2,014 research outputs found
DEEP LEARNING ARCHITECTURES - NEURAL NETWORKS AND DEEP LEARNING MODELS
No full text<p>DEEP LEARNING ARCHITECTURES - NEURAL NETWORKS AND DEEP LEARNING MODELS</p>
Body on the Barricades is a book of hope amidst curtailment of rights and freedom
No full textK. Kalyani interviews Brahma Prakash, author and academician, on his new book, Body on the Barricades: Life, Art and Resistance in Contemporary Indi
sj-docx-2-pie-10.1177_09544089221112069 - Supplemental material for Numerical analysis of non-isothermal warm deep drawing of an Al-Mg alloy using different yield criteria and experimental validation
No full textSupplemental material, sj-docx-2-pie-10.1177_09544089221112069 for Numerical analysis of non-isothermal warm deep drawing of an Al-Mg alloy using different yield criteria and experimental validation by Ved Prakash and D Ravi Kumar in Proceedings of the Institution of Mechanical Engineers, Part E: Journal of Process Mechanical Engineering</p
sj-docx-1-pie-10.1177_09544089221112069 - Supplemental material for Numerical analysis of non-isothermal warm deep drawing of an Al-Mg alloy using different yield criteria and experimental validation
No full textSupplemental material, sj-docx-1-pie-10.1177_09544089221112069 for Numerical analysis of non-isothermal warm deep drawing of an Al-Mg alloy using different yield criteria and experimental validation by Ved Prakash and D Ravi Kumar in Proceedings of the Institution of Mechanical Engineers, Part E: Journal of Process Mechanical Engineering</p
Buoyancy considerations in a fluidized bed of ground particles:
No full textThe research conducted for this thesis determines and quantifies the forces acting on a large submerged object in a gas-solid fluidized bed of Geldart A ground particles at the onset of bubbling. In particular, the additional load from the previously reported de-fluidized region of particles found above submerged objects in fluidized beds is quantified. A force model was developed and comparisons between this and the experimental data yielded properties of the de-fluidized region, such as the shape and height of the hood, and the angle to which it rises. Drag is measured by comparing forces on objects of different geometries subject to the same conditions. Buoyancy was measured by summing the pressure force acting to push the submerged object out of the bed and the counteracting pressure force of the bed over the submerged object pushing it deeper into the bed. The de-fluidized region is found to have a significant impact on the buoyancy of the object by creating an additional weight force above the object thus increasing the pressure in this region. Explanations for the hood shape and size are made through phenomenological results obtained through visual observations. Particle transport through voidage collapses at the surface, i.e. “bubbling”, create a secondary particle circulation within the bed adding particles to the top of the hood. The size of the bubbles appears to determine the extent of the hood.M.S.Includes bibliographical references (p. 78-79)by Prakash R. Ra
Beyond Modernity : Narrative Strategies in Hindi Short Stories of Uday Prakash
No full textThis thesis explores different genres and modes of writing in short stories of the contemporary Indian author Uday Prakash, such as magical realism, the fantastic, regionalism, postcolonial and postmodern writing. It poses the question: “In which literary genre should Uday Prakash’s writings best be categorised?” The study is based on a reading of Prakash’s collection, 10 Pratinidhi kahāniyāṃ – Ten Representative Stories, consisting of ten stories of the author’s own choice. Critics have often understood Uday Prakash as a writer of magical realism. This thesis, however, argues that the author fits better in the category of the fantastic since his narratives often are characterised by the “hesitation” before the supernatural, a central feature of this literary mode. The thesis further suggests that regionalism is present in his writings in the portrayal of both the rural and urban landscapes of India. Above all, Prakash portrays a “public landscape,” in which India as a whole is reflected in the local—rural or urban—regions he depicts and in which any Indian can identify himself. The postmodern perspective is also prevalent in his writings, evident through literary tropes such as metafiction, historiographic metafiction, intertextuality, self-reflexivity and extended use of metaphor. Central to his writing is a social or postcolonial critique. Together his stories write an alternative national history of India, focusing on the subaltern and the downtrodden, depicting how the old colonial structure and oppression have now re-emerged among the elite and political leadership of independent India. I have, in this thesis, understood Uday Prakash as a postcolonial experimentalist (uttaropaniveśvādī prayogvādī), standing in the tradition of the prayogvād of the 20th-century Hindi literary field since the characteristics of his authorship are the concoction of multiple literary modes or genres, the breaking with traditional forms of narration and the formation of creative and original narratives, all in the service of social and civilisational criticism
Characterization of the flow field and shape of de-fluidized regions in the wake of large submerged objects in small particle fluidized beds
No full textThe structure of the de-fluidized hood, an area of un-fluidized material found in the wake downstream of submerged objects, is studied with special interest on the internal fluidization structure and shape of the hood. Previously, the structure of the region was determined based on indirect photographic evidence rather than direct measurement. Also, the wake region had been considered to be entirely packed and focus had only been on the de-fluidized region in the near wake. In the present study, an experimental set-up was created which allowed for spheres and disks of various radii to be suspended within a fluidized bed of Geldart A particles. The force imparted by the wake region at various depths on a submerged object was measured providing insight into the downstream extent of the region, as well as the nature of the fluidization within the wake. Additionally, bed collapse with an object near the surface was studied to gain insight into particle mobility within the wake. Experiments were conducted to measure the maximum normal force supported by the wake region, as well as the shear and drag, at various depths above the submerged object surface providing an indirect measure of the velocity within the wake and its effect on neighboring structures. The results show the region to have several fluidization states: a dead region, a stagnant percolating region, a large percolating region of mobile particles, and an expanding region. The overall shape is found to resemble cylinder with a rounded top. Voids generated at the hemisphere of the submerged objects and particle circulation into and out of the region were found to play a key role in determining the shape and height of the hood, as well as playing a key role in the re-fluidization of the region. Using the data collected from the experiments, a model of the weight force imparted by the de-fluidized region on a submerged object was developed.Ph. D.Includes bibliographical referencesby Prakash R. Ra
On Improving Robustness of Deep Neural Networks for Computer Vision
No full textOver the past decade, deep learning has spearheaded a revolution in a myriad of applications, achieving remarkable success. This is particularly evident in the field of computer vision, which has undergone a transformative evolution, fueled by deep neural networks and the availability of extensive datasets. However, despite the monumental advancements in computer vision research, significant vulnerabilities persist. Numerous studies have illustrated that deep networks are fundamentally susceptible to alterations, including subtle perturbations and style changes, in input data that maintain consistent conceptual integrity in human perception. This includes phenomena such as common corruptions, adversarial perturbations, and domain shifts, all of which pose substantial challenges to the stability and reliability of these deep learning-based computer vision systems.
To address the identified challenges within computer vision, my dissertation embarked on a systematic exploration and resolution of the robustness dilemma inherent in both 2D and 3D realms through an array of approaches encompassing comprehensive benchmarking, architectural refinements, train-time strategies, test-time adaptation, and system-level methodologies. My dissertation is articulated over four segments, collectively supporting my research thesis. In the initial segment, benchmarking and enhancement of adversarial and general robustness in 2D image classification are attended to. We introduced FourierMix as a novel data augmentation technique and VPA as a test-time prompting method to ameliorate certified adversarial and general robustness, respectively. The ensuing segment revolves around benchmarking and improving the robustness of 3D point cloud recognition against adversarial attacks and common corruptions. We first proposed a novel dataset, ModelNet40-C, accompanied by systematic benchmarking results. Subsequently, we devised a new global pooling operation, DeepSym, and harnessed self-supervised learning to heighten the efficacy of adversarial training in 3D point cloud recognition. Furthermore, we discerned an inherent limitation of adversarial training in the 3D space and proposed PointDP, a diffusion-driven purifier, as a defense against adversarial point clouds. Transitioning to a more complex task, the third segment is dedicated to boosting the robustness of LiDAR-based 3D object detection. A universal vulnerability within LiDAR-based perception was identified, followed by the proposal of both system- and model-level countermeasures to mitigate the vulnerability. In the final part, we introduced CALICO, aiming at elevating 2D and 3D multimodal perception efficiency and robustness via contrastive pretraining strategies.PhDComputer Science & EngineeringUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/192386/1/jiachens_1.pd
- …
