-
Partially Stochastic Infinitely Deep Bayesian Neural Networks
Authors:
Sergio Calvo-Ordonez,
Matthieu Meunier,
Francesco Piatti,
Yuantao Shi
Abstract:
In this paper, we present Partially Stochastic Infinitely Deep Bayesian Neural Networks, a novel family of architectures that integrates partial stochasticity into the framework of infinitely deep neural networks. Our new class of architectures is designed to improve the computational efficiency of existing architectures at training and inference time. To do this, we leverage the advantages of par…
▽ More
In this paper, we present Partially Stochastic Infinitely Deep Bayesian Neural Networks, a novel family of architectures that integrates partial stochasticity into the framework of infinitely deep neural networks. Our new class of architectures is designed to improve the computational efficiency of existing architectures at training and inference time. To do this, we leverage the advantages of partial stochasticity in the infinite-depth limit which include the benefits of full stochasticity e.g. robustness, uncertainty quantification, and memory efficiency, whilst improving their limitations around computational complexity. We present a variety of architectural configurations, offering flexibility in network design including different methods for weight partition. We also provide mathematical guarantees on the expressivity of our models by establishing that our network family qualifies as Universal Conditional Distribution Approximators. Lastly, empirical evaluations across multiple tasks show that our proposed architectures achieve better downstream task performance and uncertainty quantification than their counterparts while being significantly more efficient. The code can be found at \url{https://github.com/Sergio20f/part_stoch_inf_deep}
△ Less
Submitted 2 June, 2024; v1 submitted 5 February, 2024;
originally announced February 2024.
-
Multi-domain learning CNN model for microscopy image classification
Authors:
Duc Hoa Tran,
Michel Meunier,
Farida Cheriet
Abstract:
For any type of microscopy image, getting a deep learning model to work well requires considerable effort to select a suitable architecture and time to train it. As there is a wide range of microscopes and experimental setups, designing a single model that can apply to multiple imaging domains, instead of having multiple per-domain models, becomes more essential. This task is challenging and someh…
▽ More
For any type of microscopy image, getting a deep learning model to work well requires considerable effort to select a suitable architecture and time to train it. As there is a wide range of microscopes and experimental setups, designing a single model that can apply to multiple imaging domains, instead of having multiple per-domain models, becomes more essential. This task is challenging and somehow overlooked in the literature. In this paper, we present a multi-domain learning architecture for the classification of microscopy images that differ significantly in types and contents. Unlike previous methods that are computationally intensive, we have developed a compact model, called Mobincep, by combining the simple but effective techniques of depth-wise separable convolution and the inception module. We also introduce a new optimization technique to regulate the latent feature space during training to improve the network's performance. We evaluated our model on three different public datasets and compared its performance in single-domain and multiple-domain learning modes. The proposed classifier surpasses state-of-the-art results and is robust for limited labeled data. Moreover, it helps to eliminate the burden of designing a new network when switching to new experiments.
△ Less
Submitted 20 April, 2023;
originally announced April 2023.
-
Enhanced IR Light Absorption in Group IV-SiGeSn Core-Shell Nanowires
Authors:
Anis Attiaoui,
Stephan Wirth,
André-Pierre Blanchard-Dionne,
Michel Meunier,
J. M. Hartmann,
Dan Buca,
Oussama Moutanabbir
Abstract:
Sn-containing Si and Ge alloys belong to an emerging family of semiconductors with the potential to impact group IV semiconductor devices. Indeed, the ability to independently engineer both lattice parameter and band gap holds the premise to develop enhanced or novel photonic, optoelectronic, and electronic devices. With this perspective, we present detailed investigations of the influence of Ge1-…
▽ More
Sn-containing Si and Ge alloys belong to an emerging family of semiconductors with the potential to impact group IV semiconductor devices. Indeed, the ability to independently engineer both lattice parameter and band gap holds the premise to develop enhanced or novel photonic, optoelectronic, and electronic devices. With this perspective, we present detailed investigations of the influence of Ge1-y-xSixSny layers on the optical properties of Si- and Ge-based heterostructures and nanowires. We found that adding a thin Ge1-x-ySixSny cap** layer on Si or Ge greatly enhances light absorption especially in the near IR range leading to an increase in short-circuit current density. For the Ge1-y-xSixSny structure at thicknesses below 30 nm, a 14-fold increase in the short-circuit current is predicted with respect to bare Si. This enhancement decreases by reducing the cap** layer thickness. Conversely, decreasing the shell thickness was found to improve the short-circuit current in Si/Ge1-y-xSixSny and Ge/Ge1-y-xSixSny core/shell nanowires. The optical absorption becomes very important when increasing the Sn content. Moreover, by exploiting optical antenna effect, these nanowires show an extreme light absorption reaching an enhancement factor, with respect to Si or Ge nanowires, on the order of ~104 in Si/Ge0.84Si0.04Sn0.12 and ~12 in Ge/Ge0.84Si0.04Sn0.12 core/shell nanowires. Furthermore, we analyzed the optical response of the addition of a dielectric cap** layer consisting of Si3N4 to the Si/Ge1-y-xSixSny core-shell nanowire and found about 50% increase in short-circuit current density for a dielectric layer thickness of 45 nm and a core radius and shell thickness superior to 40 nm. The core/shell optical antenna benefits from a multiplication of enhancements contributed by leaky mode resonances in the semiconductor part and antireflection effects in the dielectric part.
△ Less
Submitted 2 February, 2017;
originally announced February 2017.
-
Stability enhancement by joint phase measurements in a single cold atomic fountain
Authors:
M. Meunier,
I. Dutta,
R. Geiger,
C. Guerlin,
C. L. Garrido Alzar,
A. Landragin
Abstract:
We propose a method of joint interrogation in a single atom interferometer which overcomes the dead time between consecutive measurements in standard cold atomic fountains. The joint operation enables for a faster averaging of the Dick effect associated with the local oscillator noise in clocks and with vibration noise in cold atom inertial sensors. Such an operation allows achieving the lowest st…
▽ More
We propose a method of joint interrogation in a single atom interferometer which overcomes the dead time between consecutive measurements in standard cold atomic fountains. The joint operation enables for a faster averaging of the Dick effect associated with the local oscillator noise in clocks and with vibration noise in cold atom inertial sensors. Such an operation allows achieving the lowest stability limit due to atom shot noise. We demonstrate a multiple joint operation in which up to five clouds of atoms are interrogated simultaneously in a single setup. The essential feature of multiple joint operation, demonstrated here for a micro-wave Ramsey interrogation, can be generalized to go beyond the current stability limit associated with dead times in present-day cold atom interferometer inertial sensors.
△ Less
Submitted 8 January, 2015;
originally announced January 2015.
-
The Sagnac effect: 20 years of development in matter-wave interferometry
Authors:
Brynle Barrett,
Remi Geiger,
Indranil Dutta,
Matthieu Meunier,
Benjamin Canuel,
Alexandre Gauguet,
Philippe Bouyer,
Arnaud Landragin
Abstract:
Since the first atom interferometry experiments in 1991, measurements of rotation through the Sagnac effect in open-area atom interferometers has been studied. These studies have demonstrated very high sensitivity which can compete with state-of-the-art optical Sagnac interferometers. Since the early 2000s, these developments have been motivated by possible applications in inertial guidance and ge…
▽ More
Since the first atom interferometry experiments in 1991, measurements of rotation through the Sagnac effect in open-area atom interferometers has been studied. These studies have demonstrated very high sensitivity which can compete with state-of-the-art optical Sagnac interferometers. Since the early 2000s, these developments have been motivated by possible applications in inertial guidance and geophysics. Most matter-wave interferometers that have been investigated since then are based on two-photon Raman transitions for the manipulation of atomic wave packets. Results from the two most studied configurations, a space-domain interferometer with atomic beams and a time-domain interferometer with cold atoms, are presented and compared. Finally, the latest generation of cold atom interferometers and their preliminary results are presented.
△ Less
Submitted 1 December, 2014;
originally announced December 2014.
-
Molecular-dynamics thermal annealing model of laser ablation of silicon
Authors:
Patrick Lorazo,
Laurent J. Lewis,
Michel Meunier
Abstract:
A molecular-dynamics thermal annealing model is proposed to investigate the mechanisms involved in picosecond pulsed laser ablation of crystalline silicon. In accordance with the thermal annealing model, a detailed description of the microscopic processes which result from the interaction of a 308 nm, 10 ps, Gaussian pulse with a Si(100) substrate has been embedded into a molecular-dynamics sche…
▽ More
A molecular-dynamics thermal annealing model is proposed to investigate the mechanisms involved in picosecond pulsed laser ablation of crystalline silicon. In accordance with the thermal annealing model, a detailed description of the microscopic processes which result from the interaction of a 308 nm, 10 ps, Gaussian pulse with a Si(100) substrate has been embedded into a molecular-dynamics scheme. This was accomplished by explicitly accounting for carrier-phonon scattering and carrier diffusion. Below the predicted threshold fluence for ablation, $F_{th}=0.25 \text{J/cm}^{2}$, a surface breathing mode indicates that the solid restores internal equilibrium by the generation of pressure waves. Above $F_{th}$, our simulations reveal that matter removal is triggered by subsurface superheating effects: intense heating of the material leads to the thermal confinement of the laser-deposited energy. As a result, the material is overheated up to a temperature corresponding to the critical point of silicon and a strong pressure gradient builds up within the absorbing volume. At the same time, diffusion of the carriers into the bulk leads to the development of a steep temperature gradient beneath the surface. Matter removal is subsequently driven by the relaxation of the pressure gradient: large pieces --- several atomic layers thick --- of molten material are expelled from the surface with initial axial velocities of $\sim 1000 \text{m/s}$, their ejection following the nucleation of voids beneath the surface.
△ Less
Submitted 15 February, 2001;
originally announced February 2001.
-
Growth of three-dimensional structures by atomic deposition on surfaces containing defects : simulations and theory
Authors:
Pablo Jensen,
Hernán Larralde,
Muriel Meunier,
Alberto Pimpinelli
Abstract:
We perform a comprehensive study of the formation of three dimensional (pyramidal) structures in a large range of conditions, including the possible evaporation of adatoms from the surface and the presence of surface defects. We compare our computer simulations to theoretical calculations of the growth and find good agreement between them. This work clarifies precedent studies of three dimension…
▽ More
We perform a comprehensive study of the formation of three dimensional (pyramidal) structures in a large range of conditions, including the possible evaporation of adatoms from the surface and the presence of surface defects. We compare our computer simulations to theoretical calculations of the growth and find good agreement between them. This work clarifies precedent studies of three dimensional growth and predicts the island size distributions obtained in the different regimes. Finally, we show how our analysis can be used to interpret experimental data.
△ Less
Submitted 14 November, 1997;
originally announced November 1997.
