Archive for the ‘machine learning for financial services’ Category

ai-one’s Biologically Inspired Neural Network

Sunday, February 1st, 2015

ai-one’s Learning Algorithm: Biologically Inspired Neural Network
– Introduction to HSDS vs ANN in Text Applications

Unlike any of the traditional neural nets, the neural network based on ai-one, the HoloSemantic Data Space neural network (invented by Manfred Hoffleisch) or in short “HSDS”, are massively connected, asymmetrical graphs which are stimulated by binary spikes. HSDS do not have any neural structures pre-defined by the user. Their building blocks resemble biological neural networks: a neuron has dendrites, on which the synapses from other neurons are placed, and an axon which ends in synapses at other neurons.

The connections between the neurons emerge in an unsupervised manner while the learning input is translated into the neural graph structure. The resulting graph can be queried by means of specific stimulations of neurons. In traditional neural systems it is necessary to set up the appropriate network structure at the beginning according to what is to be learned. Moreover, the supervised learning employed by neural nets such as the perceptron requires that a teacher be present who answers specific questions. Even neural nets that employ unsupervised learning (like those of Hopfield and Kohonen) require a neighborhood function adapted to the learning issue. In contrast, HSDS require neither a teacher nor a predefined structure or neighborhood function (note that although a teacher is not required, in most applications programmatic teaching is used to insure the HSDS has learned the content needed to meet performance requirements). In the following we characterize HSDS according to their most prominent features.

Exploitation of context

In ai-one applications like BrainDocs, HSDS is used for the learning of associative networks and feature extraction. The learning input consists of documents from the application domains, which are broken down into segments rather than entered whole: all sentences may be submitted as is or segmented into sub-sentences according to grammatical markers. By way of experimenting, we have discovered that a segment should ideally consist of 7 to 8 words. This is in line with findings from cognitive psychology. Breaking down text documents into sub-sentences is the closest possible approximation to the ideal segment size. The contexts given by the sub-sentence segments help the system learn. The transitivity of term co-occurrences from the various input contexts (i.e. segments) are a crucial contribution to creating appropriate associations. This can be compared with the higher-order co-occurrences explored in the context of latent semantic indexing.

Continuously evolving structure
The neural structure of a HSDS is dynamic and changes constantly in line with neural operations. In the neural context, change means that new neurons are produced or destroyed and connections reinforced or inhibited. Connections that are not used in the processing of input into the net for some time will get gradually weaker. This effect can also be applied to querying, which then results in the weakening of connections that are rarely traversed for answering a query.

Asymmetric connections
The connections between the neurons need not be equally strong on both sides and it is not necessary that a connection should exist between all the neurons (cp. Hopfield’s correlation matrix).

Spiking neurons
The HSDS is stimulated by spikes, i.e. binary signals which either fire or do not. Thresholds do not play a role in HSDS. The stimulus directed at a neuron is coded by the sequence of spikes that arrive at the dendrite.

Massive connectivity
Whenever a new input document is processed, new (groups of) neurons are created which in turn stimulate the network by sending out a spike. Some of the neurons reached by the stimulus react and develop new connections, whereas others, which are less strongly connected, do not. The latter nevertheless contribute to the overall connectivity because they make it possible to reach neurons which could not otherwise be reached. Given the high degree of connectivity, a spike can pass through a neuron several times since it can be reached via several paths. The frequency and the chronological sequence in which this happens determine the information that is read from the net

General purpose
There is no need to define a topology before starting the learning process because the neural structure of the HSDS develops on its own. This is why it is possible to retrieve a wide range of information by means of different stimulation patterns. For example, direct associations or association chains between words can be found, the words most strongly associated with a particular word can be identified, etc.

ai-one and the Machine Intelligence Landscape

Monday, January 12th, 2015

In the sensationally titled Forbes post, Tech 2015: Deep Learning And Machine Intelligence Will Eat The World, author Anthony Wing Kosner surveys the impact of deep learning technology in 2015. This is nothing new for those in the field of AI. His post reflects the recent increase in coverage artificial intelligence (AI) technologies and companies are getting in business and mainstream media. As a core technology vendor in AI for over ten years, it’s a welcome change in perspective and attitude.

We are pleased to see ai-one correctly positioned as a core technology vendor in the Machine Intelligence Landscape chart featured in the article. The chart, created by Shivon Zilis, investor at BloombergBETA, is well done and should be incorporated into the research of anyone seriously tracking this space.

Especially significant is Zilis’ focus on “companies that will change the world of work” since these are companies applying AI technologies to innovation and productivity challenges across the public and private sectors. The resulting solutions will provide real value through the combination of domain expertise (experts and data) and innovative application development.

This investment thesis is supported by the work of Erik Brynjolfsson and Andrew McAfee in their book “The Second Machine Age”, a thorough discussion of value creation (and disruption) by the forces of innovation that is digital, exponential and combinatorial. The impact of these technologies will change the economics of every industry over years if not decades to come. Progress and returns will be uneven in their impact on industry, regional and demographic sectors. While deep learning is early in Gartner’s Hype Cycle, it is clear that the market value of machine learning companies and data science talent are climbing fast.

This need for data scientists is growing but the business impact of AI may be limited in the near future by the lack of traditional developers who can apply them. Jeff Hawkins of Numenta has spoken out on this issue and we agree. It is a fundamentally different way to create an application for “ordinary humans” and until the “killer app” Hawkin’s speaks about is created, it will be hard to attract enough developers to invest time learning new AI tools. As the chart shows, there are many technologies competing for their time. Developers can’t build applications with buzzwords and one size fits all APIs or collections of open source algorithms. Technology vendors have a lot of work to do in this respect.

Returning to Kosner’s post, what exactly is deep learning and how is it different from machine learning/artificial intelligence? According to Wikipedia,

Deep learning is a class of machine learning training algorithms that use many layers of nonlinear processing units for feature extraction and transformation. The algorithms may be supervised or unsupervised and applications include pattern recognition and statistical classification.

  • are based on the (unsupervised) learning of multiple levels of features or representations of the data. Higher level features are derived from lower level features to form a hierarchical representation.
  • are part of the broader machine learning field of learning representations of data.
  • learn multiple levels of representations that correspond to different levels of abstraction; the levels form a hierarchy of concepts.
  • form a new field with the goal of moving toward artificial intelligence. The different levels of representation help make sense of data such as images, sounds and texts.

These definitions have in common (1) multiple layers of nonlinear processing units and (2) the supervised or unsupervised learning of feature representations in each layer, with the layers forming a hierarchy from low-level to high-level features.

While in the 4th bullet this is termed a new field moving toward artificial intelligence, it is generally considered to be part of the larger field of AI already. Deep learning and machine intelligence is not the same as human intelligence. Artificial intelligence in this definition above and in the popular press usually refers to Artificial General Intelligence (AGI). AGI and the next evolution, Artificial Super Intelligence (ASI) are the forms of AI that Stephen Hawking and Elon Musk are worried about.

This is powerful stuff no question, but as an investor, user or application developer in 2015 look for the right combination of technology, data, domain expertise, and application talent applied to a compelling (valuable) problem in order to create a disruptive innovation (value). This is where the money is over the new five years and this is our focus at ai-one.


ai-one Contributes to ETH Publication on Knowledge Representation

Tuesday, June 3rd, 2014

We are pleased to announce the availability of the following publication from prestigious ETH University in Zurich.  This book will be a valuable resource to developers, data scientists, search and knowledge management educators and practitioners trying to deal with the massive amounts of information in both public and private data sources.  We are proud to have our contribution to the field acknowledged in this way.

Knowledge Organization and Representation with Digital Technologies  |  ISBN: 978-3-11-031281-2

ai-one was invited to contribute as co-author to a chapter in this technical book.

ETH Publication- Knowledge RepresentationIn the anthology readers will find very different conceptual and technological methods for modeling and digital representation of knowledge for knowledge organizations (universities, research institutes and educational institutions), and companies based on practical examples presented in a synopsis. Both basic models of the organization of knowledge and technical implementations are discussed including their limitations and difficulties in practice.  In particular the areas of knowledge representation and the semantic web are explored. Best practice examples and successful application scenarios provide the reader with a knowledge repository and a guide for the implementation of their own projects. The following topics are covered in the articles:

  •  hypertext-based knowledge management
  • digital optimization of the proven analog technology of the list box
  • innovative knowledge organization using social media
  • search process visualization for digital libraries
  • semantic events and visualization of knowledge
  • ontological mind maps and knowledge maps
  • intelligent semantic knowledge processing systems
  • fundamentals of computer-based knowledge organization and integration

The book also includes coding medical diagnoses, contributions to the automated creation of records management models, business fundamentals of computer-aided knowledge organization and integration, the concept of mega regions to support of search processes and the management of print publications in libraries.

Available in German only at this time.

Wissensorganisation und -repräsentation mit digitalen Technologien  |  ISBN: 978-3-11-031281-2

ai-one war eigeladen worden, als CO-Autor ein Kapitel in diesem Sachbuch beizusteuern.

Im Sammelband werden die sehr unterschiedlichen konzeptionellen und technologischen Verfahren zur Modellierung und digitalen Repräsentation von Wissen in Wissensorganisationen (Hochschulen, Forschungseinrichtungen und Bildungsinstitutionen) sowie in Unternehmen anhand von  praxisorientierten Beispielen in einer Zusammenschau vorgestellt. Dabei werden sowohl grundlegende Modelle der Organisation von Wissen als auch technische Umsetzungsmöglichkeiten sowie deren Grenzen und Schwierigkeiten in der Praxis insbesondere in den Bereichen der Wissensrepräsentation und des Semantic Web ausgelotet. Good practice Beispiele und erfolgreiche Anwendungsszenarien aus der Praxis bieten dem Leser einen Wissensspeicher sowie eine Anleitung zur Realisierung von eigenen Vorhaben. Folgende Themenfelder werden in den Beiträgen behandelt:

  • Hypertextbasiertes Wissensmanagement
  • digitale Optimierung der erprobten analogen Technologie des Zettelkastens
  • innovative Wissensorganisation mittels Social Media
  • Suchprozessvisualisierung für Digitale Bibliotheken
  • semantische Event- und Wissensvisualisierung
  • ontologische Mindmaps und Wissenslandkarten
  • intelligente semantische Wissensverarbeitungssysteme

sowie Grundlagen der computergestützten Wissensorganisation und -integration, das Konzept von Mega-Regionen zur Unterstützung von Suchprozessen und zum Management von Printpublikationen in Bibliotheken, automatisierte Kodierung medizinischer Diagnosen sowie Beiträge zum Records Management zur Modellbildung und Bearbeitung von Geschäftsprozessen.

AI Goes to Wall Street: Trading Platforms Get Smarter

Thursday, June 30th, 2011

Press Release

It is no secret that for many years global banks have used artificial intelligence to make better trades. Now that technology might be coming to your local independent investment advisory service.

Caapi Technologies just announced that it signed a deal to use artificial intelligence technology to build custom trading systems for small to mid-size investment firms. Caapi will build applications with software development kits (SDKs) from ai-one that enable computers to understand human language to find undervalued stocks, bonds and derivatives.

The partnership makes Caapi one of the first consulting firms to use ai-one’s machine learning technology to build trading algorithms and platforms for traders, banks and hedge funds.

Building custom trading algorithms is a huge industry propelled by the success of high-frequency trading across global markets. Originally, these algorithms were designed to find and exploit pricing differences between stocks, commodities and derivatives. Now trading algorithms are so widespread and so sophisticated that they have completely reshaped markets to the point where pricing is often driven more by speculation than it is by the underlying value of the asset class.

The challenge now is to find underpriced opportunities that generate returns based on actual performance rather than market volatility. This requires that investors sort through vast amounts of unstructured data to find undervalued assets before they are identified by the rest of the market. Often this means reading text that can’t be processed by search engines like Google. Traditional algorithmic approaches, such as Google’s, fail as they only know what they are programmed to know or programmed to find. They miss finding unexpected results that don’t fit into an equation.

ai-one’s technology is described as “biologically inspired intelligence.” It is modeled after the human brain and does not depend on algorithms. Rather, it automatically sees the inherent patterns within data and forms associations between each data element. This enables machines to learn without any human intervention. More importantly, it enables people to ask the questions they wouldn’t normally know to ask.

The CEO of Caapi, Mr. Moris Oz, sees machine learning as the key to discovering hidden investment opportunities. “a-one’s technology enables us to build semantic associative search engines for our clients that understand how the price of any given investment is related to the unstructured data found on the internet.”

Caapi’s approach is to combine proven techniques using sophisticated algorithms with machine learning that understands words.  “Language is not math,” adds Olin Hyde, VP of Business Development at ai-one. “Algorithms are fantastic at processing structured data. But human behaviors and communications are inherently unstructured and complex. We learn through words not equations. So why not enable computers to do the same?”

According to Moris Oz, CEO of Caapi, “ai-one’s SDK for machine learning could be the answer for understanding and correlating soft data driving price moves in the markets.  I’m looking forward to applying this to new applications.” The market will soon tell if it works or not.

About Caapi Technologies, Founded by Moris Oz, the company offers consulting, system engineering, Algo trading machines and rigid body physics simulations. They design, program and deliver complex algorithmic and automated trading platforms. Caapi’s expertise spans the most common technology platforms such as Java, .NET, GWT, Flex, PHP, CSS, JS, Facebook SDK etc., for building scalable, feature-rich Web applications. Based in Israel, Caapi services encompass project management, software design, software development, quality assurance, documentation, and technical support.

For more information see

Contact Moris Oz, Ph +972-9-8656875 email

About ai-one inc., ai-one provides an “API for building learning machines”.  Based in San Diego, Zurich and Berlin, ai-one’s software technology is an adaptive holosemantic data space with semiotic capabilities (“biologically inspired intelligence”).  The Topic-Mapper™ SDK for text enables developers to create intelligent applications that deliver better sense-making capabilities for semantic discovery, lightweight ontologies, knowledge collaboration, sentiment analysis, artificial intelligence and data mining.

For more information see

Contact: Olin Hyde, Ph: 1-858-531-0674, email:, web: