Dmitry Mikushin

Dmitry Mikushin

Auto-generated profile, based on publicly available information

Dmitry Mikushin

University of Lausanne, Lausanne, Switzerland

Research Focus & Metrics

Academic Career Length

17Years (2010-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
26 November 2023
Verified
Conference Paper
Open access
MBMariya Georgieva Belorgey
Mariya Georgieva Belorgey
SDSofia Dandjee
Sofia Dandjee
NGNicolas Gama
Nicolas Gama
DJDimitar Jetchev
Dimitar Jetchev
DMDmitry Mikushin
Dmitry Mikushin
We propose a novel protocol, Falkor, for secure aggregation for Federated Learning in the multi-server scenario based on masking of local models via a stream cipher based on AES in counter mode and accelerated by GPUs running on the aggregating servers. The protocol is resilient to client dropout and has reduced clients/servers communication cost by a factor equal to the number of aggregating servers (compared to the naïve baseline method). It scales simultaneously in the two major complexity aspects: 1) large number of clients; 2) highly complex machine learning models such as CNNs, RNNs, Transformers, etc. The AES-CTR-based masking function in our aggregation protocol is built on the concept of counterbased cryptographically-secure pseudorandom number generators (csPRNGs) as described in [32] and subsequently used by Facebook for their torchcsprng csPRNG. We improve upon torchcsprng by careful use of shared memory on the GPU device, a recent idea of Cihangir Tezcan [38] and obtain 100x speedup in the masking function compared to a single CPU core. Finally, we demonstrate scalability of our protocol in two realworld Federated Learning scenarios: 1) efficient training of large logistic regression models with 50 features and 50M data points distributed across 1000 clients that can dropout and securely aggregated via three servers (running secure multi-party computation (SMPC)); 2) training a recurrent neural network (RNN) model for sentiment analysis of Twitter feeds coming from a large number of Twitter users (more than 250,000 users). In case 1), our secure aggregation algorithm runs in less than a minute compared to a pure MPC computation (on 3 parties) that takes 27 hours and uses 400GB RAM machines as well as 1 gigabit-per-second network. In case 2), the total training is around 10 minutes using our GPU powered secure aggregation versus 10 hours using a single CPU core.

22 July 2019 in arXiv
Verified
Preprint
Open access
PDF file
NJNianchuan Jian
Nianchuan Jian
DMDmitry Mikushin
Dmitry Mikushin
JYJianguo Yan
Jianguo Yan
This paper introduces a phase tracking method for planetary radio science research with computational algorithm implemented fo r NVIDIA GPUs. In contrast to the phase-locked loop (PPL) phase counting method used in traditional Doppler data processing, this method fits the tracking data signal into the shape expressed by the Taylor polynomial with optimal phase and amplitude coefficients. The Differential Evolution (DE) algorithm is employed for polynomial fitting. In order to cope with high computational intensity of the proposed phase tracking method, the graphics processing units (GPUs) are employed. As a result, the method estimates the instantaneous phase, frequency, derivative of frequency (line-of-sight acceleration) and the total count phase of different integration scales. This data can be further used in planetary radio science research to analyze the planetary occultation and gravitational fields. The method has been tested on MEX (Mars Express, ESA) and Chang'E 4 relay satellite (China) tracking data. In a real experiment with 400K data block size and $\sim$80,000 DE solver objective function evaluations we were able to acheive the target convergence threshold in 6.5 seconds and do real-time processing on NVIDIA GTX580 and 2$\times$ NVIDIA K80 GPUs, respectively. The precision of integral Doppler (60s) is 2 mrad/s and 4 mrad/s for MEX(3-way) and Chang'E 4 relay satellite(3-way) respectively.

1 January 2010 in Lecture Notes in Computer Science
Verified
Book Chapter
Closed access
PDF file
DMDmitry Mikushin
Dmitry Mikushin
VSVictor Stepanenko
Victor Stepanenko