Indian Institute of Science Bangalore
etd@IISc Electronic Theses and Dissertations at Indian Institute of ScienceNot a member yet
6204 research outputs found
Sort by
Learning Filters, Filterbanks, Wavelets and Multiscale Representations
No full textThe problem of filter design is ubiquitous. Frequency selective filters are used in speech/audio processing, image analysis, convolutional neural networks for tasks such as denoising, deblurring/deconvolution, enhancement, compression, etc. While traditional filter design methods use a structured optimization formulation, the advent of deep learning techniques and associated tools and toolkits enables the learning of filters through data-driven optimization. In this thesis, we consider the filter design problem in a learning setting in both data-dependent and data-independent flavors. Data-dependent filters have properties governed by a downstream task, for instance, filters in a convolutional dictionary used for the task of image denoising. On the contrary, data-independent filters have constraints imposed on their frequency responses, such as lowpass, having diamond-shaped support, satisfying perfect reconstruction property, ability to generate wavelet functions, etc.
The contributions of this thesis are four-fold: (i) the formulation of filter, filterbank, and wavelet design as regression problems, allowing them to be designed in a learning framework; (ii) the design of contourlet-based scattering networks for image classification; (iii) the design of a deep unfolded network using composite regularization techniques for solving inverse problems in image processing; and (iv) a multiscale dictionary learning algorithm that learns one or more multiscale generator kernels to parsimoniously explain certain neural recordings.
We begin by developing learning approaches for designing filters having data-independent specifications, for instance, filters with a specified frequency response, including an ideal filter. The problem of designing such filters is formulated as a regression problem, using a training set comprising cosine signals with frequencies sampled uniformly at random. The filters are optimized using the mean-squared error loss, and generalization bounds are provided. We demonstrate the applicability of our approach for filters such as lowpass, bandpass, and highpass in 1-D, and diamond, fan and checkerboard support filters in 2-D. We then show how the methodology extends easily for designing 1-D and 2-D cosine modulated filterbanks.
Second, we consider the problems of 1-D filterbank and wavelet design through learning. Wavelets have proven to be highly successful in several signal and image processing applications. Wavelet design has been an active field of research for over two decades, with the problem often being approached analytically. We draw a parallel between convolutional autoencoders and wavelet multiresolution approximation and show how the learning angle provides a coherent computational framework for solving the design problem. We design data-independent wavelets by interpreting the corresponding perfect reconstruction filterbanks as autoencoders (what we refer to as “filterbank autoencoders”), which precludes the need for customized datasets. In fact, we show that it is possible to design them efficiently using high-dimensional Gaussian vectors as training data. Generalization bounds show that a near-zero training loss implies that the learnt filters satisfy the perfect reconstruction property with a very high probability. We show that desirable properties of a wavelet such as orthogonality, compact support, smoothness, symmetry, and vanishing moments can all be incorporated into the proposed framework by means of architectural constraints or by introducing suitable regularization functionals to the MSE cost. Notably, our approach not only recovers the well-known Daubechies family of orthogonal wavelets and the Cohen-Daubechies-Feauveau (CDF) family of symmetric biorthogonal wavelets, which are used in JPEG-2000 compression, but also learns new wavelets outside these families.
Third, we extend the ideas used for 1-D filterbank and wavelet learning to 2-D filterbank and wavelet design. A variety of efficient representations of natural images, such as wavelets and contourlets can be formulated as corresponding filterbank design problems. The design constraints on the continuous-domain wavelets have corresponding filter-domain manifestations. While most learning problems require specialized datasets, we employ 2-D random Gaussian matrices as training data and optimize filter coefficients considering the MSE loss. Design specifications such as orthogonality of the filterbank, perfect reconstruction property, symmetry, and vanishing moments are enforced through an appropriate parameterization of the convolutional units. We demonstrate several examples of learning biorthogonal and orthogonal filterbanks and wavelets having a specified number of vanishing moments, both point vanishing moments and directional vanishing moments, and symmetry constraints. Sparse recovery via composite regularization is an interesting approach proposed recently in the literature. One could design non-convex regularizers through a convex combination of sparsity-promoting penalties with known proximal operators. We develop a new algorithm, namely, convolutional proximal-averaged thresholding algorithm (C-PATA) for {\it composite-regularized} convolutional sparse coding (CR-CSC) based on the recently proposed idea of proximal averaging. We develop an autoencoder structure based on the deep-unfolding of C-PATA iterations into neural network layers, which results in the composite-regularized neural network (CoRNet) architecture. The convolutional learned iterative soft-thresholding algorithm becomes a special case of CoRNet. We demonstrate the efficacy of CoRNet considering applications to image denoising and inpainting, and compare the performance with state-of-the-art techniques such as BM3D, convolutional LISTA, and fast and flexible convolutional sparse coding (FFCSC).
The data-independent filter design technique is employed to learn a contourlet transform used within a hybrid scattering network. Hybrid scattering networks are convolutional neural networks (CNNs) where the first few layers implement a fixed windowed scattering transform, while the rest of the network is learned. Scattering networks outperform state-of-the-art deep learning models for limited-data classification tasks although the performance gains are not much for large datasets. The 2-D Morlet filterbank used in Mallat's scattering network is replaced by a contourlet filterbank, which provides sparser representations and better frequency-domain directional separation. The contourlet transform comprises a multiresolution pyramidal filterbank cascaded with directional filters. We construct directional filters using diamond-shaped quincunx filterbanks and consider two pyramidal filter variants -- square-shaped, and filters with radially isotropic frequency domain support. The performance of all variants is evaluated for natural image classification tasks on CIFAR-10 and ImageNet datasets. We show that the radial contourlet variant achieves competitive performance compared with the Morlet scattering transform on large-dataset classification tasks while performing better for the limited-dataset scenario.
We then switch over to the problem of learning data-dependent filters for sparse recovery by employing a combination of sparsity promoting regularizers. Sparse recovery via such composite regularization approaches is an interesting framework proposed recently in the literature. One could design non-convex regularizers through a convex combination of sparsity-promoting penalties with known proximal operators. We developed a new algorithm, namely, convolutional proximal-averaged thresholding algorithm (C-PATA) for composite-regularized convolutional sparse coding (CR-CSC) based on proximal averaging. We develop an autoencoder structure based on the deep-unfolding of C-PATA iterations into neural network layers, which results in the composite-regularized neural network (CoRNet) architecture. The convolutional learned iterative soft-thresholding algorithm becomes a special case of CoRNet. We demonstrate the efficacy of CoRNet considering applications to image denoising and inpainting and compare the performance with state-of-the-art techniques such as BM3D, convolutional LISTA, and fast and flexible convolutional sparse coding (FFCSC).
Finally, we conclude by developing a data-dependent method to learn filters generating a multiscale convolutional dictionary. First, the multiscale convolutional dictionary learning (MCDL) algorithm is proposed to extract a representative waveform shape from a given dataset. The proposed algorithm is based on the popularly used convolutional dictionary learning formulation with a crucial difference -- we assume that the learned atoms are scaled versions of a single generator kernel. We evaluate kernel recovery for synthetic data under noiseless and noisy data conditions. A smoothness regularizer on the learned atom is used to aid better kernel recovery under noisy conditions. Kernel recovery is shown to be robust to model choices of scales and the assumed support size of the kernel without any restrictive assumptions. The proposed approach is applied to visualizing the typical patterns present within human electrocorticogram (ECoG) measurements. The validation is carried out using publicly available ECoG data recorded from a single Parkinson's disease patient.
This thesis thus presents a cogent framework for learning filters, filterbanks, wavelets, and convolutional and multiscale dictionaries
OTFS Modulation in MIMO Visible Light Communication Systems
No full textVisible light communication (VLC) technology is emerging as an attractive technology for wireless communication in indoor and vehicular environments. In VLS systems, light emitting diodes (LED) and photo diodes (PD) are used for wireless signal transmission and reception, respectively, using the visible light spectrum. VLC systems are gaining popularity because of their ability to provide lighting and short-range communication simultaneously. Recently, a new modulation scheme called orthogonal time frequency space (OTFS) modulation has been introduced in the radio frequency (RF) communications domain, where it has been shown that OTFS achieves significantly better performance compared to the widely used orthogonal frequency division multiplexing (OFDM). OTFS is a two-dimensional (2D) modulation scheme, where MN information symbols are multiplexed in the delay-Doppler (DD) domain using M Doppler bins and N delay bins. This is in contrast to OFDM where information symbols are multiplexed in the time-frequency (TF) domain. Because of the demonstrated superiority of OTFS in the RF communication domain, research in OTFS for VLC systems has gained importance recently. In this context, it is of interest to 1) adapt OTFS waveform to VLC settings, and 2) investigate the performance OTFS in indoor VLC environments. In this thesis, we propose efficient multi-LED OTFS modulation schemes suited for visible light communication and analyze their bit error performance in indoor communication environments.
First, we propose two dual-LED transmission schemes, namely, non-DC-biased OTFS (NDC-OFTS) scheme and dual-LED complex modulation OTFS (DCM-OTFS) scheme. The NDC-OTFS scheme uses 2D Hermitian symmetry operation to convert complex signals to positive, real-valued signals suitable for transmission in the optical domain, and the DCM-OTFS scheme achieves this by exploiting the polar representation of complex signals. We obtain analytical upper bounds on the bit error performance of the proposed schemes, which are found to be tight at high signal-to-noise ratios (SNRs). Our analytical and simulation results show that the proposed OTFS schemes achieve better performance compared to their OFDM counterparts reported in the VLC literature. Further, using the ratio of the minimum distance of different normalized received signal sets as a metric, we quantify the spatial distribution of the SNR gain of the OTFS schemes compared to their OFDM counterparts.
Next, we propose two quad-LED architectures, namely, quad-LED complex modulation OTFS (QCM-OTFS) scheme and spatial modulation DCM-OTFS (SM-DCM-OTFS) scheme. The QCM-OTFS scheme sends the magnitudes of real and imaginary parts of complex signals through intensity modulation and their sign information through spatial indexing of LEDs. The proposed SM-DCM-OTFS scheme sends the magnitude and phase of complex signals (polar representation) through a pair of LEDs and frame indexing across two pairs of LEDs. We also propose two multi-LED OTFS schemes, namely, quadrature spatial modulation OTFS (QSM-OTFS) scheme and dual mode index modulation OTFS (DMIM-OTFS) scheme. The proposed schemes use the dual-LED complex modulation block as the basic building block and offer enhanced rates compared to conventional index modulation schemes. The proposed schemes have the advantage of not requiring the Hermitian symmetry and DC bias operations to obtain real, positive-valued signals suited for intensity modulation of LEDs. Our results show that the proposed OTFS based schemes perform better than their OFDM counterparts. Finally, to mitigate the effect of channel correlation and achieve improved system performance, we investigate the use of the pairwise coding technique and singular value decomposition precoding in multiple-input multiple-output (MIMO) VLC systems. Our simulation results show that the pairwise coding is effective in alleviating the effect of spatial correlation
The dual-targeting of Saccharomyces cerevisiae Pso2 into the nucleus and mitochondria is regulated by post-translational modifications: Implications for organelle-specific DNA repair
No full textThe repair of DNA interstrand crosslinks (ICLs) involves the concerted action of multiple DNA repair pathways. In eukaryotes, ICLs are sensed and repaired by replication-coupled or replication-independent mechanisms in which PSO2/SNM1 is essential for unhooking an ICL. Previous work has shown that ICL-inducing agents cause damage to Saccharomyces cerevisiae nuclear and mitochondrial DNA (mtDNA), and its pso2/snm1 mutants exhibit a petite phenotype followed by loss of mtDNA integrity and copy number. Complex as it is, the cause and underlying molecular mechanisms remains elusive. In the first part of this study, we provide direct evidence that the S. cerevisiae nuclear-encoded Pso2 is a dual-localized ICL repair factor: it is imported into mitochondria and the nucleus, and its abundance increases in the mitochondria following treatment with DNA damaging agents. We reveal that Pso2 contains one mitochondrial targeting sequence (MTS) and two nuclear localization signals (NLS1 and NLS2) in its N-terminus, although NLS1 resides within the MTS. While MTS is necessary to guide the passage of Pso2 into the mitochondrial matrix, either NLS1 or NLS2 is sufficient for its import into the nucleus, implying that the two NLS motifs are functionally redundant. Ablation of MTS abrogated mitochondrial import of Pso2, and concomitant enrichment in the nucleus. Strikingly, mutations that render both NLS1 and NLS2 inactive, blocked nuclear import of Pso2; at the same time, increased its mitochondrial localization, consistent with the notion that the relationship between MTS and NLS motifs is competitive. However, the nuclease activity of import-deficient species of Pso2 was not impaired. The potential relevance of dual targeting of Pso2 to two genome-bearing organelles is discussed
In the second part of this study, we combined a range of methods and approaches to determine the cell-intrinsic and extrinsic factors that regulate the intracellular distribution and function of Pso2 in ICL repair. Indeed, these approaches demonstrated that Pso2 is modified by ubiquitination and phosphorylation (but not SUMOylation) under normal growth conditions. Notably, we found that SUMOylation of Pso2 and its subsequent import into the mitochondria was induced by methyl methanesulfonate (MMS), but not cisplatin. We also demonstrate that the levels of MMS-induced Pso2 SUMOylation remain relatively constant throughout the synchronized cell cycle. Mutation of lysine residues (K97 and K575) within the SUMO consensus motifs in Pso2 abolished its import into the mitochondria. We further show that SIZ1 and SIZ2 SUMO E3 ligases act redundantly to catalyze SUMOylation of Pso2, while the former plays a dominant role in this process. Quite unexpectedly, we found that PSO1 (but not PSO2) was essential for attenuation of MMS-induced cell death; this implies that they function in non-overlapping ICL repair pathways or the former is epistatic to the latter. The findings reveal novel mechanistic insights into the regulation and intracellular distribution of Pso2 and shed new light on its function in DNA repair
Model-based Safe Deep Reinforcement Learning and Empirical Analysis of Safety via Attribution
No full textDuring initial iterations of training in most Reinforcement Learning (RL) algorithms, agents
perform a significant number of random exploratory steps, which in the real-world limit the
practicality of these algorithms as this can lead to potentially dangerous behavior. Hence safe
exploration is a critical issue in applying RL algorithms in the real world. This problem is well
studied in the literature under the Constrained Markov Decision Process (CMDP) Framework,
where in addition to single-stage rewards, state transitions receive single-stage costs as well.
The prescribed cost functions are responsible for mapping undesirable behavior at any given
time-step to a scalar value. Then we aim to find a feasible policy that maximizes reward returns
and keeps cost returns below a prescribed threshold during training as well as deployment.
We propose a novel On-policy Model-based Safe Deep RL algorithm in which we learn the
transition dynamics of the environment in an online manner as well as find a feasible optimal
policy using Lagrangian Relaxation-based Proximal Policy Optimization. This combination
of transition dynamics learning and a safety-promoting RL algorithm leads to 3-4 times less
environment interactions and less cumulative hazard violations compared to the model-free ap-
proach. We use an ensemble of neural networks with different initializations to tackle epistemic
and aleatoric uncertainty issues faced during environment model learning. We present our
results on a challenging Safe Reinforcement Learning benchmark - the Open AI Safety Gym.
In addition to this, we perform an attribution analysis of actions taken by the Deep Neural
Network-based policy at each time step. This analysis helps us to :
1. Identify the feature in state representation which is significantly responsible for the current
action.
2. Empirically provide the evidence of the safety-aware agent’s ability to deal with hazards in
the environment provided that hazard information is present in the state representation.
In order to perform the above analysis, we assume state representation has meaningful
information about hazards and goals. Then we calculate an attribution vector of the
same dimension as state using a well-known attribution technique known as Integrated
Gradients. The resultant attribution vector provides the importance of each state feature
for the current action.N
Strain Rate, Volume Fraction and Notch Depth Effects on Fracture Properties of Fibrous Cementitious Composites
No full textIt is known that concrete is weak in tension. Steel Fibre Reinforced Concrete (SFRC) is an alternative cementitious composite to increase the tensile strength, energy absorption capacity and durability of concrete structures. The variation of tensile stress (σ) developed in the Fracture Process Zone (FPZ) of SFRC with crack width (w) constitutes σ - w curves. The nonlinear tensile response of SFRC can be characterized using σ - w curves. Generally, the σ - w curves are obtained by direct method such as uniaxial tensile test or preferably by Three Point Bending (TPB) test using inverse analysis. However, the variation of tensile stress with crack width can be determined using Digital Image Correlation (DIC) technique and Acoustic Emission (AE) testing.
The DIC technique was used to simulate the Olesen’s cracked hinge model to locate the position of neutral axis using displacement contours in x - direction. The crack initiation point was identified using Maximum strain component in x - direction. The deformation parameters such as crack width, Crack Mouth Opening Displacement (CMOD), nonlinear hinge displacement in x - direction, neutral axis depth, curvature, cracking tensile strain, tensile strain and compressive strain were determined using DIC technique. Thus, a DIC based method was introduced to obtain above mentioned deformation parameters to determine σ - w curves. An empirical relation was introduced between experimentally obtained tensile stress and crack width to generalize σ - w relation.
Soranakom and Mobasher’s analytical model to obtain σ - ε relation of FRC in tensile and compressive region was used to determine σ - w relation for the validation of DIC based method. The strain related parameters such as µ, ω, β and were determined using DIC technique to obtain neutral axis depth ratio - normalized compressive strain response, normalized moment - normalized curvature response, Load (P) - CMOD response. The tensile strength was determined using DIC based method to compare with tensile strength predicted by Soranakom and Mobasher’s model.
A damage based model was introduced where the variation of tensile damage indices corresponding to the cementitious matrix cracking and steel fibre bridging with normalized tensile strain was obtained from cumulative AE energy of hits corresponding to tensile cracking and shear cracking respectively. The parameter a and parameter b were introduced to study the influence of steel fiber volume fraction (Vf) on the increment of tensile stress corresponding to cementitious matrix cracking and steel fibre bridging respectively.
The influence of Vf, strain rate, notch to depth ratio on the fracture energy, elastic modulus, ultimate peak load (Pu), midspan deflection and CMOD at Pu, fictitious crack tip displacement, crack propagation rate, parameter a, parameter b, tensile stress and damage indices corresponding to the cementitious matrix cracking and steel fibre bridging were studied under Mode I fracture. The σ - w curves were used to determine P - CMOD response using sigw - Concrete© to validate the DIC based method and damage based model
Probabilistic design and optimization of foundations on cohesionless soil using surrogate models
No full textThe design of foundations depends on the properties of soil, which are affected by various forms of uncertainties. These uncertainties affect the foundation performance and hence are considered in the deterministic design using safety factors. Such factors may not be able to cover all design scenarios involving different degrees of variation of soil properties and hence may lead to an inefficient design. These variabilities can be assessed and quantified systematically using a probabilistic framework to provide an optimized design with a consistent level of safety using analytical or numerical models. The use of numerical modelling techniques enhances the analytical capabilities by providing numerical solutions considering interactions of the different elements involved, employing appropriate constitutive models and input properties. It also facilitates the modelling of random processes and random fields by providing random properties as inputs to the model and can be used for analysis of geotechnical structures. Thus, the conjunction of probabilistic and numerical modelling techniques provides enormous possibilities for considering variabilities in foundation design and its optimization. But the computational cost associated with the use of probabilistic methods and uncertainty assessment has been a constant hindrance to their incorporation in the foundation design methods.
This thesis proposes a methodology utilizing probabilistic techniques with analytical and numerical models for the design and optimization of foundations. The probabilistic technique incorporates the variability of the soil properties such as cohesion, angle of friction, soil modulus, and unit weight for design and analysis. The issue of computational cost is overcome by using surrogate modelling techniques that replace the numerical models to improve computational efficiency. Thus, the variability of soil properties is considered in the numerical models using surrogate-based numerical modelling techniques for the probabilistic design of foundations. The optimization algorithm developed for different discrete values of design dimensions can be used to arrive at the optimum design curve from which the final design dimensions can be chosen based on the desired level of failure probability. Further, the possibility of improving the computational efficiency of the surrogate-based design and optimization methodology using advanced techniques are explored.
The first objective of the thesis deals with the design and analysis of a shallow foundation resting on cohesionless soil to study the influence of different forms of variabilities on foundation performance. The reliability-based design is performed by considering the mean value of the soil properties, their COV, and the variation of COV. The robust geotechnical design methodology is modified to incorporate system reliability along with the consideration of variability of soil properties to simplify the multi-objective design problem. Constrained optimization is performed to determine the variability of failure probability of the foundation and its cost and obtain the optimized design curve. The results from the robust design show the importance of consideration of variations of COV in highly variable soil deposits as it affects the final design. Further, spatial variability studies are performed to assess their contribution to the foundation performance by considering numerical methods in conjunction with a random field modelling technique known as Karhunen-Loeve expansion. The spatial variability study considering different COV values and correlation distances of the soil properties shows the importance of incorporating them in the design and analysis of foundations as it is observed to affect the bearing capacity of foundations.
The second objective deals with developing a reliability-based design methodology to optimize foundations using numerical modelling techniques by utilizing surrogate models. Kriging models are used as surrogate models due to their efficiency in capturing the stochastic nature of the problem and good predictive capabilities. Kriging models are developed for different foundation dimensions, and their performance is assessed using the statistics of bearing capacity of foundation or failure probability. The methodology developed is explained with the example of a shallow foundation. The methodology is extended to study the variation of different factors and their effects on foundation design. The variation of COV of the soil properties and correlation among them are examined to study their effect on the failure probability of the foundation. The effect of B/L ratio on the failure probability is also analyzed to understand its importance in foundation design. The reliability-based design is also extended to the estimation of partial factors by comparing the design and mean values of the soil properties obtained from the reliability analysis. This can help to understand the effect of variability on the partial factors as well. Finally, the optimization algorithm is used to arrive at the design curve from which the final design dimensions can be chosen, corresponding to the desired level of reliability.
As the third objective, the reliability-based design methodology using kriging surrogate models is developed and applied to geogrid reinforced soil, thereby providing an opportunity to consider the variability of soil and geogrid properties in the design of foundations. Geogrid reinforcement can be considered when the limit states are not satisfied due to constraints in foundation dimensions. Three-dimensional numerical modelling is utilized, facilitating the consideration of the foundation shape and extent of reinforcement to be provided. The optimized design curve is obtained for the discrete foundation dimensions using the optimization algorithm, from which the design that satisfies the desired failure probability can be obtained. The study shows the applicability of the design and optimization methodology for various problems by using surrogate-based numerical modelling techniques. Reliability analysis also enables the estimation of partial factors from the design values of soil properties obtained.
The fourth objective addresses the improvement in computational efficiency of the reliability-based design and optimization methodology using advanced techniques such as adaptive kriging methods. The adaptive kriging algorithm is modified to bring in the variability of foundation dimensions to obtain an optimized design of the foundation from the different possible values of foundation dimensions. It improves the applicability of surrogate models by reducing the number of numerical simulations required to develop the surrogate model. Quantile-based adaptive kriging methodology is used, which considers the variation of foundation dimensions and adds new simulation points at the area of interest, i.e., at the threshold value of the performance function. This methodology provides the possibility of arriving at economic design using quantile values. The effectiveness of the adaptive kriging method for the design and optimization of shallow foundations and the robustness of the kriging model are illustrated. The optimized design curve is obtained using adaptive kriging from which the optimum design can be selected that satisfies the design requirements.
The fifth and final objective deals with the quantile-based design methodology which provides the design properties based on quantile estimates using simple statistical calculations. The quantile-based design methodology proves to be computationally efficient, reducing the number of computations required to obtain a probabilistic design. The study also shows the possibility of arriving at optimised design using simple calculations.
Thus, this thesis proposes a design methodology that can be used for probabilistic design and optimization of foundations using analytical or numerical models, along with a detailed study on different types of variability that affect the foundation performance. The major issue of computational cost encountered while using probabilistic methods in conjunction with numerical models is sorted out by using kriging surrogates or adaptive kriging methodology. The quantile-based design methods can be used for probabilistic design using simple statistical calculations. The probabilistic framework developed in this thesis is applied to the design of shallow and pile foundations, to illustrate their applicability and efficiency in the design process
Information Rates over Point-to-Point and Multi-user Wireless Channels with Energy and Delay Constraints
No full textIn this thesis, we consider communication systems having energy, delay and reliability constraints.
We characterize optimal communication rates achievable over these systems.
First we consider point-to-point communication setting. In this context, first, we characterize
achievable rates for an energy harvesting point-to-point channel with additive Gaussian
noise. These rates are shown to be close to the optimal rates under various assumptions on the
system architecture. Next, we consider a non-energy harvesting, point-to-point block fading
wireless channel with Gaussian noise, subjected to canonical peak and average power constraints.
We characterize lower and upper bounds on the maximal channel coding rate with
channel state information at the transmitter and the receiver, at a given codeword length and
average probability of error. The bounds characterize back-off from the water-filling capacity in
the finite block length regime. Subsequently, we extend the finite block length results derived
for the non-energy harvesting channel, to the case where the transmitter is energy harvesting.
Next, we consider multi-user communication scenario. In this setting, first, we consider a
Gaussian multiple access channel with energy harvesting transmitters. We obtain the capacity
region of the channel with transmitters having infinite energy buffer. In addition, with transmitters
assumed to have finite energy buffers, we characterize achievable rate regions.
Next, we obtain inner bounds to the ergodic capacity region of a block fading Gaussian multiple
access channel, with finite codeword length, non-vanishing average error probability and peak
power constraint on the codewords. Subsequently, we consider a fading Gaussian broadcast
channel under the assumption that both the transmitter and the receivers harvests energy and
the receivers treat the transmitter as a radio frequency energy source. This corresponds to
the paradigm of simultaneous wireless information and power transfer. In this scenario, we
characterize the fundamental limits of simultaneous wireless information and power transfer
under a minimum-rate constraint. Finally, we consider an energy harvesting, fading Gaussian
multiple access channel and the receivers treat transmitter as an radio-frequency energy source.
In this setting as well, we characterize the information theoretic limits under a minimum-rate
constraint at each user
Interpolation of Digital Elevation Models using Generative Adversarial Networks
No full textA digital elevation model (DEM) is a three-dimensional representation of elevation data of a terrain such as a terrestrial terrain acquired by a reconnaissance aircraft or a lunar terrain acquired using a Chandrayaan rover. Terrestrial DEMs are used in hydrological modeling, geomorphology, and glaciology. Lunar DEMs can be used to locate natural resources and to identify prospective landing sites for exploratory missions. Hence, high quality, reliable DEMs are of great significance. DEMs are generally captured using LiDAR (Light Detection and Ranging), stereophotogrammetry, and time-of-flight cameras. However, a reliable DEM cannot be constructed if there are no adequate landmark points/features. This is the case with smooth terrains, occlusions, presence of multiple voids. The measurements are nonuniform in general. Hence, there is a need for interpolation and void-filling techniques that can estimate the elevation with a high accuracy.
Inverse distance weighting (IDW), De Launay triangulation, and Kriging are some of the popular benchmark algorithms for interpolating scattered and nonuniformly spaced data. Manual parameter tuning, inability to recover high-frequency information, and high computational complexity are some of the issues that these traditional interpolation techniques suffer from. Deep Learning (DL) has proven to be effective in providing excellent results in the field of image processing and computer vision, specifically in the tasks of super resolution, image in-painting, extrapolation, and segmentation. With the massive success of DL in several image processing and computer vision applications, its applicability has been explored for solving the DEM interpolation problem as well. However, convolutions are not readily defined if the measurements are nonuniform. Hence, the recent DL based research on DEM interpolation has only focused only on regularly spaced data.
We address the realistic problem of DEM interpolation from irregularly spaced measurements, with the density of measurements varying spatially. This is a new and unexplored direction in the deep learning setting. We propose a new and a robust DL architecture based on Generative Adversarial Networks (GANs) to perform interpolation and result in a uniform DEM with a user-specified resolution. The generator comprises three modules: Learnable Distance Weighting Module (LDW), DEM in-painting architecture, and Continuous Convolution (CC) modules. We design the novel LDW module as a learnable counterpart to the popular IDW algorithm that operates on the distances between the measurements and grid locations. This reduces the problem to that of inpainting post the LDW transformation. The proposed method is evaluated on synthetically generated data and on standard publicly available NASA (LOLA LRO) datasets using the mean relative error and PSNR as performance metrics. Extensive experiments justify the effectiveness and accuracy of the proposed approach in comparison with the benchmark techniques. We conclude the thesis by discussing possible future directions for DL based DEM interpolation.ISRO - IISc Space Technology Cel
Turbulent Free Moist Convection over Horizontal Surfaces
No full textThe heat and moisture loss from water surfaces are important in many natural and industrial systems. For example, the natural water cycle where water evaporates from the lakes, rivers, and seas, rises in the atmosphere, forms a cloud, and returns to the earth's surface in the form of rain. In addition, evaporation is a significant source of atmospheric water. In the above cases, heat and mass transport are coupled with each other, making the problem more complex. Even though few studies have been conducted to understand the flow dynamics and scalar transport near the surface, our understanding of this problem is still insufficient.
The current thesis deals with turbulent free moist convection over the horizontal surface. The study is carried out in two separate parts. The first part proposes a model to describe the near-wall dynamics in turbulent moist convection over the horizontal surface. In the second part, experiments are conducted to verify the model and study heat transfer and evaporation rates. The experimental set-up consists of an open tank of heated water kept in still ambient air. Diagnostics include temperature and humidity measurements near the water surface and visualization of the flow field, which typically consists of small water droplets from condensed vapour.
Our experiments of turbulent moist convection over the heated water surface indicate that the mechanics of heat and mass transfer near the surface is line plumes. Based on this evidence, our model consists of a 2-D periodic array of line plumes and associated boundary layers. The model provides the value for the average plume spacing. The predicted value of the plume spacing is in reasonable agreement with the experiments. Moreover, it predicts the distribution of mean temperature, vapour density and supersaturation, and rms of fluctuation of the above quantities. To understand the effects of the condensation, we also solve the model using a numerical simulation where we include the condensation in the plume and the boundary layer. The distribution of various quantities mentioned above changed slightly, while the critical plume spacing remained constant. The simulation gives the liquid water concentration field and its mean and rms fluctuation.
The experiments are carried out over the heated water surface to understand the planform structure and the heat & mass transfer for different input conditions. Visualization done using a laser sheet oriented vertically or horizontally just above the water surface confirms the existence of the line plumes close to the surface. The plumes are oriented randomly and move in the lateral direction and merge due to an entrainment field created by the upward moving plume and an incoming flow in the downward direction. At a higher Grashof number (surface temperature), we observed the mushroom-type structures in the side-view and spiral structures in the top-view. The origin and elongation of the plumes are similar to that observed in single component natural convection.
Plume length and plume spacing are found manually using software from the top view images of the flow field. A continuous laser sheet and High-speed camera are used to visualize the particle motion, and Particle Image Velocimetry (PIV) is used to get the rms of horizontal and vertical velocity fluctuation. Here, the particles are the droplets formed due to condensation in the plume and the boundary layer. Temperature, vapour density, and relative humidity profiles are obtained with the help of thermocouples and humidity sensors. The distribution of mean temperature, vapour density, supersaturation, rms of temperature, and the plume spacing are in good agreement with the results from the model.
A siphon system is used to get the evaporation rate. A power-law correlation is developed between the Sherwood number and Rayleigh number (Sh = 0.373 Sc1/3 Raea0.291). It is difficult to get the exact values of sensible heat flux directly from the experiments, so we used the model that relates the sensible heat Nusselt number, NuT and Raea. The values obtained from these equations are in good agreement with the experiments in the literature. The heat and mass transfer are also represented in non-dimensional form (GrδT−1/3and GrδM−1/3), where Gr is Grashof number that includes density contributions due to temperature and water vapour content, and δT and δM are conduction layer thickness based on temperature and vapour density. The experimental values of both the parameters lie between 0.13 and 0.15. These values are nearly independent of the Grashof number
Isoform Specific Synthetic Glutathione Peroxidase Mimetics in Redox Regulation of Cell Death and Diseases
No full textReactive oxygen species (ROS)-induced oxidative stress has been implicated in various pathologies and cell death processes. Cells have antioxidant systems to neutralize the harmful effects of excess ROS. Glutathione peroxidases (GPXs) are the key antioxidant enzymes that catalyze the reduction of H2O2 or organic hydroperoxides to water or the corresponding alcohols using glutathione (GSH). In humans, eight GPX isoforms are found, of which GPX1, GPX2, GPX3, GPX4, and GPX6 are selenoproteins containing selenocysteine (Sec) in their active site. Each isoform has unique characteristics and specific tissue localization that determines the precise biological role. Insufficiency of the GPX antioxidant system has been implicated in various diseases, including neurodegenerative disorders, stroke, pulmonary embolism, ischemia/reperfusion injury (IRI), brain damage, cancer, organ failure, and cell death processes such as apoptosis and ferroptosis. My work entails the development of isoform-specific small molecule GPX mimetics and the validation of GPX mimetics in cell death and disease models where insufficiency of GPX activity is the underlying cause of cell death and disease progression.
Blood platelets, primary mediators of homeostasis and thrombosis, play a crucial role in angiogenesis, repair of the endothelium, and wound healing. Platelets can be targets of ROS produced within or by the vasculature, which dramatically affects platelet physiology. A decrease in the circulating platelet population leads to a condition referred to as thrombocytopenia, which can be life-threatening due to trouble stopping bleeding. Reduced intracellular antioxidant capacity in platelets contributes to thrombocytopenia via ROS mediated apoptotic signalling pathways. Our study demonstrates that compd.1, a selenium-based small molecule having GPX1/3-like activity protects platelets from apoptosis by regulating ROS-induced ASK1 (Apoptosis Signal-Regulating kinase 1) activation. GPX deficiency has also been associated with an increased risk of platelet-dependent thrombosis and vascular dysfunction which results in vessel occlusion leading to stroke, pulmonary embolism, and heart attack leading cause of death worldwide. We found that Compd.1, a GPX1/3 mimetic, inhibits platelet activation in response to various agonists such as collagen, thrombin, and ADP. We further show that Compd.1 effectively suppresses platelet activation-mediated platelet neutrophil aggregates and neutrophil extracellular trap formation, which are associated with inflammatory and thrombotic disorders such as sepsis, vascular inflammation, and COVID-19. Importantly, we show that Compd.1 potently inhibits platelet-dependent thrombosis in mouse models without causing bleeding complications.
Ferroptosis is an iron-dependent, non-apoptotic form of regulated cell death caused by aberrant accumulation of phospholipid hydroperoxides in cellular membranes. Ferroptosis has been implicated in several pathological conditions, including neurodegenerative disorders, ischemia/reperfusion injury (IRI), brain damage, cancer, and organ failure. The selenoenzyme glutathione peroxidase 4 (GPX4) is the master regulator of ferroptosis suppresses ferroptosis by reducing phospholipid hydroperoxides to their corresponding alcohol. We have synthesized a novel organoselenium compound (Compd.5) that effectively mimics the activity of GPX4 and suppresses ferroptosis in primary neurons. Compd.5 actively reduces lipid hydroperoxides and suppresses ferroptosis induced by classic ferroptosis inducers such as RSL3, FIN56, FINO2 Erastin, and HCA in primary cortical neurons.
Overall, our studies led to the development of isoform-specific synthetic GPX mimetics that can be harnessed to combat platelet activation-mediated thrombotic diseases, ROS-mediated thrombocytopenia in platelets, and neurological disorders associated with ferroptosis.Indian Institute of Science (IISc