Switching Neural Networks: A New Connectionist Model for Classification

A new connectionist model, called Switching Neural Network (SNN), for the solution of classification problems is presented. SNN includes a first layer containing a particular kind of A/D converters, called latticizers, that suitably transform input vectors into binary strings. Then, the subsequent two layers of an SNN realize a positive Boolean function that solve in a lattice domain the original classification problem. Every function realized by an SNN can be written in terms of intelligible rules. Training can be performed by adopting a proper method for positive Boolean function reconstruction, called Shadow Clustering (SC). Simulation results obtained on the StatLog benchmark show the good quality of the SNNs trained with SC.

Connectionist Models

Connectionism is a theoretical framework for cognition whose principal tenets are (1) that all cognitive phenomena arise from the propagation of activation among simple neuronlike processing units and (2) that such propagation is mediated by weighted synapselike connections between units. Connectionist theories are typically instantiated as computer models, that is, computer programs that simulate how activation propagates through the system of interconnected units specified by the theory. Such models have profoundly influenced virtually every subdomain of cognitive science: perception and attention; word recognition, reading, derivational morphology, and other aspects of language; episodic, semantic, and short-term memory; action and other forms of sequential processing; executive function and cognitive control; many aspects of cognitive development; and even emotion. Many researchers, however, consider the connectionist framework to be incommensurable with the mainstream view that the human mind is a symbol-processing system and, for this reason, consider connectionist theories to be fundamentally flawed, or at least insufficient to serve as a general framework for cognition. This article considers the historical roots of connectionism, reviews key aspects of the approach, and addresses some of the criticisms it faces.

Multivariate Regression with Neural Networks: Unique, Exact and Generic Models

Michael Nielsen provides a visual demonstration in his web book Neural Networks and Deep Learning that a 1-layer deep neural network can match any function y = f(x). It is just a matter of the number of neurons to get a prediction that is arbitrarily close – the more the neurons the better the approximation. There is the Universal Approximation Theorem as well that supplies a rigorous proof of the same.But the known issues with overfitting remain and the obtained network model is only good for the range of the training data. That is, if the training data consisted only of inputs with x_1 < x < x_2 there would be no reason to expect the obtained network model to work outside of that range.
This series of posts are about obtaining network models that are unique, generic and exact. That is,
• they predict the correct output (the exact part)
• they generalize to all inputs irrespective of the data range used to train the model (the generic part)
• they can be obtained from any initial guess of the weights and biases (the unique part)

Gradients support in PyTorch

In this article by Maxim Lapan, the author of Deep Reinforcement Learning Hands-On,we are going to discuss about gradients in PyTorch. Gradients support in tensors is one of the major changes in PyTorch 0.4.0. In previous versions, graph tracking and gradients accumulation were done in a separate, very thin class Variable, which worked as a wrapper around the tensor and automatically performed saving of the history of computations in order to be able to backpropagate. Now gradients are a built-in tensor property, which makes the API much cleaner.

Using Topological Data Analysis to Understand the Behavior of Convolutional Neural Networks

TLDR: Neural Networks are powerful but complex and opaque tools. Using Topological Data Analysis, we can describe the functioning and learning of a convolutional neural network in a compact and understandable way. The implications of the finding are profound and can accelerate the development of a wide range of applications from self-driving everything to GDPR.

The Hidden Revolution in Data Science

When someone asks you if you´re taking full advantage of data science, you shouldn´t assume that you need to add chatbots to your product, or to throw money at a consultant who says they´re using deep learning. Rather, you should look at how people within your company are taking advantage of data analysis and visualization. It´s great that a lab at Google can use AI to build a self-driving car. But it´s also exciting when a marketing manager can use R to visualize how effective their ads are, or when a finance team can use Python to predict future revenue.

Let R/Python send messages when the algorithms are done training

As Data Scientists, we often train complex algorithms in order to tackle certain business problems and generate value. These algorithms, however, can take a while to train. Sometimes they take a couple of hours, hours which I´m not going to spend just sitting and waiting. But regularly checking whether the training is done, is also not the most efficient way. Now I started to use Telegram to send me notifications from R and Python to let me know when training is done. Furthermore, I´m also using it for example to send me notifications when pipelines / ETLs fail, which allows me to repair them as soon as they fail. It´s really easy, so I thought I´ll share my code!

Apache Pulsar 2.0 Brings Enterprise-Class Scale, Speed and Functionality to Streaming Data Processing

Streamlio, the intelligent platform for fast data, announced availability of Apache Pulsar 2.0, a significant new release of the streaming messaging solution at the core of the Streamlio platform. Building on the proven Apache Pulsar foundation, this release adds new capabilities to Apache Pulsar that support easy development and deployment of modern data-driven applications and demonstrate the maturity and enterprise-class capabilities of Pulsar while delivering significantly better performance, scalability and durability than older messaging platforms such as Apache Kafka, as verified in real-world OpenMessaging benchmark tests.

30 Free Resources for Machine Learning, Deep Learning, NLP & AI

This is a collection of free resources beyond the regularly shared books, MOOCs, and courses, mostly from over the past year. They start from zero and progress accordingly, and are suitable for individuals looking to pick up some of the basic ideas, before hopefully branching out further (see the final 2 resources listed below for more on that). These resources are not presented in any particular order, so feel free to pursue those which look most enticing to you. All credit goes the the individual authors of the respective materials, without whose hard work we would not have the benefit of learning from such great content.

Why the Data Lake Matters

This post outlines why the data lake matters, outlining the complexity of a data lake and taking a look at its evolution over time. H2O.ai: Automatic Machine Learning for Enterprise – Free 21 day trial

Extracting a Reference Grid of your Data for Machine Learning Models Visualization

Extracting a Reference Grid of your Data for Machine Learning Models Visualization

Handling Strings with R

This book aims to provide a panoramic perspective of the wide array of string manipulations that you can perform with R. If you are new to R, or lack experience working with character data, this book will help you get started with the basics of handling strings. Likewise, if you are already familiar with R, you will find material that shows you how to do more advanced string and text processing operations. Despite the fact that R may not be as rich and diverse as other scripting languages when it comes to string manipulation, it can take you very far if you know how. Sadly, documentation on how to manipulate strings and text data in R is very scarce. This work is my two cents to increase the number of available resources about this indispensable topic for any data scientist.