Deep learning is a form of machine learning that enables computers to learn from experience and understand the world in terms of a hierarchy of concepts. Because the computer gathers knowledge from experience, there is no need for a human computer operator to formally specify all the knowledge that the computer needs. The hierarchy of concepts allows the computer to learn complicated concepts by building them out of simpler ones; a graph of these hierarchies would be many layers deep. This book introduces a broad range of topics in deep learning. The text offers mathematical and conceptual background, covering relevant concepts in linear algebra, probability theory and information theory, numerical computation, and machine learning. It describes deep learning techniques used by practitioners in industry, including deep feedforward networks, regularization, optimization algorithms, convolutional networks, sequence modeling, and practical methodology; and it surveys such applications as natural language processing, speech recognition, computer vision, online recommendation systems, bioinformatics, and videogames. Finally, the book offers research perspectives, covering such theoretical topics as linear factor models, autoencoders, representation learning, structured probabilistic models, Monte Carlo methods, the partition function, approximate inference, and deep generative models. Deep Learning can be used by undergraduate or graduate students planning careers in either industry or research, and by software engineers who want to begin using deep learning in their products or platforms. A website offers supplementary material for both readers and instructors.

Deep Learning

“Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks”の学習報告

The science and technology of ultracapacitors are reviewed for a number of electrode materials, including carbon, mixed metal oxides, and conducting polymers. More work has been done using microporous carbons than with the other materials and most of the commercially available devices use carbon electrodes and an organic electrolytes. The energy density of these devices is 3¯5 Wh/kg with a power density of 300¯500 W/kg for high efficiency (90¯95%) charge/discharges. Projections of future developments using carbon indicate that energy densities of 10 Wh/kg or higher are likely with power densities of 1¯2 kW/kg. A key problem in the fabrication of these advanced devices is the bonding of the thin electrodes to a current collector such the contact resistance is less than 0.1 cm2. Special attention is given in the paper to comparing the power density characteristics of ultracapacitors and batteries. The comparisons should be made at the same charge/discharge efficiency.

Ultracapacitors: Why, How, and Where is the Technology

Much recent research has been devoted to learning algorithms for deep architectures such as Deep Belief Networks and stacks of auto-encoder variants, with impressive results obtained in several areas, mostly on vision and language data sets. The best results obtained on supervised learning tasks involve an unsupervised learning component, usually in an unsupervised pre-training phase. Even though these new algorithms have enabled training deep models, many questions remain as to the nature of this difficult learning problem. The main question investigated here is the following: how does unsupervised pre-training work? Answering this questions is important if learning in deep architectures is to be further improved. We propose several explanatory hypotheses and test them through extensive simulations. We empirically show the influence of pre-training with respect to architecture depth, model capacity, and number of training examples. The experiments confirm and clarify the advantage of unsupervised pre-training. The results suggest that unsupervised pre-training guides the learning towards basins of attraction of minima that support better generalization from the training data set; the evidence from these results supports a regularization explanation for the effect of pre-training.

/pdf/why-does-unsupervised-pre-training-help-deep-learning-4k2kgnets0.pdf

Why Does Unsupervised Pre-training Help Deep Learning?

http://www.diva-portal.org/smash/get/diva2:315882/FULLTEXT02

Nonlinear black-box modeling in system identification: a unified overview

Nonlinear black-box models in system identification: mathematical foundations

Neural networks for modelling and control of dynamic systems: M. Nørgaard, O. Ravn, N. K. Poulsen and L. K. Hansen. Springer-Verlag, London Berlin Heidelberg, 2000, pp. xiv+246

https://publications.lib.chalmers.se/records/fulltext/155128/local_155128.pdf

Component sizing of a plug-in hybrid electric powertrain via convex optimization

This paper presents a model-based algorithm that estimates how the driver of a vehicle can either steer, brake, or accelerate to avoid colliding with an arbitrary object. In this algorithm, the motion of the vehicle is described by a linear bicycle model, and the perimeter of the vehicle is represented by a rectangle. The estimated perimeter of the object is described by a polygon that is allowed to change size, shape, position, and orientation at sampled time instances. Potential evasive maneuvers are modeled, parameterized, and approximated such that an analytical expression can be derived to estimate the set of maneuvers that the driver can use to avoid a collision. This set of maneuvers is then assessed to determine if the driver needs immediate assistance to avoid or mitigate an accident. The proposed threat-assessment algorithm is evaluated using authentic data from both real traffic conditions and collision situations on a test track and by using simulations with a detailed vehicle model. The evaluations show that the algorithm outperforms conventional threat-assessment algorithms at rear-end collisions in terms of the timing of autonomous brake activation. This is crucial for increasing the performance of collision-avoidance systems and for decreasing the risk of unnecessary braking. Moreover, the algorithm is computationally efficient and can be used to assist the driver in avoiding or mitigating collisions with all types of road users in all kinds of traffic scenarios.

Jonas Sjöberg

Papers

Nonlinear black-box modeling in system identification: a unified overview

Nonlinear black-box models in system identification: mathematical foundations

Neural networks for modelling and control of dynamic systems: M. Nørgaard, O. Ravn, N. K. Poulsen and L. K. Hansen. Springer-Verlag, London Berlin Heidelberg, 2000, pp. xiv+246

Component sizing of a plug-in hybrid electric powertrain via convex optimization

Model-Based Threat Assessment for Avoiding Arbitrary Vehicle Collisions