• Refine Query
  • Source
  • Publication year
  • to
  • Language
  • 638
  • 218
  • 133
  • 77
  • 42
  • 39
  • 20
  • 13
  • 10
  • 9
  • 8
  • 8
  • 7
  • 5
  • 5
  • Tagged with
  • 1454
  • 174
  • 165
  • 163
  • 137
  • 114
  • 100
  • 95
  • 93
  • 87
  • 70
  • 70
  • 67
  • 66
  • 66
  • About
  • The Global ETD Search service is a free service for researchers to find electronic theses and dissertations. This service is provided by the Networked Digital Library of Theses and Dissertations.
    Our metadata is collected from universities around the world. If you manage a university/consortium/country archive and want to be added, details can be found on the NDLTD website.
341

Breaking Privacy in Model-Heterogeneous Federated Learning

Haldankar, Atharva Amit 14 May 2024 (has links)
Federated learning (FL) is a communication protocol that allows multiple distrustful clients to collaboratively train a machine learning model. In FL, data never leaves client devices; instead, clients only share locally computed gradients or model parameters with a central server. As individual gradients may leak information about a given client's dataset, secure aggregation was proposed. With secure aggregation, the server only receives the aggregate gradient update from the set of all sampled clients without being able to access any individual gradient. One challenge in FL is the systems-level heterogeneity that is quite often present among client devices. Specifically, clients in the FL protocol may have varying levels of compute power, on-device memory, and communication bandwidth. These limitations are addressed by model-heterogeneous FL schemes, where clients are able to train on subsets of the global model. Despite the benefits of model-heterogeneous schemes in addressing systems-level challenges, the implications of these schemes on client privacy have not been thoroughly investigated. In this thesis, we investigate whether the nature of model distribution and the computational heterogeneity among client devices in model-heterogeneous FL schemes may result in the server being able to recover sensitive information from target clients. To this end, we propose two novel attacks in the model-heterogeneous setting, even with secure aggregation in place. We call these attacks the Convergence Rate Attack and the Rolling Model Attack. The Convergence Rate Attack targets schemes where clients train on the same subset of the global model, while the Rolling Model Attack targets schemes where model-parameters are dynamically updated each round. We show that a malicious adversary is able to compromise the model and data confidentiality of a target group of clients. We evaluate our attacks on the MNIST dataset and show that using our techniques, an adversary can reconstruct data samples with high fidelity. / Master of Science / Federated learning (FL) is a communication protocol that allows multiple distrustful users to collaboratively train a machine learning model together. In FL, data never leaves user devices; instead, users only share locally computed gradients or model parameters (e.g. weight and bias values) with an aggregation server. As individual gradients may leak information about a given user's dataset, secure aggregation was proposed. Secure aggregation is a protocol that users and the server run together, where the server only receives the aggregate gradient update from the set of all sampled users instead of each individual user update. In FL, users often have varying levels of compute power, on-device memory, and communication bandwidth. These differences between users are collectively referred to as systems-level (or system) heterogeneity. While there are a number of techniques to address system heterogeneity, one popular approach is to have users train on different subsets of the global model. This approach is known as model-heterogeneous FL. Despite the benefits of model-heterogeneous FL schemes in addressing systems-level challenges, the implications of these schemes on user privacy have not been thoroughly investigated. In this thesis, we investigate whether the nature of model distribution and the differences in compute power between user devices in model-heterogeneous FL schemes may result in the server being able to recover sensitive information. To this end, we propose two novel attacks in the model-heterogeneous setting with secure aggregation in place. We call these attacks the Convergence Rate Attack and the Rolling Model Attack. The Convergence Rate Attack targets schemes where users train on the same subset of the global model, while the Rolling Model Attack targets schemes where model-parameters may change each round. We first show that a malicious server is able to obtain individual user updates, despite secure aggregation being in place. Then, we demonstrate how an adversary can utilize those updates to reverse engineer data samples from users. We evaluate our attacks on the MNIST dataset, a commonly used dataset of handwritten digits and their labels. We show that by running our attacks, an adversary can accurately identify what images a user trained on.
342

Relational Learning approaches for Recommender Systems

Pellegrini, Giovanni 07 October 2021 (has links)
Learning on relational data is a relevant task in the machine learning community. Extracting information from structured data is a non-trivial task due to the combinatorial complexity of the domain and the necessity to construct methods that work on collections of values of different sizes rather than fixed representations. Relational data can naturally be interpreted as graphs, a class of flexible and expressive structures that can model data from diverse domains,from biology to social interactions. Graphs have been used in a huge variety of contexts, such as molecular modelling, social networks, image processing and recommendation systems. In this manuscript, we tackle some challenges in learning on relational data by developing new learning methodologies. Specifically, in our first contribution, we introduce a new class of metrics for relational data based on relational features extraction technique called Type ExtensionTrees. This class of metrics defines the (dis)similarity of two nodes in a graph by exploiting the nested structure of their relational neighborhood at different depth steps. In our second contribution, we developed a new strategy to collect the information of multisets of data values by introducing a new framework of learnable aggregators called Learning Aggregation Functions.We provide a detailed description of the methodologies and an extensive experimental evaluation on synthetic and real world data to assess the expressiveness of the proposed models. A particular focus is given to the application of these methods to the recommendation systems domain, exploring the combination of the proposed methods with recent techniques developed for Constructive Preference Elicitation and Group Recommendation tasks.
343

A Parallel Aggregation Algorithm for Inter-Grid Transfer Operators in Algebraic Multigrid

Garcia Hilares, Nilton Alan 13 September 2019 (has links)
As finite element discretizations ever grow in size to address real-world problems, there is an increasing need for fast algorithms. Nowadays there are many GPU/CPU parallel approaches to solve such problems. Multigrid methods can be used to solve large-scale problems, or even better they can be used to precondition the conjugate gradient method, yielding better results in general. Capabilities of multigrid algorithms rely on the effectiveness of the inter-grid transfer operators. In this thesis we focus on the aggregation approach, discussing how different aggregation strategies affect the convergence rate. Based on these discussions, we propose an alternative parallel aggregation algorithm to improve convergence. We also provide numerous experimental results that compare different aggregation approaches, multigrid methods, and conjugate gradient iteration counts, showing that our proposed algorithm performs better in serial and parallel. / Modeling real-world problems incurs a high computational cost because these mathematical models involve large-scale data manipulation. Thus we need fast and efficient algorithms. Nowadays there are many high-performance approaches for these problems. One such method is called the Multigrid algorithm. This approach models a physical domain using a hierarchy of grids, and so the effectiveness of these approaches relies on how well data can be transferred from grid to grid. In this thesis, we focus on the aggregation approach, which clusters a grid’s vertices according to its connections. We also provide an alternative parallel aggregation algorithm to give a faster solution. We show numerous experimental results that compare different aggregation approaches and multigrid methods, showing that our proposed algorithm performs better in serial and parallel than other popular implementations.
344

Aggregation, courtship, and behavioral interactions in European earwigs, Forficula auricularia L. (Dermaptera: Forficulidae)

Walker, Karen Ann 02 October 2007 (has links)
Due to its relatively cool, humid summers, southwestern Virginia provides an ideal climate for European earwigs, Forficula auricularia. In 1990 - 1992, nymphs were captured in wooden groove-board traps beginning in late May, adults were captured beginning in mid-June, and disappeared from sampling sites by September or October. Sex ratios were significantly female-biased most of the season, becoming more marked by the fall. The pest status of F. auncularia is exacerbated by its gregarious nature. Gas chromatography-mass spectroscopy and accompanying behavioral bioassays showed that aggregation occurred as a result of a pheromone located on the male cuticle, which is probably a minor component of the hydrocarbon profile. Approximately 88% of the detected volatiles on the cuticle were identified as a series of normal and branched alkanes. Fatty acids and hydrocarbons were also identified in nymphal and adult legs, but these extracts were not attractive. Frass, which also contained fatty acids and hydrocarbons, was attractive, but likely acquired its attractancy through the earwigs' proclivity for consuming carcasses and exuviae. The defensive quinones produced by F aunculana repel conspecifics. A study of the behavioral repertoire of F. aunculana showed that, contrary to previous reports, only nymphs are nocturnal. Many differences in behavior were due to gender, age, and partner age. (e.g., females spent more time feeding than did males, adults fed more when paired with nymphs than when paired with adults). Social behaviors (communal feeding, aggression, contact, and dorsal palpation) comprised <10% of the insect's behavioral repertoire. Since dorsal palpation, a previously undescribed behavior and a form of allogrooming, occurred more frequently during reproductive periods, it may have a sexual significance. Dorsal palpation also may augment the distribution of defensive quinones on the cuticle of F. auricularia. An analysis of nymphal group dynamics demonstrated that as group size increased, nymphs spent significantly less time feeding alone and grooming, but more time resting. Antennal contact rates between group members increased significantly with group size. Detailed observations of the courtship and mating of F. auricularia revealed a complex of sexual behaviors for both males and females. Receptive females were behaviorally active during courtship. The significance of the male cerci was demonstrated by their use in early courtship with displays, and later use as a tactile stimulus for the female; and study of males from which the cerci had been removed, which showed no mating by amputated males. Male forcep length was bimodally distributed and positively allometric, while female forcep length was normally distributed. Males with longer forceps did not have a mating advantage. Further research is needed to identify the chemical composition of the aggregation pheromone, and to quantify any advantages of body and forcep size on mating success. / Ph. D.
345

Role of the N- and C-terminal strands of beta 2-microglobulin in amyloid formation at neutral pH.

Jones, Susan, Smith, D.P., Radford, S.E. January 2003 (has links)
No / Beta 2-microglobulin (ß2m) is known to form amyloid fibrils de novo in vitro under acidic conditions (below pH 4.8). Fibril formation at neutral pH, however, has only been observed by deletion of the N-terminal six residues; by the addition of pre-assembled seeds; or in the presence of Cu2+. Based on these observations, and other structural data, models for fibril formation of ß2m have been proposed that involve the fraying of the N and C-terminal ß-strands and the consequent loss of edge strand protective features. Here, we examine the role of the N and C-terminal strands in the initiation of fibrillogenesis of ß2m by creating point mutations in strands A and G and comparing the properties of the resulting proteins with variants containing similar mutations elsewhere in the protein. We show that truncation of buried hydrophobic side-chains in strands A and G promotes rapid fibril formation at neutral pH, even in unseeded reactions, and increases the rate of fibril formation under acidic conditions. By contrast, similar mutations created in the remaining seven ß-strands of the native protein have little effect on the rate or pH dependence of fibril formation. The data are consistent with the view that perturbation of the N and C-terminal edge strands is an important feature in the generation of assembly-competent states of ß2m.
346

Characterisation of aggregates of cyclodextrin-drug complexes using Taylor Dispersion Analysis

Zaman, Hadar, Bright, A.G., Adams, Kevin, Goodall, D.M., Forbes, Robert T. 06 February 2017 (has links)
Yes / There is a need to understand the nature of aggregation of cyclodextrins (CDs) with guest molecules in increasingly complex formulation systems. To this end an innovative application of Taylor dispersion analysis (TDA) and comparison with dynamic light scattering (DLS) have been carried out to probe the nature of ICT01-2588 (ICT-2588), a novel tumor-targeted vascular disrupting agent, in solvents including a potential buffered formulation containing 10% hydroxypropyl-β-cyclodextrin. The two hydrodynamic sizing techniques give measurement responses are that fundamentally different for aggregated solutions containing the target molecule, and the benefits of using TDA in conjunction with DLS are that systems are characterised through measurement of both mass- and z-average hydrodynamic radii. Whereas DLS measurements primarily resolve the large aggregates of ICT01-2588 in its formulation medium, methodology for TDA is described to determine the size and notably to quantify the proportion of monomers in the presence of large aggregates, and at the same time measure the formulation viscosity. Interestingly TDA and DLS have also distinguished between aggregate profiles formed using HP-β-CD samples from different suppliers. The approach is expected to be widely applicable to this important class of drug formulations where drug solubility is enhanced by cyclodextrin and other excipients.
347

An Evaluation of Bed Bug (Cimex lectularius L.) Host Location and Aggregation Behavior

Reis, Matthew Douglas 11 January 2011 (has links)
This study attempts to elucidate bed bug behavior in response to host cues and aggregation cues from conspecifics. Both fed and unfed bed bugs were evaluated to determine differences in behavior with regard to their circadian activities. Arena bioassays were conducted to determine the bed bug's ability to locate a host from different distances and if the antennae were essential for host location. Starved bed bugs were able to locate a host from 100 cm away. The bed bugs search path became more directed towards the host as the bed bug was placed at closer distances. The bed bugs' mean searching speed was found to be 1.7 cm/s. The bed bugs were able to locate a host even when their antennae were completely removed. Fed and unfed bed bugs were tested both individually and in groups to determine their attraction towards aggregation cues. Both fed and unfed bed bugs, regardless of sex, were significantly attracted to feces of conspecifics and exuvia of fifth instars. Finally, bed bugs were observed throughout the night to document their circadian activities after successfully taking a blood meal or failing to take a blood meal. Unfed bed bugs continued to search for a host throughout the night until aggregating together 2 hours prior to photophase. Alternatively, fed bed bugs immediately aggregated together within 30 minutes of a bloodmeal. / Master of Science in Life Sciences
348

Stabilization of weakly charged microparticles using highly charged nanoparticles

Herman, David Joel 22 August 2011 (has links)
An experimental investigation was conducted to evaluate the possible use of highly-charged spherical nanoparticles to stabilize an aqueous dispersion of weakly-charged microspheres. At low pH values, the surface of silica is weakly charged, which leads to flocculation of colloidal suspensions of silica microspheres. Binary solutions of weakly charged silica microspheres and highly charged polystyrene latex nanoparticles result in adsorption of the nanoparticles onto the surface of the silica microspheres. This effectively "recharges" the silica spheres, with effective zeta potentials increased to the range that is unfavorable for flocculation of microspheres in a silica-only solution. However, this does not guarantee stability, and comparisons between positively charged amidine latex nanoparticles and negatively charged sulfate latex nanoparticles indicate that the degree of coverage plays an important role in the restabilization. The sulfate latex nanoparticles do not cover the surface sufficiently, and though they seemingly provide sufficient charge, the weakly charged patches of the exposed silica substrate can lead to flocculation. The amidine latex nanoparticles, on the other hand, cover the surface more completely, and effectively prevent flocculation of the silica microspheres. The mechanisms responsible for this different adsorption and stabilizing behavior are not entirely understood, as both the amidine and sulfate latex nanoparticles are of similar size and the magnitude of the zeta potentials of the different particle types are comparable. / Master of Science
349

Aggregated sensor payload submission model for token-based access control in the Web of Things

Amir, Mohammad, Pillai, Prashant, Hu, Yim Fun 26 October 2015 (has links)
Yes / Web of Things (WoT) can be considered as a merger of newly emerging paradigms of Internet of Things (IoT) and cloud computing. Rapidly varying, highly volatile and heterogeneous data traffic is a characteristic of the WoT. Hence, the capture, processing, storage and exchange of huge volumes of data is a key requirement in this environment. The crucial resources in the WoT are the sensing devices and the sensing data. Consequently, access control mechanisms employed in this highly dynamic and demanding environment need to be enhanced so as to reduce the end-to-end latency for capturing and exchanging data pertaining to these underlying resources. While there are many previous studies comparing the advantages and disadvantages of access control mechanisms at the algorithm level, vary few of these provide any detailed comparison the performance of these access control mechanisms when used for different data handling procedures in the context of data capture, processing and storage. This study builds on previous work on token-based access control mechanisms and presents a comparison of two different approaches used for handling sensing devices and data in the WoT. It is shown that the aggregated data submission approach is around 700% more efficient than the serial payload submission procedure in reducing the round-trip response time.
350

Aggregation and pattern formation in charged granular gases

Singh, Chamkor 02 September 2019 (has links)
No description available.

Page generated in 0.0809 seconds