arXiv subjects

Abhishek Sharma

Publications and source records attributed to Abhishek Sharma.

136 records · Page 8Linked to original sources

arXiv2016

VConv-DAE: Deep Volumetric Shape Learning Without Object Labels

With the advent of affordable depth sensors, 3D capture becomes more and more ubiquitous and already has made its way into commercial products. Yet, capturing the geometry or complete shapes of everyday objects using scanning devices (e.g. Kinect) still comes with several challenges that result in noise or even incomplete shapes. Recent success in deep learning has shown how to learn complex shape distributions in a data-driven way from large scale 3D CAD Model collections and to utilize them for 3D processing on volumetric representations and thereby circumventing problems of topology and tessellation. Prior work has shown encouraging results on problems ranging from shape completion to recognition. We provide an analysis of such approaches and discover that training as well as the resulting representation are strongly and unnecessarily tied to the notion of object labels. Thus, we propose a full convolutional volumetric auto encoder that learns volumetric representation from noisy data by estimating the voxel occupancy grids. The proposed method outperforms prior work on challenging tasks like denoising and shape completion. We also show that the obtained deep embedding gives competitive performance when used for classification and promising results for shape interpolation.

cs.CV↗

arXiv2016

Composites of FeCl3 and TiO2 with Bismaleimide resins

Ferric Chloride-Bismaleimide (FeCl3-BMI) and Titania-Bismaleimide (TiO2-BMI) composite were synthesized using PVA as a binder. The composite systems were deposited on glass slide as a homogenous coating. XRD and FTIR studies of the composite system showed its crystalline and structural configuration. A mixed phase of TiO2 and BMI as well as short range crystallinity was observed. An increase in crystallinity with temperature was also seen. The percentage of N-H symmetric stretching was also found to increase with temperature.

cond-mat.mtrl-sci↗

arXiv2016

Substrate effect and nanoindentation fracture toughness based on pile up and failure

The effect of substrate was studied using nanoindentation on thin films. Soft films on hard substrate showed more pile up than usual which was attributed to the dislocation pile up at the film substrate interface. The effect of tip blunting on the load depth and hardness plots of nanoindentation was shown. The experimental date of variation of Vickers hardness with film thickness and loads were fitted and new parameters were analyzed. The delaminated area was analyzed using geometrical shapes using optical view of the failure region along with the load displacement Indentation fracture using Nanoindentation using Berkovich indenter has been studied. Indentation fracture toughness (KR) was analyzed based on computational programs. The contact mechanics during nanoindentation was studied with parameters related to indenter shape and tip sharpness. Elastic, plastic and total energies were computationally determined. The energy difference was related to shear stress being generated with elastic to plastic transition. Change in the nature of residual stress was related to film thickness.

cond-mat.mtrl-sci↗

arXiv2016

Controlling Search in Very large Commonsense Knowledge Bases: A Machine Learning Approach

Very large commonsense knowledge bases (KBs) often have thousands to millions of axioms, of which relatively few are relevant for answering any given query. A large number of irrelevant axioms can easily overwhelm resolution-based theorem provers. Therefore, methods that help the reasoner identify useful inference paths form an essential part of large-scale reasoning systems. In this paper, we describe two ordering heuristics for optimization of reasoning in such systems. First, we discuss how decision trees can be used to select inference steps that are more likely to succeed. Second, we identify a small set of problem instance features that suffice to guide searches away from intractable regions of the search space. We show the efficacy of these techniques via experiments on thousands of queries from the Cyc KB. Results show that these methods lead to an order of magnitude reduction in inference time.

cs.AI↗

arXiv2016

Ultra-sensitive nanoscale magnetic field sensors based on resonant spin filtering

Solid state magnetic field sensors based on magneto-resistance modulation find direct applications in communication devices, specifically in proximity detection, rotational reference detection and current sensing. In this work, we propose sensor structures based on the magneto-resistance physics of resonant spin-filtering and present device designs catered toward exceptional magnetic field sensing capabilities. Using the non-equilibrium Green's function spin transport formalism self consistently coupled to the Poisson's equation, we present highly-tunable pentalayer magnetic tunnel junction structures that are capable of exhibiting an ultra-high peak tunnel magneto resistance $(\approx 2500 \%$). We show how this translates to device designs featuring an ultra-high current sensitivity enhancement of over 300\% in comparison with typical trilayer MTJ sensors, and a wider tunable range of field sensitivity. We also demonstrate that a dynamic variation in sensor functionalities with the structural landscape enables a superior design flexibility over typical trilayer sensors. An optimal design exhibiting close to a 700\% sensitivity increase as a result of angle dependent spin filtering is then presented.This work sets a stage to engineer spintronic building blocks via the design of functional structures tailored to exhibit ultra-sensitive spin filtering.

cond-mat.mes-hall↗

arXiv2015

Time Critical Multitasking for Multicore Microcontroller using XMOS Kit

This paper presents the research work on multicore microcontrollers using parallel, and time critical programming for the embedded systems. Due to the high complexity and limitations, it is very hard to work on the application development phase on such architectures. The experimental results mentioned in the paper are based on xCORE multicore microcontroller form XMOS. The paper also imitates multi-tasking and parallel programming for the same platform. The tasks assigned to multiple cores are executed simultaneously, which saves the time and energy. The relative study for multicore processor and multicore controller concludes that micro architecture based controller having multiple cores illustrates better performance in time critical multi-tasking environment. The research work mentioned here not only illustrates the functionality of multicore microcontroller, but also express the novel technique of programming, profiling and optimization on such platforms in real time environments.

cs.DC↗

arXiv2015

Deep Hierarchical Parsing for Semantic Segmentation

This paper proposes a learning-based approach to scene parsing inspired by the deep Recursive Context Propagation Network (RCPN). RCPN is a deep feed-forward neural network that utilizes the contextual information from the entire image, through bottom-up followed by top-down context propagation via random binary parse trees. This improves the feature representation of every super-pixel in the image for better classification into semantic categories. We analyze RCPN and propose two novel contributions to further improve the model. We first analyze the learning of RCPN parameters and discover the presence of bypass error paths in the computation graph of RCPN that can hinder contextual propagation. We propose to tackle this problem by including the classification loss of the internal nodes of the random parse trees in the original RCPN loss function. Secondly, we use an MRF on the parse tree nodes to model the hierarchical dependency present in the output. Both modifications provide performance boosts over the original RCPN and the new system achieves state-of-the-art performance on Stanford Background, SIFT-Flow and Daimler urban datasets.

cs.CV↗

arXiv2014

Locally Scale-Invariant Convolutional Neural Networks

Convolutional Neural Networks (ConvNets) have shown excellent results on many visual classification tasks. With the exception of ImageNet, these datasets are carefully crafted such that objects are well-aligned at similar scales. Naturally, the feature learning problem gets more challenging as the amount of variation in the data increases, as the models have to learn to be invariant to certain changes in appearance. Recent results on the ImageNet dataset show that given enough data, ConvNets can learn such invariances producing very discriminative features [1]. But could we do more: use less parameters, less data, learn more discriminative features, if certain invariances were built into the learning process? In this paper we present a simple model that allows ConvNets to learn features in a locally scale-invariant manner without increasing the number of model parameters. We show on a modified MNIST dataset that when faced with scale variation, building in scale-invariance allows ConvNets to learn more discriminative features with reduced chances of over-fitting.

cs.CV↗

arXiv2010

Innovative Rated-Resource Peer-to-Peer Network

Peer-to-Peer (P2P) networks provide a significant solution for file sharing among peers connected to Internet. It is fast and completely decentralised system with robustness. But due to absence of a server documents on a P2P network are not rated which makes it difficult for a peer to obtain precise information in result of a query. In past, some researchers tried to attach ratings to the peers itself but it was complex and less effective. In this paper, a novel P2P architecture is proposed which attaches ratings to the uploaded document directly. These ratings then become as element in its XML advertisement which has several child elements for information classification. The attached element is extracted from the advertisement in real time and the document is then sorted accordingly. Therefore, the information can be easily sorted based on a request by a peer according to the relevance of matter. The information regarding relevance is obtained by the peer issuing the query. This research leads to a smart P2P model, the Rated-Resource P2P network (R2P2P).

cs.NI↗

arXiv2008

Dynamic Data Compression with Distortion Constraints for Wireless Transmission over a Fading Channel

We consider a wireless node that randomly receives data from different sensor units. The arriving data must be compressed, stored, and transmitted over a wireless link, where both the compression and transmission operations consume power. Specifically, the controller must choose from one of multiple compression options every timeslot. Each option requires a different amount of power and has different compression ratio properties. Further, the wireless link has potentially time-varying channels, and transmission rates depend on current channel states and transmission power allocations. We design a dynamic algorithm for joint compression and transmission, and prove that it comes arbitrarily close to minimizing average power expenditure, with an explicit tradeoff in average delay. Our approach uses stochastic network optimization together with a concept of place holder bits to provide efficient energy-delay performance. The algorithm is simple to implement and does not require knowledge of probability distributions for packet arrivals or channel states. Extensions that treat distortion constraints are also considered.

math.OC↗