-
Deep learning and face recognition: the state of the art
Authors:
Stephen Balaban
Abstract:
Deep Neural Networks (DNNs) have established themselves as a dominant technique in machine learning. DNNs have been top performers on a wide variety of tasks including image classification, speech recognition, and face recognition. Convolutional neural networks (CNNs) have been used in nearly all of the top performing methods on the Labeled Faces in the Wild (LFW) dataset. In this talk and accompa…
▽ More
Deep Neural Networks (DNNs) have established themselves as a dominant technique in machine learning. DNNs have been top performers on a wide variety of tasks including image classification, speech recognition, and face recognition. Convolutional neural networks (CNNs) have been used in nearly all of the top performing methods on the Labeled Faces in the Wild (LFW) dataset. In this talk and accompanying paper, I attempt to provide a review and summary of the deep learning techniques used in the state-of-the-art. In addition, I highlight the need for both larger and more challenging public datasets to benchmark these systems. The high accuracy (99.63% for FaceNet at the time of publishing) and utilization of outside data (hundreds of millions of images in the case of Google's FaceNet) suggest that current face verification benchmarks such as LFW may not be challenging enough, nor provide enough data, for current techniques. There exist a variety of organizations with mobile photo sharing applications that would be capable of releasing a very large scale and highly diverse dataset of facial images captured on mobile devices. Such an "ImageNet for Face Recognition" would likely receive a warm welcome from researchers and practitioners alike.
△ Less
Submitted 9 February, 2019;
originally announced February 2019.
-
RenderNet: A deep convolutional network for differentiable rendering from 3D shapes
Authors:
Thu Nguyen-Phuoc,
Chuan Li,
Stephen Balaban,
Yong-Liang Yang
Abstract:
Traditional computer graphics rendering pipeline is designed for procedurally generating 2D quality images from 3D shapes with high performance. The non-differentiability due to discrete operations such as visibility computation makes it hard to explicitly correlate rendering parameters and the resulting image, posing a significant challenge for inverse rendering tasks. Recent work on differentiab…
▽ More
Traditional computer graphics rendering pipeline is designed for procedurally generating 2D quality images from 3D shapes with high performance. The non-differentiability due to discrete operations such as visibility computation makes it hard to explicitly correlate rendering parameters and the resulting image, posing a significant challenge for inverse rendering tasks. Recent work on differentiable rendering achieves differentiability either by designing surrogate gradients for non-differentiable operations or via an approximate but differentiable renderer. These methods, however, are still limited when it comes to handling occlusion, and restricted to particular rendering effects. We present RenderNet, a differentiable rendering convolutional network with a novel projection unit that can render 2D images from 3D shapes. Spatial occlusion and shading calculation are automatically encoded in the network. Our experiments show that RenderNet can successfully learn to implement different shaders, and can be used in inverse rendering tasks to estimate shape, pose, lighting and texture from a single image.
△ Less
Submitted 1 April, 2019; v1 submitted 18 June, 2018;
originally announced June 2018.
-
Bipolaron Binding in Quantum Wires
Authors:
E. P. Pokatilov,
V. M. Fomin,
J. T. Devreese,
S. N. Balaban,
S. N. Klimin
Abstract:
A theory of bipolaron states in quantum wires with a parabolic potential well is developed applying the Feynman variational principle. The basic parameters of the bipolaron ground state (the binding energy, the number of phonons in the bipolaron cloud, the effective mass, and the bipolaron radius) are studied as a function of sizes of the potential well. Two cases are considered in detail: a cyl…
▽ More
A theory of bipolaron states in quantum wires with a parabolic potential well is developed applying the Feynman variational principle. The basic parameters of the bipolaron ground state (the binding energy, the number of phonons in the bipolaron cloud, the effective mass, and the bipolaron radius) are studied as a function of sizes of the potential well. Two cases are considered in detail: a cylindrical quantum wire and a planar quantum wire. Analytical expressions for the bipolaron parameters are obtained at large and small sizes of the quantum well. It is shown that at $R\gg 1$ [where $R$ means the radius (halfwidth) of a cylindrical (planar) quantum wire, expressed in Feynman units], the influence of confinement on the bipolaron binding energy is described by the function $\sim 1/R^{2}$ for both cases, while at small sizes this influence is different in each case. In quantum wires, the bipolaron binding energy $W(R) $ increases logarithmically with decreasing radius. The shapes and the sizes of a nanostructure, which are favorable for observation of stable bipolaron states, are determined.
△ Less
Submitted 19 April, 2000;
originally announced April 2000.
-
Quantum transport in the cylindrical nanosize silicon-based MOSFET
Authors:
S. N. Balaban,
E. P. Pokatilov,
V. M. Fomin,
V. N. Gladilin,
J. T. Devreese,
W. Magnus,
W. Schoenmaker,
M. Van Rossum,
B. Soree
Abstract:
A model is developed for a detailed investigation of the current flowing through a cylindrical nanosize MOSFET with a close gate electrode. The quantum mechanical features of the lateral charge transport are described by Wigner distribution function which is explicitly dealing with electron scattering due to acoustic phonons and acceptor impurities. A numerical simulation is carried out to obtai…
▽ More
A model is developed for a detailed investigation of the current flowing through a cylindrical nanosize MOSFET with a close gate electrode. The quantum mechanical features of the lateral charge transport are described by Wigner distribution function which is explicitly dealing with electron scattering due to acoustic phonons and acceptor impurities. A numerical simulation is carried out to obtain a set of I-V characteristics for various channel lengths. It is demonstrated that inclusion of the collision term in the numerical simulation is important for low values of the source-drain voltage. The calculations have further shown that the scattering leads to an increase of the electron density in the channel thereby smoothing out the threshold kink in I-V characteristics. An analysis of the electron phase-space distribution shows that scattering does not prevent electrons from flowing through the channel as a narrow stream, and that features of both ballistic and diffusive transport may be observed simultaneously.
△ Less
Submitted 2 April, 2000;
originally announced April 2000.
