Search

Fine-tuning pretrained self-supervised language models is widely adopted for transfer learning to downstream tasks. Fine-tuning can be achieved by freezing gradients of the pretrained network and only updating gradients of a newly added classification layer, or by performing gradient updates on all parameters. Gradual unfreezing makes a trade-off between the two by gradually unfreezing gradients of whole layers during training. This has been an effective strategy to trade-off between storage and training speed with generalization performance. However, it is not clear whether gradually unfreezing layers throughout training is optimal, compared to sparse variants of gradual unfreezing which may improve fine-tuning performance. In this paper, we propose to stochastically mask gradients to regularize pretrained language models for improving overall fine-tuned performance. We introduce GradDrop and variants thereof, a class of gradient sparsification methods that mask gradients during the backward pass, acting as gradient noise. GradDrop is sparse and stochastic unlike gradual freezing. Extensive experiments on the multilingual XGLUE benchmark with XLMR-Large show that GradDrop is competitive against methods that use additional translated data for intermediate pretraining and outperforms standard fine-tuning and gradual unfreezing. A post-analysis shows how GradDrop improves performance with languages it was not trained on, such as under-resourced languages., Accepted to IJCNN 2023

Extended Reality (XR) is one of the most important 5G/6G media applications that will fundamentally transform human interactions. However, ensuring low latency, high data rate, and reliability to support XR services poses significant challenges. This letter presents a novel AI-assisted service provisioning scheme that leverages predicted frames for processing rather than relying solely on actual frames. This method virtually increases the network delay budget and consequently improves service provisioning, albeit at the expense of minor prediction errors. The proposed scheme is validated by extensive simulations demonstrating a multi-fold increase in supported XR users and also provides crucial network design insights.

We investigate the effects of post-training quantization and quantization-aware training on the generalization of Transformer language models. We present a new method called self-distilled quantization (SDQ) that minimizes accumulative quantization errors and outperforms baselines. We apply SDQ to multilingual models XLM-R-Base and InfoXLM-Base and demonstrate that both models can be reduced from 32-bit floating point weights to 8-bit integer weights while maintaining a high level of performance on the XGLUE benchmark. Our results also highlight the challenges of quantizing multilingual models, which must generalize to languages they were not fine-tuned on.

Despite being the current de-facto models in most NLP tasks, transformers are often limited to short sequences due to their quadratic attention complexity on the number of tokens. Several attempts to address this issue were studied, either by reducing the cost of the self-attention computation or by modeling smaller sequences and combining them through a recurrence mechanism or using a new transformer model. In this paper, we suggest to take advantage of pre-trained sentence transformers to start from semantically meaningful embeddings of the individual sentences, and then combine them through a small attention layer that scales linearly with the document length. We report the results obtained by this simple architecture on three standard document classification datasets. When compared with the current state-of-the-art models using standard fine-tuning, the studied method obtains competitive results (even if there is no clear best model in this configuration). We also showcase that the studied architecture obtains better results when freezing the underlying transformers. A configuration that is useful when we need to avoid complete fine-tuning (e.g. when the same frozen transformer is shared by different applications). Finally, two additional experiments are provided to further evaluate the relevancy of the studied architecture over simpler baselines.

We report the measurement of hyperfine splitting in the $7d$ $^{2}D_{3/2}$ state of $^{133}$Cs using high resolution Doppler-free two-photon spectroscopy in a Cs vapor cell. We determine the hyperfine coupling constants $A = 7.3509(9)$ MHz and $B = -0.041(8)$ MHz, which represent an order of magnitude improvement in the precision. We also obtain bounds on the magnitude of the nuclear magnetic octupole coupling constant $C$. Additionally, we measure the ac Stark shift of the $6s$ $^{2}S_{1/2} \rightarrow 7d$ $^{2}D_{3/2}$ transition at 767.8 nm to be $-49 \pm 5$ Hz/(W/cm$^2$), in agreement with theoretical calculations. We further report the measurement of collisional shift [$-32.6 \pm 2.0$ kHz/mTorr] and pressure broadening for the individual hyperfine levels of the $6s$ $^{2}S_{1/2} \rightarrow 7d$ $^{2}D_{3/2}$ transition. These measurements provide valuable inputs for analysis of systematic effects in optical frequency standards based on the cesium $6s$ $^{2}S_{1/2} \rightarrow 7d$ $^{2}D_{3/2}$ two-photon transition, 14 pages, 7 figures, to appear in Physical Review A

This paper describes a new approach for approximating the inverse kinematics of a manipulator using an Ant Colony Optimization (ACO) based RBFN (Radial Basis Function Network). In this paper, a training solution using the ACO and the LMS (Least Mean Square) algorithm is presented in a two-phase training procedure. To settle the problem that the cluster results of k-mean clustering Radial Basis Function (RBF) are easy to be influenced by the selection of initial characters and converge to a local minimum, Ant Colony Optimization (ACO) for the RBF neural networks which will optimize the center of RBF neural networks and reduce the number of the hidden layer neurons nodes is presented. The result demonstrates that the accuracy of Ant Colony Optimization for the Radial Basis Function (RBF) neural networks is higher, and the extent of fitting has been improved.

Supplement 1: contains plots of HFS measured at different laser powers and notes on global fitting of hyperfine coupling constants

Aspect Based Sentiment Analysis is a dominant research area with potential applications in social media analytics, business, finance, and health. Prior works in this area are primarily based on supervised methods, with a few techniques using weak supervision limited to predicting a single aspect category per review sentence. In this paper, we present an extremely weakly supervised multi-label Aspect Category Sentiment Analysis framework which does not use any labelled data. We only rely on a single word per class as an initial indicative information. We further propose an automatic word selection technique to choose these seed categories and sentiment words. We explore unsupervised language model post-training to improve the overall performance, and propose a multi-label generator model to generate multiple aspect category-sentiment pairs per review sentence. Experiments conducted on four benchmark datasets showcase our method to outperform other weakly supervised baselines by a significant margin., Comment: to be published in EMNLP 2022

We describe the EdinSaar submission to the shared task of Multilingual Low-Resource Translation for North Germanic Languages at the Sixth Conference on Machine Translation (WMT2021). We submit multilingual translation models for translations to/from Icelandic (is), Norwegian-Bokmal (nb), and Swedish (sv). We employ various experimental approaches, including multilingual pre-training, back-translation, fine-tuning, and ensembling. In most translation directions, our models outperform other submitted systems., Comment: To be published WMT2021

Pruning aims to reduce the number of parameters while maintaining performance close to the original network. This work proposes a novel \emph{self-distillation} based pruning strategy, whereby the representational similarity between the pruned and unpruned versions of the same network is maximized. Unlike previous approaches that treat distillation and pruning separately, we use distillation to inform the pruning criteria, without requiring a separate student network as in knowledge distillation. We show that the proposed {\em cross-correlation objective for self-distilled pruning} implicitly encourages sparse solutions, naturally complementing magnitude-based pruning criteria. Experiments on the GLUE and XGLUE benchmarks show that self-distilled pruning increases mono- and cross-lingual language model performance. Self-distilled pruned models also outperform smaller Transformers with an equal number of parameters and are competitive against (6 times) larger distilled networks. We also observe that self-distillation (1) maximizes class separability, (2) increases the signal-to-noise ratio, and (3) converges faster after pruning steps, providing further insights into why self-distilled pruning improves generalization.

Frequently-Asked-Question (FAQ) retrieval provides an effective procedure for responding to user's natural language based queries. Such platforms are becoming common in enterprise chatbots, product question answering, and preliminary technical support for customers. However, the challenge in such scenarios lies in bridging the lexical and semantic gap between varied query formulations and the corresponding answers, both of which typically have a very short span. This paper proposes TI-S2S, a novel learning framework combining TF-IDF based keyword extraction and Word2Vec embeddings for training a Sequence-to-Sequence (Seq2Seq) architecture. It achieves high precision for FAQ retrieval by better understanding the underlying intent of a user question captured via the representative keywords. We further propose a variant with an additional neural network module for guiding retrieval via relevant candidate identification based on similarity features. Experiments on publicly available dataset depict our approaches to provide around 92% precision-at-rank-5, exhibiting nearly 13% improvement over existing approaches., Comment: 6 pages

Silicon ferroelectric field-effect transistors (FeFETs) with low-k interfacial layer (IL) between ferroelectric gate stack and silicon channel suffers from high write voltage, limited write endurance and large read-after-write latency due to early IL breakdown and charge trapping and detrapping at the interface. We demonstrate low voltage, high speed memory operation with high write endurance using an IL-free back-end-of-line (BEOL) compatible FeFET. We fabricate IL-free FeFETs with 28nm channel length and 126nm width under a thermal budget

In this article we study the presence of chaos in supersymmetric(SUSY) quantum mechanics. For that purpose we present a form of 4-point out of time order correlator(OTOC) for SUSY quantum mechanical systems using tensor product formalism. We calculate the 4-point OTOC for SUSY 1D harmonic oscillator and find that the OTOC is exactly equal to that of 1D bosonic harmonic oscillator system. In similar manner using the eigenstate representation of supersymmetric systems we calculate the generalized higher order out of time order correlator. The higher order OTOC is a more sensitive measure of chaos than the usual 4-point correlator used in literature. Finally, we calculate the generalized 2N-point OTOC for SUSY 1D harmonic oscillator.

In this article we study the presence of chaos in supersymmetric (SUSY) quantum mechanical systems. We present a form of 4-point out-of-time-order correlator (OTOC) for SUSY quantum mechanical systems using both Tensor Product and Partner Hamiltonian formalisms. We calculate the 4-point OTOC for SUSY 1D harmonic oscillator and find that the OTOC is precisely equal to that of the 1D bosonic harmonic oscillator system. Using the eigenstate representation of supersymmetric systems, we extend the definition for generalised higher-order out-of-time-order correlators. The higher-order OTOC is a more sensitive measure of chaos than the usual 4-point correlator used in the literature. We present a compact form of the generalised 2N-point OTOC in SUSY quantum mechanics using both formalisms. Finally, we calculate the generalised 2N-point OTOC for SUSY 1D harmonic oscillator and show their equivalence., 8 pages

Path planning in dynamic environments is essential to high-risk applications such as unmanned aerial vehicles, self-driving cars, and autonomous underwater vehicles. In this paper, we generate collision-free trajectories for a robot within any given environment with temporal and spatial uncertainties caused due to randomly moving obstacles. We use two Poisson distributions to model the movements of obstacles across the generated trajectory of a robot in both space and time to determine the probability of collision with an obstacle. Measures are taken to avoid an obstacle by intelligently manipulating the speed of the robot at space-time intervals where a larger number of obstacles intersect the trajectory of the robot. Our method potentially reduces the use of computationally expensive collision detection libraries. Based on our experiments, there has been a significant improvement over existing methods in terms of safety, accuracy, execution time and computational cost. Our results show a high level of accuracy between the predicted and actual number of collisions with moving obstacles.

Combinatorial optimization problems arise in a wide range of applications from diverse domains. Many of these problems are NP-hard and designing efficient heuristics for them requires considerable time and experimentation. On the other hand, the number of optimization problems in the industry continues to grow. In recent years, machine learning techniques have been explored to address this gap. We propose a framework for leveraging machine learning techniques to scale-up exact combinatorial optimization algorithms. In contrast to the existing approaches based on deep-learning, reinforcement learning and restricted Boltzmann machines that attempt to directly learn the output of the optimization problem from its input (with limited success), our framework learns the relatively simpler task of pruning the elements in order to reduce the size of the problem instances. In addition, our framework uses only interpretable learning models based on intuitive features and thus the learning process provides deeper insights into the optimization problem and the instance class, that can be used for designing better heuristics. For the classical maximum clique enumeration problem, we show that our framework can prune a large fraction of the input graph (around 99 % of nodes in case of sparse graphs) and still detect almost all of the maximum cliques. This results in several fold speedups of state-of-the-art algorithms. Furthermore, the model used in our framework highlights that the chi-squared value of neighborhood degree has a statistically significant correlation with the presence of a node in a maximum clique, particularly in dense graphs which constitute a significant challenge for modern solvers. We leverage this insight to design a novel heuristic for this problem outperforming the state-of-the-art. Our heuristic is also of independent interest for maximum clique detection and enumeration., Comment: Integrates three works which appeared at AAAI'19 [arXiv:1902.08455], the DSO workshop at IJCAI'19 [arXiv:1910.00517] and CIKM'19

The switching dynamics of a single-domain BiFeO3/CoFe heterojunction is modeled and key parameters such as interface exchange coupling coefficient are extracted from experimental results. The lower limit of the magnetic order response time of CoFe in the BiFeO3/CoFe heterojunction is theoretically quantified to be on to the order of 100 ps. Our results indicate that the switching behavior of CoFe in the BiFeO3/CoFe heterojunction is dominated by the rotation of the Neel vector in BiFeO3 rather than the unidirectional exchange bias at the interface. We also quantify the magnitude of the interface exchange coupling coefficient J_int to be 0.32 pJ/m by comparing our simulation results with the giant magnetoresistance (GMR) curves and the magnetic hysteresis loop in the experiments. To the best of our knowledge, this is the first time that J_int is extracted quantitatively from experiments. Furthermore, we demonstrate that the switching success rate and the thermal stability of the BiFeO3/CoFe heterojunction can be improved by reducing the thickness of CoFe and increasing the length to width aspect ratio of the BiFeO3/CoFe heterojunction. Our theoretical model provides a comprehensive framework to study the magnetoelectric properties and the manipulation of the magnetic order of CoFe in the BiFeO3/CoFe heterojunction.

Computationally hard problems, including combinatorial optimization, can be mapped into the problem of finding the ground-state of an Ising Hamiltonian. Building physical systems with collective computational ability and distributed parallel processing capability can accelerate the ground-state search. Here, we present a continuous-time dynamical system (CTDS) approach where the ground-state solution appears as stable points or attractor states of the CTDS. We harness the emergent dynamics of a network of phase-transition nano-oscillators (PTNO) to build an Ising Hamiltonian solver. The hardware fabric comprises of electrically coupled injection-locked stochastic PTNOs with bi-stable phases emulating artificial Ising spins. We demonstrate the ability of the stochastic PTNO-CTDS to progressively find more optimal solution by increasing the strength of the injection-locking signal - akin to performing classical annealing. We demonstrate in silico that the PTNO-CTDS prototype solves a benchmark non-deterministic polynomial time (NP)-hard Max-Cut problem with high probability of success. Using experimentally calibrated numerical simulations and incorporating non-idealities, we investigate the performance of our Ising Hamiltonian solver on dense Max-Cut problems with increasing graph size. We report a high energy-efficiency of 1.3x10^7 solutions/sec/Watt for 100-node dense Max-cut problems which translates to a 5x improvement over the recently demonstrated memristor-based Hopfield network and several orders of magnitude improvement over other candidates such as CPU and GPU, quantum annealer and photonic Ising solver approaches. Such an energy efficient hardware exhibiting high solution-throughput/Watt can find applications in industrial planning and manufacturing, defense and cyber-security, bioinformatics and drug discovery., Comment: 23 pages, 6 figures, 1 table

Cross-lingual alignment of word embeddings play an important role in knowledge transfer across languages, for improving machine translation and other multi-lingual applications. Current unsupervised approaches rely on similarities in geometric structure of word embedding spaces across languages, to learn structure-preserving linear transformations using adversarial networks and refinement strategies. However, such techniques, in practice, tend to suffer from instability and convergence issues, requiring tedious fine-tuning for precise parameter setting. This paper proposes BioSpere, a novel framework for unsupervised mapping of bi-lingual word embeddings onto a shared vector space, by combining adversarial initialization and refinement procedure with point set registration algorithm used in image processing. We show that our framework alleviates the shortcomings of existing methodologies, and is relatively invariant to variable adversarial learning performance, depicting robustness in terms of parameter choices and training losses. Experimental evaluation on parallel dictionary induction task demonstrates state-of-the-art results for our framework on diverse language pairs.

Multiplicative stochasticity such as Dropout improves the robustness and generalizability of deep neural networks. Here, we further demonstrate that always-on multiplicative stochasticity combined with simple threshold neurons are sufficient operations for deep neural networks. We call such models Neural Sampling Machines (NSM). We find that the probability of activation of the NSM exhibits a self-normalizing property that mirrors Weight Normalization, a previously studied mechanism that fulfills many of the features of Batch Normalization in an online fashion. The normalization of activities during training speeds up convergence by preventing internal covariate shift caused by changes in the input distribution. The always-on stochasticity of the NSM confers the following advantages: the network is identical in the inference and learning phases, making the NSM suitable for online learning, it can exploit stochasticity inherent to a physical substrate such as analog non-volatile memories for in-memory computing, and it is suitable for Monte Carlo sampling, while requiring almost exclusively addition and comparison operations. We demonstrate NSMs on standard classification benchmarks (MNIST and CIFAR) and event-based classification benchmarks (N-MNIST and DVS Gestures). Our results show that NSMs perform comparably or better than conventional artificial neural networks with the same architecture.

There has been great success recently in tackling challenging NLP tasks by neural networks which have been pre-trained and fine-tuned on large amounts of task data. In this paper, we investigate one such model, BERT for question-answering, with the aim to analyze why it is able to achieve significantly better results than other models. We run DeepLIFT on the model predictions and test the outcomes to monitor shift in the attention values for input. We also cluster the results to analyze any possible patterns similar to human reasoning depending on the kind of input paragraph and question the model is trying to answer., 6 pages, 13 figures

Environmental and economic concerns promote research on designing energy-efficient Time and Wavelength Division Multiplexed Ethernet Passive Optical Network (TWDM-EPON), which is the future extension to TDM-EPON. In TDM-EPON, a plethora of research is already present to achieve energy savings at Optical Network Units (ONUs) which can easily be applied for TWDM-EPON ONUs. However, TWDM-EPON provides an additional opportunity for saving energy at the Optical Line Terminal (OLT). All existing protocols have primarily been designed for saving energy at the OLT receivers. The protocols to save energy at the OLT receives depends only on the Up-Stream(US) traffic scheduling while its transmitter counterpart depends on both US and Down-Stream (DS) scheduling since the OLT transmits GATE message along with DS traffic. The US and DS scheduling have a basic difference. The MAC protocol doesn't allow scheduling of US traffic of an ONU after its REPORT arrival at multiple disjoint time slots. However, this restriction is absent for DS traffic and hence, the grant-size of an ONU can be partitioned and every part can be scheduled at different times. In this paper, we propose a method for saving energy at the OLT transmitters in TWDM-EPON while satisfying the SLAs. This includes a heuristic algorithm to partition the DS grant and schedule them. Through extensive simulations, we demonstrate that the proposed method provides a significant improvement in energy efficiency as compared to existing protocols (up to 45%).

We propose a multi-stage learning approach for pruning the search space of maximum clique enumeration, a fundamental computationally difficult problem arising in various network analysis tasks. In each stage, our approach learns the characteristics of vertices in terms of various neighborhood features and leverage them to prune the set of vertices that are likely not contained in any maximum clique. Furthermore, we demonstrate that our approach is domain independent -- the same small set of features works well on graph instances from different domain. Compared to the state-of-the-art heuristics and preprocessing strategies, the advantages of our approach are that (i) it does not require any estimate on the maximum clique size at runtime and (ii) we demonstrate it to be effective also for dense graphs. In particular, for dense graphs, we typically prune around 30 \% of the vertices resulting in speedups of up to 53 times for state-of-the-art solvers while generally preserving the size of the maximum clique (though some maximum cliques may be lost). For large real-world sparse graphs, we routinely prune over 99 \% of the vertices resulting in several tenfold speedups at best, typically with no impact on solution quality., Appeared at the Data Science Meets Optimization Workshop (DSO) at IJCAI'19

Economical and environmental concerns necessitate research on designing energy-efficient optical access network especially Ethernet Passive Optical Network (EPON) which is one of the most widely accepted and deployed last-mile access network. In this paper, our primary focus is on designing a protocol for saving energy at Optical Network Units (ONUs) while satisfying the Service Label Agreement (SLA). The SLA of both Up-Stream (US) and Down-Stream (DS) traffic can be satisfied only if the EPON network can react to their instantaneous load change during sleep periods of ONUs and to the best of our knowledge, there doesn't exist any such proposal. Towards this target, we propose a mechanism that allows the Optical Line Terminal (OLT) to force ONUs to wake-up from sleep mode. Here, we demonstrate that if the OLT can distribute the active ONUs (transceivers are active) fairly among cycles then it provides a significant improvement in energy-efficiency. To achieve this, we formulate an ILP for fairly distributing active ONUs among cycles while satisfying the SLA of both US and DS traffic at the same time. A polynomial time $1$-approximation algorithm is proposed for solving this ILP. The convergence and the complexity analysis of the algorithm are also performed. Extensive simulations depict that fair distribution of ONUs reduces the power consumption and average delay figure at the same time and the reduction increases with an increment of the number of ONUs and round-trip time.

The present work deals with scalar field cosmology in the framework of a quantum gravity modified Einstein-Hilbert Lagrangian with variable $G$ and $��$. Using Renormalization group, variable $G$ behaves as a minimally coupled filed (not the scalar-tensor theory) and variable $��$ can be interpreted as a potential function. The point Lagrangian for this model in the background of homogeneous and isotropic flat FLRW space-time model experiences point-like Noether symmetry and equivalent potential function $��(G)$ is determined. Using a point transformation in the $3D$ augmented space is found that one of the variable become cyclic and as a consequence there is considerable simplification to the physical system. Lastly, the constants of motion can be written in compact form and it is possible to have analytic cosmological solutions in the present context., 10 pages, 6 figures

It is very challenging to work with low-resource languages due to the inadequate availability of data. Using a dictionary to map independently trained word embeddings into a shared vector space has proved to be very useful in learning bilingual embeddings in the past. Here we have tried to map individual embeddings of words in English and their corresponding translated words in low-resource languages like Estonian, Slovenian, Slovakian, and Hungarian. We have used a supervised learning approach. We report accuracy scores through various retrieval strategies which show that it is possible to approach challenging tasks in Natural Language Processing like machine translation for such languages, provided that we have at least some amount of proper bilingual data. We also conclude that we can follow an unsupervised learning path on monolingual text data as that is more suitable for low-resource languages., Comment: 7 pages, 4 tables

Magnetic skyrmions have been the focus of intense research with promising applications in memory, logic and interconnect technology. Several schemes have been recently proposed and demonstrated to nucleate skyrmions. However, they either result in an uncontrolled skyrmion bubble production or are mostly targeted towards integration with racetrack memory device. In this work, we propose a novel scheme for a controlled single skyrmion nucleation in a confined nanowire geometry with sub-100 nm width using a generalized approach of "localized spin current injection" technique in material systems exhibiting low Dzyaloshinskii-Moriya interaction (DMI). Our proposed nucleation mechanism follows a pathway involving the creation of a reversed magnetic domain containing one or more pairs of vertical Bloch lines (VBLs) that form an edge-to-edge domain wall as the VBLs get annihilated at the edge of the nanowire. However, pinning of the edge domain walls within a narrow gap using notches or anti-notches results in the creation of a magnetic bubble with defect-free domain wall that eventually relaxes into a circular skyrmion structure. Our simulations predict that the proposed mechanism allows skyrmion nucleation on sub-nanosecond timescale, shows robustness to variations like local pinning sites and is applicable for any skyrmion-based logic, memory and interconnect application.

The paper deals with $f(R)$ gravity theory in the background of inhomogeneous FLRW--type space time model. With proper choice of the inhomogeneous metric function it is possible to have an emergent scenario for the $f(R)$--cosmology. Explicit form of $f(R)$ is obtained for power law expansion. It has been shown that the present $f(R)$ gravity model is equivalent to some particle creation mechanism in Einstein gravity. Further a complete cosmic evolution from inflation to present late time acceleration is possible with proper continuous choices of the $f(R)$--functions. Finally, in the perspective of thermodynamical analysis a form of $f(R)$ has been evaluated using the unified first law.

Design of energy-efficient access networks has emerged as an important area of research, since access networks consume $80-90\%$ of the overall Internet power consumption. TWDM-PON is envisaged to be one of the widely accepted future access technologies. TWDM-PON offers an additional opportunity to save energy at the OLT along with the existing energy-efficient ONU design. In this paper, we focus on the energy-efficient OLT design in a TWDM-PON. While most of the conventional methods employ a minimization of the number of wavelengths, we propose a novel approach which aims at minimizing the number of voids created due to scheduling. In the process, for the first time, we present a low-complexity on-line scheduling algorithm for the upstream traffic considering delay constraints. Our extensive simulations demonstrate a significant improvement in energy efficiency of $\sim 25\%$ for high load at the OLT receivers. Furthermore, we provide an analytical upper-bound on the energy-efficiency of the OLT receivers and demonstrate that the proposed protocol achieves an energy efficiency very close to the bound with a maximum deviation $\sim 2\%$ for $64$ ONUs.

Online reviews provide viewpoints on the strengths and shortcomings of products/services, influencing potential customers' purchasing decisions. However, the proliferation of non-credible reviews -- either fake (promoting/ demoting an item), incompetent (involving irrelevant aspects), or biased -- entails the problem of identifying credible reviews. Prior works involve classifiers harnessing rich information about items/users -- which might not be readily available in several domains -- that provide only limited interpretability as to why a review is deemed non-credible. This paper presents a novel approach to address the above issues. We utilize latent topic models leveraging review texts, item ratings, and timestamps to derive consistency features without relying on item/user histories, unavailable for "long-tail" items/users. We develop models, for computing review credibility scores to provide interpretable evidence for non-credible reviews, that are also transferable to other domains -- addressing the scarcity of labeled data. Experiments on real-world datasets demonstrate improvements over state-of-the-art baselines.

The direct photodissociation of trapped $^{85}$Rb$_2^+$ (rubidium) molecular ions by the cooling light for the $^{85}$Rb magneto-optical trap (MOT) is studied, both experimentally and theoretically. Vibrationally excited Rb$_{2}^{+}$ ions are created by photoionization of Rb$_{2}$ molecules formed photoassociatively in the Rb MOT and are trapped in a modified spherical Paul trap. The decay rate of the trapped Rb$_{2}^{+}$ ion signal in the presence of the MOT cooling light is measured and agreement with our calculated rates for molecular ion photodissociation is observed. The photodissociation mechanism due to the MOT light is expected to be active and therefore universal for all homonuclear diatomic alkali metal molecular ions.

Many Web applications require efficient querying of large Knowledge Graphs (KGs). We propose KOGNAC, a dictionary-encoding algorithm designed to improve SPARQL querying with a judicious combination of statistical and semantic techniques. In KOGNAC, frequent terms are detected with a frequency approximation algorithm and encoded to maximise compression. Infrequent terms are semantically grouped into ontological classes and encoded to increase data locality. We evaluated KOGNAC in combination with state-of-the-art RDF engines, and observed that it significantly improves SPARQL querying on KGs with up to 1B edges.

We extend our recently developed inversion method to extract excited state potentials from fluorescence line positions and line strengths. We consider a previous limitation of the method arising due to insufficient input data in cases where the relatively weaker emission data are not experimentally available. We develop a solution to this problem by "regenerating" these weak transition lines via applying a model potential, e.g. a Morse potential. The result of this procedure, illustrated for the Q-branch emission from the lowest three vibrational levels of the B($^1 \Pi)$ state of LiRb, is shown to have an error of $0.29$ cm$^{-1}$ in the classically allowed region and a global error of $5.67$ cm$^{-1}$ for $V\le E(\nu'=10)$. The robustness of this procedure is also demonstrated by considering the statistical error in the measured line intensities.

Heteronuclear polar molecules have recently attracted enormous attention owing to their ground state having a large electric dipole moment. The long range anisotropic dipole-dipole interaction in such systems is the basis for a variety of applications including quantum computing, precision measurements, ultracold chemistry and quantum simulations. Heteronuclear bi-alkali molecules, only a small subset of polar molecules, have received special attention mainly because the constituent alkali atoms are easy to laser cool and can be relatively easily associated to form molecules at ultracold temperatures. Our choice, the LiRb molecule, is motivated by the relatively high dipole moment (4.1 Debye) of the LiRb molecule in its rovibronic ground state. In this thesis, we study the LiRb molecule using laser spectroscopy and report, for the first time, the production of ultracold LiRb molecules by photoassociation (PA). The LiRb molecule is the least studied among all bi-alkali molecules and the first spectroscopic measurements on hot vapor phase LiRb molecules were reported only in 2011. We describe these measurements and their significance in chapter 2, after a brief introduction to the field of ultracold polar molecules in chapter 1. In chapter 3, we discuss our apparatus for simultaneous cooling and trapping of Li and Rb atoms in a dual-species magneto-optical trap (MOT) and report the measurement of interspecies collision-induced losses. In chapter 4, we describe the production of ultracold LiRb molecules in excited electronic states by photoassociation (PA). We report the measurements of the C6 coefficients for the Li (2s 2S1/2) + Rb (5p 2P1/2) and the Li (2s 2S1/2) + Rb (5p 2P3/2) asymptotes. We find a molecule formation rate (PLiRb) as high as 3.5x107 s-1 and a PA rate coefficient (KPA) as high as 1.3x10-10 cm3/s, the highest among heteronuclear bi-alkali molecules. In chapter 5, we discuss results on two-photon PA and we conclude, in chapter 6, with a road roadmap of future experiments for the production and detection of ultracold LiRb molecules in their rovibronic ground state.

We report the production of ultracold 7Li85Rb molecules by photoassociation (PA) below the Li (2s 2S1/2) + Rb (5p 2P1/2) asymptote. We perform PA spectroscopy in a dual-species 7Li-85Rb magneto-optical trap (MOT) and detect the PA resonances using trap loss spectroscopy. We observe several strong PA resonances corresponding to the last few bound states, assign the lines and derive the long range C6 dispersion coefficients for the Li (2s 2S1/2) + Rb (5p 2P1/2) asymptote. We also report an excited-state molecule formation rate (P_LiRb) of ~10^7 s^-1 and a PA rate coefficient (K_PA) of ~4x10^-11 cm^3/s, which are both among the highest observed for heteronuclear bi-alkali molecules. These suggest that PA is a promising route for the creation of ultracold ground state LiRb molecules., Comment: 6 pages

Applications involving telecommunication call data records, web pages, online transactions, medical records, stock markets, climate warning systems, etc., necessitate efficient management and processing of such massively exponential amount of data from diverse sources. De-duplication or Intelligent Compression in streaming scenarios for approximate identification and elimination of duplicates from such unbounded data stream is a greater challenge given the real-time nature of data arrival. Stable Bloom Filters (SBF) addresses this problem to a certain extent. . In this work, we present several novel algorithms for the problem of approximate detection of duplicates in data streams. We propose the Reservoir Sampling based Bloom Filter (RSBF) combining the working principle of reservoir sampling and Bloom Filters. We also present variants of the novel Biased Sampling based Bloom Filter (BSBF) based on biased sampling concepts. We also propose a randomized load balanced variant of the sampling Bloom Filter approach to efficiently tackle the duplicate detection. In this work, we thus provide a generic framework for de-duplication using Bloom Filters. Using detailed theoretical analysis we prove analytical bounds on the false positive rate, false negative rate and convergence rate of the proposed structures. We exhibit that our models clearly outperform the existing methods. We also demonstrate empirical analysis of the structures using real-world datasets (3 million records) and also with synthetic datasets (1 billion records) capturing various input distributions., 41 pages

The tremendous expanse of search engines, dictionary and thesaurus storage, and other text mining applications, combined with the popularity of readily available scanning devices and optical character recognition tools, has necessitated efficient storage, retrieval and management of massive text databases for various modern applications. For such applications, we propose a novel data structure, INSTRUCT, for efficient storage and management of sequence databases. Our structure uses bit vectors for reusing the storage space for common triplets, and hence, has a very low memory requirement. INSTRUCT efficiently handles prefix and suffix search queries in addition to the exact string search operation by iteratively checking the presence of triplets. We also propose an extension of the structure to handle substring search efficiently, albeit with an increase in the space requirements. This extension is important in the context of trie-based solutions which are unable to handle such queries efficiently. We perform several experiments portraying that INSTRUCT outperforms the existing structures by nearly a factor of two in terms of space requirements, while the query times are better. The ability to handle insertion and deletion of strings in addition to supporting all kinds of queries including exact search, prefix/suffix search and substring search makes INSTRUCT a complete data structure., Comment: International Conference on Management of Data (COMAD), 2010

Allocation of balls into bins is a well studied abstraction for load balancing problems.The literature hosts numerous results for sequential(single dimensional) allocation case when m balls are thrown into n bins. In this paper we study the symmetric multiple choice process for both unweighted and weighted balls as well as for both multidimensional and scalar models.Additionally,we present the results on bounds on gap for (1+beta) choice process with multidimensional balls and bins. We show that for the symmetric d choice process and with m=O(n), the upper bound on the gap is O(lnln(n)) w.h.p.This upper bound on the gap is within D=f factor of the lower bound. This is the first such tight result.For the general case of m>>n the expected gap is bounded by O(lnln(n)).For variable f and non-uniform distribution of the populated dimensions,we obtain the upper bound on the expected gap as O(log(n)). Further,for the multiple round parallel balls and bins,we show that the gap is also bounded by O(loglog(n)) for m=O(n).The same bound holds for the expected gap when m>>n. Our analysis also has strong implications in the sequential scalar case.For the weighted balls and bins and general case m>>n,we show that the upper bound on the expected gap is O(log(n)) which improves upon the best prior bound of n^c.Moreover,we show that for the (1 + beta) choice process and m=O(n) the upper bound(assuming uniform distribution of f populated dimensions over D total dimensions) on the gap is O(log(n)/beta),which is within D=f factor of the lower bound.For fixed f with non-uniform distribution and for random f with Binomial distribution the expected gap remains O(log(n)/beta) independent of the total number of balls thrown. This is the first such tight result for (1 +beta) paradigm with multidimensional balls and bins.

Balanced allocation of online balls-into-bins has long been an active area of research for efficient load balancing and hashing applications.There exists a large number of results in this domain for different settings, such as parallel allocations~\cite{parallel}, multi-dimensional allocations~\cite{multi}, weighted balls~\cite{weight} etc. For sequential multi-choice allocation, where $m$ balls are thrown into $n$ bins with each ball choosing $d$ (constant) bins independently uniformly at random, the maximum load of a bin is $O(\log \log n) + m/n$ with high probability~\cite{heavily_load}. This offers the current best known allocation scheme. However, for $d = ��(\log n)$, the gap reduces to $O(1)$~\cite{soda08}.A similar constant gap bound has been established for parallel allocations with $O(\log ^*n)$ communication rounds~\cite{lenzen}. In this paper we propose a novel multi-choice allocation algorithm, \emph{Improved D-choice with Estimated Average} ($IDEA$) achieving a constant gap with a high probability for the sequential single-dimensional online allocation problem with constant $d$. We achieve a maximum load of $\lceil m/n \rceil$ with high probability for constant $d$ choice scheme with \emph{expected} constant number of retries or rounds per ball. We also show that the bound holds even for an arbitrary large number of balls, $m>>n$. Further, we generalize this result to (i)~the weighted case, where balls have weights drawn from an arbitrary weight distribution with finite variance, (ii)~multi-dimensional setting, where balls have $D$ dimensions with $f$ randomly and uniformly chosen filled dimension for $m=n$, and (iii)~the parallel case, where $n$ balls arrive and are placed parallely in the bins. We show that the gap in these case is also a constant w.h.p. (independent of $m$) for constant value of $d$ with expected constant number of retries per ball.

We have studied the X 1��+ and B 1�� states of 7Li85Rb using Laser Induced Fluorescence (LIF) spectroscopy and Fluorescence Excitation Spectroscopy (FES). We extract molecular constants for levels v" = 0-2 of the X 1��+ state and levels v' = 0-20 of the B 1�� state. For the B 1�� state, we have observed rotational perturbations in the e-parity component of the v' = 2 level, and determined the dissociation energy. We discuss implications of our measurements in finding efficient photoassociation pathways for production of ultra-cold ground state LiRb molecules, and their detection via state selective ionization., 20 pages, 6 figures, 2 tables, List of all experimentally observed transitions

Allocation of balls into bins is a well studied abstraction for load balancing problems.The literature hosts numerous results for sequential(single dimensional) allocation case when m balls are thrown into n bins. In this paper we study the symmetric multiple choice process for both unweighted and weighted balls as well as for both multidimensional and scalar models.Additionally,we present the results on bounds on gap for (1+beta) choice process with multidimensional balls and bins. We show that for the symmetric d choice process and with m=O(n), the upper bound on the gap is O(lnln(n)) w.h.p.This upper bound on the gap is within D=f factor of the lower bound. This is the first such tight result.For the general case of m>>n the expected gap is bounded by O(lnln(n)).For variable f and non-uniform distribution of the populated dimensions,we obtain the upper bound on the expected gap as O(log(n)). Further,for the multiple round parallel balls and bins,we show that the gap is also bounded by O(loglog(n)) for m=O(n).The same bound holds for the expected gap when m>>n. Our analysis also has strong implications in the sequential scalar case.For the weighted balls and bins and general case m>>n,we show that the upper bound on the expected gap is O(log(n)) which improves upon the best prior bound of n^c.Moreover,we show that for the (1 + beta) choice process and m=O(n) the upper bound(assuming uniform distribution of f populated dimensions over D total dimensions) on the gap is O(log(n)/beta),which is within D=f factor of the lower bound.For fixed f with non-uniform distribution and for random f with Binomial distribution the expected gap remains O(log(n)/beta) independent of the total number of balls thrown. This is the first such tight result for (1 +beta) paradigm with multidimensional balls and bins.