Browsing by Author "Loppi, Niki"
Now showing 1 - 5 of 5
- Results Per Page
- Sort Options
- d3p - A Python Package for Differentially-Private Probabilistic Programming
A4 Artikkeli konferenssijulkaisussa(2022-04-01) Prediger, Lukas; Loppi, Niki; Kaski, Samuel; Honkela, AnttiWe present d3p, a software package designed to help fielding runtime efficient widely-applicable Bayesian inference under differential privacy guarantees. d3p achieves general applicability to a wide range of probabilistic modelling problems by implementing the differentially private variational inference algorithm, allowing users to fit any parametric probabilistic model with a differentiable density function. d3p adopts the probabilistic programming paradigm as a powerful way for the user to flexibly define such models. We demonstrate the use of our software on a hierarchical logistic regression example, showing the expressiveness of the modelling approach as well as the ease of running the parameter inference. We also perform an empirical evaluation of the runtime of the private inference on a complex model and find an ~10 fold speed-up compared to an implementation using TensorFlow Privacy. - Multi-node Training for StyleGAN2
A4 Artikkeli konferenssijulkaisussa(2021) Loppi, Niki; Kynkäänniemi, TuomasStyleGAN2 is a Tensorflow-based Generative Adversarial Network (GAN) framework that represents the state-of-the-art in generative image modelling. The current release of StyleGAN2 implements multi-GPU training via Tensorflow’s device contexts which limits data parallelism to a single node. In this work, a data-parallel multi-node training capability is implemented in StyleGAN2 via Horovod which enables harnessing the compute capability of larger cluster architectures. We demonstrate that the new Horovod-based communication outperforms the previous context approach on a single node. Furthermore, we demonstrate that the multi-node training does not compromise the accuracy of StyleGAN2 for a constant effective batch size. Finally, we report strong and weak scaling of the new implementation up to 64 NVIDIA Tesla A100 GPUs distributed across eight NVIDIA DGX A100 nodes, demonstrating the utility of the approach at scale. - Spatio-temporal variational Gaussian processes
A4 Artikkeli konferenssijulkaisussa(2021) Hamelijnck, Oliver; Wilkinson, William; Loppi, Niki; Solin, Arno; Damoulas, TheodorosWe introduce a scalable approach to Gaussian process inference that combines spatio-temporal filtering with natural gradient variational inference, resulting in a non-conjugate GP method for multivariate data that scales linearly with respect to time. Our natural gradient approach enables application of parallel filtering and smoothing, further reducing the temporal span complexity to be logarithmic in the number of time steps. We derive a sparse approximation that constructs a state-space model over a reduced set of spatial inducing points, and show that for separable Markov kernels the full and sparse cases exactly recover the standard variational GP, whilst exhibiting favourable computational properties. To further improve the spatial scaling we propose a mean-field assumption of independence between spatial locations which, when coupled with sparsity and parallelisation, leads to an efficient and accurate method for large spatio-temporal problems. - Ulkoisten kuormien vaikutukset lentokoneiden ominaisuuksiin
Insinööritieteiden korkeakoulu | Bachelor's thesis(2012-04-25) Loppi, Niki - Wall-modeling for large eddy simulation of rotating flows
Insinööritieteiden korkeakoulu | Master's thesis(2015-03-23) Loppi, NikiFlows affected by system rotation are common phenomena in engineering applications as well as in nature. The coordinate transformation into a rotating frame introduces two fictitious accelerations: the Coriolis acceleration and the centrifugal acceleration, which need to be included in the simulation. In this thesis, a spanwise rotating turbulent channel flow is studied through Large Eddy Simulations (LES). LES is a numerical modeling approach which is based on the decomposition of the turbulence spectrum into dynamically important large scales and homogeneous small scales. In LES, the large scales are resolved directly, while the effects of small-scales are modeled. In turbulent shear flows, the dynamically important scales are highly proportional to the Reynolds number within the inner boundary layer, which causes LES to be almost as expensive as Direct Numerical Simulation. By modeling the inner layer approximately, it is possible to bypass the very strict requirements of wall-resolved LES. In this thesis, firstly, the DNS case by Kristoffersen & Anderson [1] is reproduced to validate the implemented Coriolis source terms. After, a database for the wall-model analysis is established by performing wall-resolved high Reynolds number simulations with three different rotation rates. The wall-modeling approach by Kawai & Larsson [2] is then tested through an a priori analysis in which a standalone wall model is applied to wall-resolved results. Based on these results, a rotation correction, which adapts to the stability effects resulting from the system rotation, is proposed. Finally, this new rotation corrected wall-model is validated by performing a Wall-Modeled Large Eddy Simulation (WMLES). The WMLES results were found to be in good agreement with the wall-resolved data.