187 research outputs found
Recommended from our members
Towards Resource-Efficient Trustworthy Distributed Learning
No full textFederated learning (FL) is a collaborative training framework, where multiple edge-devices (users) jointly train a global model, without sharing their private data. Instead, users share the local gradients with a central server after training with their on-device dataset. Several major challenges limit the deployment of conventional FL protocols in real-world large-scale networks. For example, mobile users may drop out from the protocol due to having limited communication and computation resources. Moreover, a single global model may not be optimal for users with heterogeneous data distributions. Lastly, user data can be retrieved even from the shared local gradients through gradient inversion attacks.
Clustered FL is a recent distributed learning mechanism to tackle data heterogeneity in FL, which allows training of multiple global models concurrently, each designated for a cluster of users, i.e., users with similar data distributions. To reduce communication latency in FL, a notable approach is over-the-air aggregation, where users share the same spectrum for sending gradients, enabling gradient aggregation directly over the wireless medium. Gradient sparsification is another widely adopted technique to alleviate communication cost, where each user sends only a small fraction of the entire gradient. For resource-efficient training of large models, a popular approach is submodel learning, which allows users to train a portion of the global model based on their available computing resources and bandwidth. With the prevalence of large foundation models in practical applications, parameter-efficient fine-tuning has recently gained attention, where users only train a few lightweight modules while keeping the pretrained model frozen, significantly reducing resource consumption.
The dissertation addresses several key challenges associated with trustworthy deployment of FL in resource-limited settings. Firstly, we propose an over-the-air aggregation mechanism for clustered FL, which uses the wireless medium to align the local gradients from the same cluster during aggregation, enabling the server to later correctly decode the gradient aggregate for each cluster. The underlying motivation is to tackle data heterogeneity and communication bottleneck simultaneously in large-scale FL settings. Next, we discover that user privacy can be breached when conventional secure aggregation (SA) protocols are naively applied to clustered FL or resource-aware FL protocols. This is attributed to the server's access to auxiliary information like cluster identity (in clustered FL), selected gradient coordinates (in gradient sparsification) or selected submodel indices (in submodel learning) of the user. To prevent associated privacy threats, we propose novel SA protocols, which ensure the information-theoretic privacy of both the auxiliary information and local gradients. Finally, we investigate the privacy risks associated with parameter-efficient fine-tuning, which has been largely underexplored. Accordingly, we identify a novel gradient inversion attack arising from a maliciously crafted pretrained model and fine-tuning modules, which demonstrates high-fidelity data recovery from the shared fine-tuning gradients
Lossy coding of correlated sources over a multiple access channel: necessary conditions and separation results
No full textLossy coding of correlated sources over a multiple access channel (MAC) is studied. First, a joint source-channel coding scheme is presented when the decoder has correlated side information. Next, the optimality of separate source and channel coding, that emerges from the availability of a common observation at the encoders, or side information at the encoders and the decoder, is investigated. It is shown that separation is optimal when the encoders have access to a common observation whose lossless recovery is required at the decoder, and the two sources are independent conditioned on this common observation. Optimality of separation is also proved when the encoder and the decoder have access to shared side information conditioned on which the two sources are independent. These separation results obtained in the presence of side information are then utilized to provide a set of necessary conditions for the transmission of correlated sources over a MAC without side information. Finally, by specializing the obtained necessary conditions to the transmission of binary and Gaussian sources over a MAC, it is shown that they can potentially be tighter than the existing results in the literature, providing a novel converse for this fundamental problem
On the necessary conditions for transmitting correlated sources over a multiple access channel
No full textWe study the lossy communication of correlated
sources over a multiple access channel (MAC). In particular, we
provide a new set of necessary conditions for the achievability of
a distortion pair over a given channel. The necessary conditions
are then specialized to the case of bivariate Gaussian sources and
doubly symmetric binary sources over a Gaussian multiple access
channel. Our results indicate that the new necessary conditions
provide the tightest conditions to date in certain cases
A multiphase phase-field study of three-dimensional martensitic twinned microstructures at large strains
No full textA thermodynamically consistent multiphase phase-field approach for stress and temperature-induced martensitic phase transformation at the nanoscale and under large strains is developed. A total of N independent order parameters are considered for materials with N variants, where one of the order parameters describes A M transformations and the remaining N-1 independent order parameters describe the transformations between the variants. A non-contradictory gradient energy is used within the free energy of the system to account for the energies of the interfaces. In addition, a non-contradictory kinetic relationships for the rate of the order parameters versus thermodynamic driving forces is suggested. As a result, a system of consistent coupled Ginzburg-Landau equations for the order parameters are derived. The crystallographic solution for twins within twins is presented for the cubic to tetragonal transformations. A 3D complex twins within twins microstructure is simulated using the developed phase-field approach and a large-strain-based nonlinear finite element method. A comparative study between the crystallographic solution and the simulation result is presented.This is a pre-print of the article Basak, Anup, and Valery I. Levitas. "A multiphase phase-field study of three-dimensional martensitic twinned microstructures at large strains." arXiv preprint arXiv:2206.12576 (2022).
DOI: 10.48550/arXiv.2206.12576.
Copyright 2022 The Author(s).
Attribution 4.0 International (CC BY 4.0).
Posted with permission
Failure mechanisms in lithium silicon batteries
No full textLithium silicon (Li-Si) batteries offer more than ten times the theoretical specific capacity compared to current lithium ion battery technologies, by using a silicon anode. In practice however, the cycle life of Li-Si batteries is very limited. The large volume change of the silicon anode is known to be the main reason for this. Research on the volume changes during varying cell cycles and voltages is presented in this thesis and an experimental set up for a quasi in situ study of the SEI layer is suggested. Cycling tests with an amorphous silicon thin film of 220 nm deposited using magnetron sputtering on a copper foil current collector confirmed that the major cause of capacity loss is swelling of the silicon during lithiation, causing the silicon to detach from the current collector and resulting in significant capacity loss. Increasing the lower cut off voltage from 0 V to 0.2 V resulted in a slight improvement of cycle life. Silicon detachment also decreased as determined by SEM images. EFTEM and EDX mapping showed a clear split between a partially lithiated silicon layer on the surface and a pure silicon layer on the current collector side. It can be concluded that discharging Li-Si batteries to 0.2 V instead of 0 V is a promising method to reduce the swelling of silicon during lithiation.HREMQuantum NanoscienceApplied Science
- …
