arXiv subjects

Abhishek Singh

Publications and source records attributed to Abhishek Singh.

90 records · Page 5Linked to original sources

arXiv2020

Ownership preserving AI Market Places using Blockchain

We present a blockchain based system that allows data owners, cloud vendors, and AI developers to collaboratively train machine learning models in a trustless AI marketplace. Data is a highly valued digital asset and central to deriving business insights. Our system enables data owners to retain ownership and privacy of their data, while still allowing AI developers to leverage the data for training. Similarly, AI developers can utilize compute resources from cloud vendors without loosing ownership or privacy of their trained models. Our system protocols are set up to incentivize all three entities - data owners, cloud vendors, and AI developers to truthfully record their actions on the distributed ledger, so that the blockchain system provides verifiable evidence of wrongdoing and dispute resolution. Our system is implemented on the Hyperledger Fabric and can provide a viable alternative to centralized AI systems that do not guarantee data or model privacy. We present experimental performance results that demonstrate the latency and throughput of its transactions under different network configurations where peers on the blockchain may be spread across different datacenters and geographies. Our results indicate that the proposed solution scales well to large number of data and model owners and can train up to 70 models per second on a 12-peer non optimized blockchain network and roughly 30 models per second in a 24 peer network.

cs.DC↗

arXiv2019

Clinical XLNet: Modeling Sequential Clinical Notes and Predicting Prolonged Mechanical Ventilation

Clinical notes contain rich data, which is unexploited in predictive modeling compared to structured data. In this work, we developed a new text representation Clinical XLNet for clinical notes which also leverages the temporal information of the sequence of the notes. We evaluated our models on prolonged mechanical ventilation prediction problem and our experiments demonstrated that Clinical XLNet outperforms the best baselines consistently.

cs.CL↗

arXiv2019

NASIB: Neural Architecture Search withIn Budget

Neural Architecture Search (NAS) represents a class of methods to generate the optimal neural network architecture and typically iterate over candidate architectures till convergence over some particular metric like validation loss. They are constrained by the available computation resources, especially in enterprise environments. In this paper, we propose a new approach for NAS, called NASIB, which adapts and attunes to the computation resources (budget) available by varying the exploration vs. exploitation trade-off. We reduce the expert bias by searching over an augmented search space induced by Superkernels. The proposed method can provide the architecture search useful for different computation resources and different domains beyond image classification of natural images where we lack bespoke architecture motifs and domain expertise. We show, on CIFAR10, that itis possible to search over a space that comprises of 12x more candidate operations than the traditional prior art in just 1.5 GPU days, while reaching close to state of the art accuracy. While our method searches over an exponentially larger search space, it could lead to novel architectures that require lesser domain expertise, compared to the majority of the existing methods.

cs.LG↗

arXiv2019

Maximal adversarial perturbations for obfuscation: Hiding certain attributes while preserving rest

In this paper we investigate the usage of adversarial perturbations for the purpose of privacy from human perception and model (machine) based detection. We employ adversarial perturbations for obfuscating certain variables in raw data while preserving the rest. Current adversarial perturbation methods are used for data poisoning with minimal perturbations of the raw data such that the machine learning model's performance is adversely impacted while the human vision cannot perceive the difference in the poisoned dataset due to minimal nature of perturbations. We instead apply relatively maximal perturbations of raw data to conditionally damage model's classification of one attribute while preserving the model performance over another attribute. In addition, the maximal nature of perturbation helps adversely impact human perception in classifying hidden attribute apart from impacting model performance. We validate our result qualitatively by showing the obfuscated dataset and quantitatively by showing the inability of models trained on clean data to predict the hidden attribute from the perturbed dataset while being able to predict the rest of attributes.

cs.LG↗

arXiv2019

Detailed comparison of communication efficiency of split learning and federated learning

We compare communication efficiencies of two compelling distributed machine learning approaches of split learning and federated learning. We show useful settings under which each method outperforms the other in terms of communication efficiency. We consider various practical scenarios of distributed learning setup and juxtapose the two methods under various real-life scenarios. We consider settings of small and large number of clients as well as small models (1M - 6M parameters), large models (10M - 200M parameters) and very large models (1 Billion-100 Billion parameters). We show that increasing number of clients or increasing model size favors split learning setup over the federated while increasing the number of data samples while keeping the number of clients or model size low makes federated learning more communication efficient.

cs.LG↗

arXiv2019

Giant longitudinal negative magneto-resistance under perpendicular magnetic field in Bi$_{2-x}$Fe$_x$Se$_{3-x}$S$_x$ Topological insulators

The magnetic, magneto-transport and ARPES studies of Fe and S co-doped Bi2Se3 were investigated. With doping concentration magneto-resistance (MR) gradually decreases and for a certain doping concentration giant negative MR is observed which persists up to room temperature. Magnetic measurement indicates that the negative MR is observed when ferromagnetic ordering is induced with Fe doping. The magnetic ordering can be attributed with the RKKY interaction. Positive MR reappears with larger doping concentration which may be attributed to the decrease of FM ordering due to the turning off of the spin-orbit coupling leading to the destruction of non-trivial bulk state. This in-effect de-hybridizes the conduction band with the Fe spin. The ARPES data also indicates that above a critical doping concentration (x>0.09) the non-trivial bulk state is completely destroyed.

cond-mat.str-el↗

arXiv2019

Room temperature large spontaneous exchange bias in hard-soft antiferromagnetic composite BiFeO3-TbMnO3

We report the presence of giant spontaneous exchange bias (HSEB) in a hard and soft antiferromagnetic composite of BiFeO3-TbMnO3 (BFO-TMO in 7:3 and 8:2 ratio). The HSEB varies between 5-778Oe, but persists up to room temperature with a maximum near a spin reorientation transition temperature observed from magnetization vs. temperature measurement in Zero-field cooled (ZFC) and Field cooled (FC) modes. Isothermal remnant magnetization measurements at room temperature indicate the presence of an interfacial layer of a 2 dimensional dilute antiferromagnet in a field (2D DAFF). A stable value of the exchange bias has been observed via training effect measurements which signify the role of interfacial exchange coupling in the system. Based on the experimental results we explain the presence of the giant spontaneous exchange bias on the basis of a strong strain-mediated magnetoelectriccoupling induced exchange interaction and the creation of 2D DAFF layer at the interface. Theproperties of this layer are defined by canting and pinning of BFO spins at the interface with TMO due to Fe and Mn interaction. X-ray Magnetic Circular Dichroism (XMCD) confirms the presence of canted antiferromagnetic ordering of BiFeO3, charge transfer between Mn ions and different magnetically coupled layers which play vital role in getting the exchange bias.

cond-mat.str-el↗

arXiv2018

"When and Where?": Behavior Dominant Location Forecasting with Micro-blog Streams

The proliferation of smartphones and wearable devices has increased the availability of large amounts of geospatial streams to provide significant automated discovery of knowledge in pervasive environments, but most prominent information related to altering interests have not yet adequately capitalized. In this paper, we provide a novel algorithm to exploit the dynamic fluctuations in user's point-of-interest while forecasting the future place of visit with fine granularity. Our proposed algorithm is based on the dynamic formation of collective personality communities using different languages, opinions, geographical and temporal distributions for finding out optimized equivalent content. We performed extensive empirical experiments involving, real-time streams derived from 0.6 million stream tuples of micro-blog comprising 1945 social person fusion with graph algorithm and feed-forward neural network model as a predictive classification model. Lastly, The framework achieves 62.10% mean average precision on 1,20,000 embeddings on unlabeled users and surprisingly 85.92% increment on the state-of-the-art approach.

cs.SI↗

arXiv2018

Fast Neural Architecture Construction using EnvelopeNets

Fast Neural Architecture Construction (NAC) is a method to construct deep network architectures by pruning and expansion of a base network. In recent years, several automated search methods for neural network architectures have been proposed using methods such as evolutionary algorithms and reinforcement learning. These methods use a single scalar objective function (usually accuracy) that is evaluated after a full training and evaluation cycle. In contrast NAC directly compares the utility of different filters using statistics derived from filter featuremaps reach a state where the utility of different filters within a network can be compared and hence can be used to construct networks. The training epochs needed for filters within a network to reach this state is much less than the training epochs needed for the accuracy of a network to stabilize. NAC exploits this finding to construct convolutional neural nets (CNNs) with close to state of the art accuracy, in < 1 GPU day, faster than most of the current neural architecture search methods. The constructed networks show close to state of the art performance on the image classification problem on well known datasets (CIFAR-10, ImageNet) and consistently show better performance than hand constructed and randomly generated networks of the same depth, operators and approximately the same number of parameters.

cs.NE↗

arXiv2018

Anomalous and Topological Hall effect in Cu doped Sb2Te3 Topological Insulator

The magneto-transport and magnetization measurements of Sb1.90Cu0.10Te3 were performed at different temperatures and different fields. Magneto-transport measurement at high field indicates the coexistence of both bulk and surface states. The magnetization shows the induced antiferromagnetic ordering with Cu doping and the observed quantum oscillation in it indicates that magnetization in Sb1.90Cu0.10Te3 is the bulk property. The non linearity in Hall data suggests the existence of anomalous and topological Hall effect. The anomalous and topological Hall effect (THE) from measured hall data of Cu doped Sb2Te3 topological insulator have been evaluated.

cond-mat.mes-hall↗

arXiv2017

The Evolution of Kicked Stellar-Mass Black Holes in Star Cluster Environments

We consider how dynamical friction acts on black holes that receive a velocity kick while located at the center of a gravitational potential, analogous to a star cluster, due to either a natal kick or the anisotropic emission of gravitational waves during a black hole-black hole merger. Our investigation specifically focuses on how well various Chandrasekhar-based dynamical friction models can predict the orbital decay of kicked black holes with $m_{bh} \lesssim 100 M_\odot$ due to an inhomogeneous background stellar field. In general, the orbital evolution of a kicked black hole follows that of a damped oscillator where two-body encounters and dynamical friction serve as sources of damping. However, we find models for approximating the effects of dynamical friction do not accurately predict the amount of energy lost by the black hole if the initial kick velocity $v_{k}$ is greater than the stellar velocity dispersion $σ$. For all kick velocities, we also find that two-body encounters with nearby stars can cause the energy evolution of a kicked BH to stray significantly from standard dynamical friction theory as encounters can sometimes lead to an energy gain. For larger kick velocities, we find the orbital decay of a black hole departs from classical theory completely as the black hole's orbital amplitude decays linearly with time as opposed to exponentially. Therefore, we have developed a linear decay formalism which scales linearly with black hole mass and $\frac{v_{k}}σ$ in order to account for the variations in the local gravitational potential.

astro-ph.GA↗

arXiv2017

High temperature spin-freezing transition in Pyrochlore Eu2Ti2O7: A new observation from ac-susceptibility

The magnetization (both dc and ac) of pyrochlore Eu2Ti2O7 have been investigated. Ac susceptibility measurements reveal a new unusual spin freezing below 35K for pure Eu2Ti2O7. Frequency dependence of these ac peak positions indicate slow spin relaxation near this temperature and it follows the Arrhenius formula suggesting a thermally activated relaxation process. The origin of this spin freezing has been attributed essentially to a single ion process which is associated to Eu3+ spin relaxation. Non-magnetic dilution by Y3+ ions also confirms the single ion freezing.

cond-mat.mtrl-sci↗

arXiv2017

Certificate Transparency with Enhancements and Short Proofs

Browsers can detect malicious websites that are provisioned with forged or fake TLS/SSL certificates. However, they are not so good at detecting malicious websites if they are provisioned with mistakenly issued certificates or certificates that have been issued by a compromised certificate authority. Google proposed certificate transparency which is an open framework to monitor and audit certificates in real time. Thereafter, a few other certificate transparency schemes have been proposed which can even handle revocation. All currently known constructions use Merkle hash trees and have proof size logarithmic in the number of certificates/domain owners. We present a new certificate transparency scheme with short (constant size) proofs. Our construction makes use of dynamic bilinear-map accumulators. The scheme has many desirable properties like efficient revocation, low verification cost and update costs comparable to the existing schemes. We provide proofs of security and evaluate the performance of our scheme.

cs.CR↗

arXiv2016

Change-over of carrier type and magneto-transport property in Cu doped Bi2Te3 Topological Insulators

Structural, resistivity, thermoelectric power and magneto-transport properties of Cu doped Bi2Te3 topological insulators have been investigated. The occurrence of the tuning of charge carriers from n type to p type by Cu doping at Te sites of Bi2Te3 is observed both from Hall effect and thermoelectric power measurements. Carrier mobility decreases with the doping of Cu which provides evidence of the movement of Fermi level from bulk conduction band to the bulk valence band. Thermoelectric power also increaseswith doping of Cu.Moreover linear magnetoresistance (LMR) has been observed at high magnetic field in pure Bi2Te3 which is associated to the gapless topological surface states protected by time reversal symmetry (TRS), whereas doping of Cu breaks TRS and an opening of band gap occurs which quenches the LMR.

cond-mat.mtrl-sci↗

arXiv2014

Carbon Irradiated SI-GaAs for Photoconductive THz Detection

We report here a photoconductive material for THz generation and detection with sub-picosecond carrier lifetime made by C12 (Carbon) irradiation on commercially available semi-insulating (SI) GaAs. We are able to reduce the carrier lifetime of SI-GaAs down to sub-picosecond by irradiating it with various irradiation dosages of Carbon (C12) ions. With an increase of the irradiation dose from ~1012 /cm2 to ~1015 /cm2 the carrier lifetime of SI-GaAs monotonously decreases to 0.55 picosecond, resulting in strongly improved THz pulse detection compared with normal SI-GaAs.

cond-mat.mtrl-sci↗

arXiv2014

Microlensless Interdigitated Photoconductive Terahertz Emitters

We report here fabrication of interdigitated photoconductive antenna (iPCA) terahertz (THz) emitters based on plasmonic electrode design. Novel design of this iPCA enables it to work without microlens array focusing, which is otherwise required for photo excitation of selective photoconductive regions to avoid the destructive interference of emitted THz radiation from oppositely biased regions. Benefit of iPCA over single active region PCA is that photo excitation can be done at larger area, hence avoiding the saturation effected at higher optical excitation density. The emitted THz radiation power from plasmonic-iPCAs is ~ 2 times more than the single active region plasmonic PCA at 200 mW optical excitation, which will further increase at higher optical powers. This design is expected to reduce fabrication cost of photoconductive THz sources and detectors.

cond-mat.mes-hall↗

arXiv2013

Highly Efficient and Electrically Robust Carbon Irradiated SI-GaAs Based Photoconductive THz Emitters

We demonstrate here an efficient THz source with low electrical power consumption. We have increased the maximum THz radiation power emitted from SI-GaAs based photoconductive emitters by two orders of magnitude. By irradiating the SI-GaAs substrate with Carbon-ions up to 2 micron deep, we have created lot of defects and decreased the life time of photo-excited carriers inside the substrate. Depending on the irradiation dose we find 1 to 2 orders of magnitude decrease in total current flowing in the substrate, resulting in subsequent decrease of heat dissipation in the antenna. This has resulted in increasing maximum cut-off of the applied voltage across Photo-Conductive Emitter (PCE) electrodes to operate the device without thermal breakdown from ~35 V to > 150 V for the 25 micron electrode gaps. At optimum operating conditions, carbon irradiated (10^14 ions/cm^2) PCEs give THz pulses with power about 100 times higher in comparison to the usual PCEs on SI-GaAs and electrical to THz power conversion efficiency has improved by a factor of ~ 800.

cond-mat.mtrl-sci↗

arXiv2013

Improved Efficiency of Photoconductive THz Emitters by Increasing the Effective Contact Length of Electrodes

We study the effect of a surface modification at the interface between metallic electrodes and semiconducting substrate in Semi-Insulating GaAs (SI-GaAs) based photoconductive emitters (PCE) on the emission of Tera-Hertz (THz) radiation. We partially etch out 500 nm thick layer of SI-GaAs in grating like pattern with various periods before the contact deposition. By depositing the electrodes on the patterned surface, the electrodes follow the contour of the grating period. This increases the effective contact length of the electrodes per unit area of the active regions on the PCE. The maxima of the electric field amplitude of the THz pulses emitted from the patterned surface are enhanced by up to more than a factor 2 as compared to an un-patterned surface. We attribute this increase to the increase of the effective contact length of the electrode due to surface patterning.

cond-mat.mes-hall↗