework with the goal of answering counterfactual questions in the context of causal inference (CI), where the factual observations are obtained from several homogeneous subgroups. While the CI network is designed to generalize from factual to counterfactual distribution in order to tackle covariate shift, MetaCI employs the meta-learning paradigm to tackle the shift in data distributions between training and test phase due to the presence of heterogeneity in the population, and due to drifts in the target distribution, also known as concept shift. We benchmark the performance of the MetaCI algorithm using the mean absolute percentage error over the average treatment effect as the metric, and demonstrate that meta initialization has significant gains compared to randomly initialized networks, and other methods.

cs.LG↗

arXiv2020

Parallelized Instantaneous Velocity and Heading Estimation of Objects using Single Imaging Radar

The development of high-resolution imaging radars introduce a plethora of useful applications, particularly in the automotive sector. With increasing attention on active transport safety and autonomous driving, these imaging radars are set to form the core of an autonomous engine. One of the most important tasks of such high-resolution radars is to estimate the instantaneous velocities and heading angles of the detected objects (vehicles, pedestrians, etc.). Feasible estimation methods should be fast enough in real-time scenarios, bias-free and robust against micro-Dopplers, noise and other systemic variations. This work proposes a parallel-computing scheme that achieves a real-time and accurate implementation of vector velocity determination using frequency modulated continuous wave (FMCW) radars. The proposed scheme is tested against traffic data collected using an FMCW radar at a center frequency of 78.6 GHz and a bandwidth of 4 GHz. Experiments show that the parallel algorithm presented performs much faster than its conventional counterparts without any loss in precision.

eess.SP↗

arXiv2020

Effect of lockdown interventions to control the COVID-19 epidemic in India

The pandemic caused by the novel Coronavirus SARS-CoV2 has been responsible for life threatening health complications, and extreme pressure on healthcare systems. While preventive and definite curative medical interventions are yet to arrive, Non-Pharmaceutical Interventions (NPIs) like physical isolation, quarantine and drastic social measures imposed by governing agencies are effective in arresting the spread of infections in a population. In densely populated countries like India, lockdown interventions are partially effective due to social and administrative complexities. Using detailed demographic data, we present an agent based model to imitate the behavior of the population and its mobility features, even under intervention. We demonstrate the effectiveness of contact tracing policies and how our model efficiently relates to empirical findings on testing efficiency. We also present various lockdown intervention strategies for mitigation - using the bare number of infections, the effective reproduction rate, as well as using reinforcement learning. Our analysis can help assess the socio-economic consequences of such interventions, and provide useful ideas and insights to policy makers for better decision making.

physics.soc-ph↗

arXiv2020

CCA: Exploring the Possibility of Contextual Camouflage Attack on Object Detection

Deep neural network based object detection hasbecome the cornerstone of many real-world applications. Alongwith this success comes concerns about its vulnerability tomalicious attacks. To gain more insight into this issue, we proposea contextual camouflage attack (CCA for short) algorithm to in-fluence the performance of object detectors. In this paper, we usean evolutionary search strategy and adversarial machine learningin interactions with a photo-realistic simulated environment tofind camouflage patterns that are effective over a huge varietyof object locations, camera poses, and lighting conditions. Theproposed camouflages are validated effective to most of the state-of-the-art object detectors.

cs.CV↗

arXiv2020

Fast Griffin Lim based Waveform Generation Strategy for Text-to-Speech Synthesis

The performance of text-to-speech (TTS) systems heavily depends on spectrogram to waveform generation, also known as the speech reconstruction phase. The time required for the same is known as synthesis delay. In this paper, an approach to reduce speech synthesis delay has been proposed. It aims to enhance the TTS systems for real-time applications such as digital assistants, mobile phones, embedded devices, etc. The proposed approach applies Fast Griffin Lim Algorithm (FGLA) instead Griffin Lim algorithm (GLA) as vocoder in the speech synthesis phase. GLA and FGLA are both iterative, but the convergence rate of FGLA is faster than GLA. The proposed approach is tested on LJSpeech, Blizzard and Tatoeba datasets and the results for FGLA are compared against GLA and neural Generative Adversarial Network (GAN) based vocoder. The performance is evaluated based on synthesis delay and speech quality. A 36.58% reduction in speech synthesis delay has been observed. The quality of the output speech has improved, which is advocated by higher Mean opinion scores (MOS) and faster convergence with FGLA as opposed to GLA.

eess.AS↗

arXiv2020

Low field electron mobility in $α-Ga_{2}O_{3}$: An ab-initio approach

The $α$ phase of $Ga_{2}O_{3}$ is an ultra-wideband semiconductor with potential power electronics applications. In this work, we calculate the low field electron mobility in $α-Ga_{2}O_{3}$ from first principles. The 10 atom unit cell contributes to 30 phonon modes and the effect of each mode is taken into account for the transport calculation. The phonon dispersion and the Raman spectrum are calculated under the density functional perturbation theory formalism and compared with experiments. The IR strength is calculated from the dipole moment at the $Γ$ point of the Brillouin zone. The electron-phonon interaction elements (EPI) on a dense reciprocal space grid is obtained using the Wannier interpolation technique. The polar nature of the material is accounted for by interpolating the non-polar and polar EPI elements independently as the localized nature of the Wannier functions are not suitable for interpolating the long-range polar interaction elements. For polar interaction the full phonon dispersion is taken into account. The electron mobility is then calculated including the polar, non-polar and ionized impurity scattering.

physics.app-ph↗

arXiv2019

Slim-CNN: A Light-Weight CNN for Face Attribute Prediction

We introduce a computationally-efficient CNN micro-architecture Slim Module to design a lightweight deep neural network Slim-Net for face attribute prediction. Slim Modules are constructed by assembling depthwise separable convolutions with pointwise convolution to produce a computationally efficient module. The problem of facial attribute prediction is challenging because of the large variations in pose, background, illumination, and dataset imbalance. We stack these Slim Modules to devise a compact CNN which still maintains very high accuracy. Additionally, the neural network has a very low memory footprint which makes it suitable for mobile and embedded applications. Experiments on the CelebA dataset show that Slim-Net achieves an accuracy of 91.24% with at least 25 times fewer parameters than comparably performing methods, which reduces the memory storage requirement of Slim-net by at least 87%.

cs.CV↗

arXiv2019

Algorithms and Improved bounds for online learning under finite hypothesis class

Online learning is the process of answering a sequence of questions based on the correct answers to the previous questions. It is studied in many research areas such as game theory, information theory and machine learning. There are two main components of online learning framework. First, the learning algorithm also known as the learner and second, the hypothesis class which is essentially a set of functions which learner uses to predict answers to the questions. Sometimes, this class contains some functions which have the capability to provide correct answers to the entire sequence of questions. This case is called realizable case. And when hypothesis class does not contain such functions is called unrealizable case. The goal of the learner, in both the cases, is to make as few mistakes as that could have been made by most powerful functions in hypothesis class over the entire sequence of questions. Performance of the learners is analysed by theoretical bounds on the number of mistakes made by them. This paper proposes three algorithms to improve the mistakes bound in the unrealizable case. Proposed algorithms perform highly better than the existing ones in the long run when most of the input sequences presented to the learner are likely to be realizable.

cs.LG↗

arXiv2018

Chemically stabilized epitaxial wurtzite-BN thin film

We report on the chemically stabilized epitaxial w-BN thin film grown on c-plane sapphire by pulsed laser deposition under slow kinetic condition. Traces of no other allotropes such as cubic (c) or hexagonal (h) BN phases are present. Sapphire substrate plays a significant role in stabilizing the metastable w-BN from h-BN target under unusual PLD growth condition involving low temperature and pressure and is explained based on density functional theory calculation. The hardness and the elastic modulus of the w-BN film are 37 & 339 GPa, respectively measured by indentation along <0001> direction. The results are extremely promising in advancing the microelectronic and mechanical tooling industry.

cond-mat.mtrl-sci↗

arXiv2017

Leveraging web resources for keyword assignment to short text documents

Assigning relevant keywords to documents is very important for efficient retrieval, clustering and management of the documents. Especially with the web corpus deluged with digital documents, automation of this task is of prime importance. Keyword assignment is a broad topic of research which refers to tagging of document with keywords, key-phrases or topics. For text documents, the keyword assignment techniques have been developed under two sub-topics: automatic keyword extraction (AKE) and automatic key-phrase abstraction. However, the approaches developed in the literature for full text documents cannot be used to assign keywords to low text content documents like twitter feeds, news clips, product reviews or even short scholarly text. In this work, we point out several practical challenges encountered in tagging such low text content documents. As a solution to these challenges, we show that the proposed approaches which leverage knowledge from several open source web resources enhance the quality of the tags (keywords) assigned to the low text content documents. The performance of the proposed approach is tested on real world corpus consisting of scholarly documents with text content ranging from only the text in the title of the document (5-10 words) to the summary text/abstract (100- 150 words). We find that the proposed approach not just improves the accuracy of keyword assignment but offer a computationally efficient solution which can be used in real world applications.

cs.IR↗

arXiv2017

Joint Backhaul-Access Analysis of Full Duplex Self-Backhauling Heterogeneous Networks

With the successful demonstration of in-band full-duplex (IBFD) transceivers, a new research dimension has been added to wireless networks. This paper proposes an interesting use case of this capability for IBFD self-backhauling heterogeneous networks (HetNet). IBFD self-backhauling in a HetNet refers to IBFD-enabled small cells backhauling themselves with macro cells over the wireless channel. Owing to their IBFD capability, the small cells simultaneously communicate over the access and backhaul links, using the same frequency band. The idea is doubly advantageous, as it obviates the need for fiber backhauling small cells every hundred meters and allows the access spectrum to be reused for backhauling at no extra cost. This work considers the case of a two-tier cellular network with IBFD-enabled small cells, wirelessly backhauling themselves with conventional macro cells. For clear exposition, the case considered is that of FDD network, where within access and backhaul links, the downlink (DL) and uplink (UL) are frequency duplexed ($f1$, $f2$ respectively), while the total frequency spectrum used at access and backhaul ($f1+f2$) is the same. Analytical expressions for coverage and average downlink (DL) rate in such a network are derived using tools from the field of stochastic geometry. It is shown that DL rate in such networks could be close to double that of a conventional TDD/FDD self-backhauling network, at the expense of reduced coverage due to higher interference in IBFD networks. For the proposed IBFD network, the conflicting aspects of increased interference on one side and high spectral efficiency on the other are captured into a mathematical model. The mathematical model introduces an end-to-end joint analysis of backhaul (or fronthaul) and access links, in contrast to the largely available access-centric studies.

cs.NI↗

arXiv2016

A Linearization Technique for Self-Interference Cancellation in Full-Duplex Radios

The fundamental problem in the design of a full-duplex radio is the cancellation of the self-interference (SI) signal generated by the transmitter.Current techniques for suppressing SI rely on generating a copy of the SI signal and subtracting it partly in the RF (radio frequency) and digital domains. A critical step in replicating the self-interference is the estimation of the multi-path channel through which the transmitted signal propagates to the antenna. Since there is no prior model on the number of multipath reflections, current techniques assume a tap delay line filter (in the RF and digital domain) with a large number of taps, and estimate the taps in the analog and the digital domain. Assuming such a model leads to a large form-factor for the analog and RF circuits and increased complexity in the digital domain. In this paper, using a linearization technique, we show that the self-interference channel in an indoor environment can be effectively modelled as $H(f)=C_0 + C_1f$ in the frequency domain. Thus, the effective self-interference channel can be represented by two parameters $C_0$ and $C_1$, irrespective of the multipath environment. We also provide experimental evidence to verify the above channel model and propose novel low-complexity designs for self-interference cancellation. Linearization not only aids in the practicality of analog cancellation by reducing the form factor, but also results in a simpler SI filter model in the digital domain due to dimensionality reduction of the channel parameters. Therefore this method can enable the widespread adoption of full-duplex techniques to portable devices in addition to infrastructure base-stations.

cs.NI↗

arXiv2015

Predicting Small Group Accretion in Social Networks: A topology based incremental approach

Small Group evolution has been of central importance in social sciences and also in the industry for understanding dynamics of team formation. While most of research works studying groups deal at a macro level with evolution of arbitrary size communities, in this paper we restrict ourselves to studying evolution of small group (size $\leq20$) which is governed by contrasting sociological phenomenon. Given a previous history of group collaboration between a set of actors, we address the problem of predicting likely future group collaborations. Unfortunately, predicting groups requires choosing from $n \choose r$ possibilities (where $r$ is group size and $n$ is total number of actors), which becomes computationally intractable as group size increases. However, our statistical analysis of a real world dataset has shown that two processes: an external actor joining an existing group (incremental accretion (IA)) or collaborating with a subset of actors of an exiting group (subgroup accretion (SA)), are largely responsible for future group formation. This helps to drastically reduce the $n\choose r$ possibilities. We therefore, model the attachment of a group for different actors outside this group. In this paper, we have built three topology based prediction models to study these phenomena. The performance of these models is evaluated using extensive experiments over DBLP dataset. Our prediction results shows that the proposed models are significantly useful for future group predictions both for IA and SA.

cs.SI↗

arXiv2015

Ignorance is Almost Bliss: Near-Optimal Stochastic Matching With Few Queries

The stochastic matching problem deals with finding a maximum matching in a graph whose edges are unknown but can be accessed via queries. This is a special case of stochastic $k$-set packing, where the problem is to find a maximum packing of sets, each of which exists with some probability. In this paper, we provide edge and set query algorithms for these two problems, respectively, that provably achieve some fraction of the omniscient optimal solution. Our main theoretical result for the stochastic matching (i.e., $2$-set packing) problem is the design of an \emph{adaptive} algorithm that queries only a constant number of edges per vertex and achieves a $(1-ε)$ fraction of the omniscient optimal solution, for an arbitrarily small $ε>0$. Moreover, this adaptive algorithm performs the queries in only a constant number of rounds. We complement this result with a \emph{non-adaptive} (i.e., one round of queries) algorithm that achieves a $(0.5 - ε)$ fraction of the omniscient optimum. We also extend both our results to stochastic $k$-set packing by designing an adaptive algorithm that achieves a $(\frac{2}{k} - ε)$ fraction of the omniscient optimal solution, again with only $O(1)$ queries per element. This guarantee is close to the best known polynomial-time approximation ratio of $\frac{3}{k+1} -ε$ for the \emph{deterministic} $k$-set packing problem [Furer and Yu, 2013] We empirically explore the application of (adaptations of) these algorithms to the kidney exchange problem, where patients with end-stage renal failure swap willing but incompatible donors. We show on both generated data and on real data from the first 169 match runs of the UNOS nationwide kidney exchange that even a very small number of non-adaptive edge queries per vertex results in large gains in expected successful matches.

cs.DS↗

arXiv2014

Privacy-Preserving Public Information for Sequential Games

In settings with incomplete information, players can find it difficult to coordinate to find states with good social welfare. For example, in financial settings, if a collection of financial firms have limited information about each other's strategies, some large number of them may choose the same high-risk investment in hopes of high returns. While this might be acceptable in some cases, the economy can be hurt badly if many firms make investments in the same risky market segment and it fails. One reason why many firms might end up choosing the same segment is that they do not have information about other firms' investments (imperfect information may lead to `bad' game states). Directly reporting all players' investments, however, raises confidentiality concerns for both individuals and institutions. In this paper, we explore whether information about the game-state can be publicly announced in a manner that maintains the privacy of the actions of the players, and still suffices to deter players from reaching bad game-states. We show that in many games of interest, it is possible for players to avoid these bad states with the help of privacy-preserving, publicly-announced information. We model behavior of players in this imperfect information setting in two ways -- greedy and undominated strategic behaviours, and we prove guarantees on social welfare that certain kinds of privacy-preserving information can help attain. Furthermore, we design a counter with improved privacy guarantees under continual observation.

cs.GT↗