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

Artificial Neural Networks have dramatically improved performance for many machine learning tasks. We demonstrate a new architecture for a fully optical neural network that enables a computational speed enhancement of at least two orders of magnitude and three orders of magnitude in power efficiency over state-of-the-art electronics.

Deep learning with coherent nanophotonic circuits

Blockchain is a decentralized transaction and data management technology developed first for Bitcoin cryptocurrency. The interest in Blockchain technology has been increasing since the idea was coined in 2008. The reason for the interest in Blockchain is its central attributes that provide security, anonymity and data integrity without any third party organization in control of the transactions, and therefore it creates interesting research areas, especially from the perspective of technical challenges and limitations. In this research, we have conducted a systematic mapping study with the goal of collecting all relevant research on Blockchain technology. Our objective is to understand the current research topics, challenges and future directions regarding Blockchain technology from the technical perspective. We have extracted 41 primary papers from scientific databases. The results show that focus in over 80% of the papers is on Bitcoin system and less than 20% deals with other Blockchain applications including e.g. smart contracts and licensing. The majority of research is focusing on revealing and improving limitations of Blockchain from privacy and security perspectives, but many of the proposed solutions lack concrete evaluation on their effectiveness. Many other Blockchain scalability related challenges including throughput and latency have been left unstudied. On the basis of this study, recommendations on future research directions are provided for researchers.

/pdf/where-is-current-research-on-blockchain-technology-a-530we6hlvi.pdf

Where Is Current Research on Blockchain Technology?-A Systematic Review.

Besides attracting a billion dollar economy, Bitcoin revolutionized the field of digital currencies and influenced many adjacent areas. This also induced significant scientific interest. In this survey, we unroll and structure the manyfold results and research directions. We start by introducing the Bitcoin protocol and its building blocks. From there we continue to explore the design space by discussing existing contributions and results. In the process, we deduce the fundamental structures and insights at the core of the Bitcoin protocol and its applications. As we show and discuss, many key ideas are likewise applicable in various other fields, so that their impact reaches far beyond Bitcoin itself.

Bitcoin and Beyond: A Technical Survey on Decentralized Digital Currencies

Recent advances in physical reservoir computing: A review

https://www.cse.iitk.ac.in/users/isaha/Publications/Journals/NC10.pdf/

Artificial neural networks in hardware: A survey of two decades of progress

Bitcoin prevents double-spending using the blockchain, a public ledger kept with every client. Every single transaction till date is present in this ledger. Due to this, true anonymity is not present in bitcoin. We present a method to enhance anonymity in bitcoin-type cryptocurrencies. In the blockchain, each block holds a list of transactions linking the sending and receiving addresses. In our modified protocol the transactions (and blocks) do not contain any such links. Using this, we obtain a far higher degree of anonymity. Our method uses a new primitive known as composite signatures. Our security is based on the hardness of the Computation Diffie-Hellman assumption in bilinear maps.

http://fc14.ifca.ai/bitcoin/papers/bitcoin14_submission_19.pdf

Increasing Anonymity in Bitcoin

This article presents a detailed overview of the hardware realizations for Artiflcial Neural Network (ANN) models and their variants appearing in academic studies as well as in commercial use. We outline underlying design issues encountered while mapping an ideal ANN model onto a compact and reliable hardware implementation and survey a wide range of illustrative examples with a speciflc focus on recent advancements. Issues related to the evaluation parameters for hardware implementations and hardware speciflc learning methodologies are outlined. Chip designs at neuronal level and as complete systems including neurocomputers built from accelerator boards as well as neurochips are studied. Parallel implementations including bit slice, systolic, and SIMD architectures have been described. Also discussions on optical implementations as well as Neuro-Fuzzy hardware are presented in addition to an outline of the recent trends and potential future research directions.

Artiflcial Neural Networks in Hardware: A Survey

Software clustering is an important technique for extracting high level component architecture from the underlying source code. One of the limitations of the existing approaches is that most of the proposed techniques use only similar types of features for estimating distance between source code elements. Therefore, in cases, where the selected features are poorly present in the source code, these techniques may not produce good quality results in absence of adequate inputs to work on. In this paper we propose an approach to overcome this limitation. Proposed approach uses a combination of multiple types of features together and applies automated weighing on the extracted features to enhance their information quality and to reduce noise. We define a way to estimate distance between code elements in terms of combination of multiple types of features. Weighted graph partitioning with a multi-objective global modularity criterion is used to select the clusters as architectural components. We describe methods for automated labeling of the extracted components and for generating inter-component interactions. We further discuss how the suggested approach extends to clustering at multiple hierarchical levels, to application portfolios, and even for improving precision for the feature location problem.

Software Clustering: Unifying Syntactic and Semantic Features

A method for component discovery from source code may include receiving source code, and determining business classes by excluding packages and classes in the source code identified as belonging to a presentation layer, as belonging to a data access layer, as models and/or as utilities. The method may further include extracting multi-dimensional features from the business classes, estimating similarity for business class pairs based on the extracted multi-dimensional features, clustering the business classes based on the similarity and mapping functional concepts to the clusters. The clusters generated by the clustering may represent components of the source code. The method may also include determining interfaces for the components based on the clustering.

Janardan Misra

Papers

Artificial neural networks in hardware: A survey of two decades of progress

Increasing Anonymity in Bitcoin

Artiflcial Neural Networks in Hardware: A Survey

Software Clustering: Unifying Syntactic and Semantic Features

Component discovery from source code