machine learning

Robotic laboratories are becoming commonplace, so computational data analysis has to as well

Andrew Walkingshaw — Fri, 02 May 2008 12:29:03 -0700

Description

It's not news that robots are better at routine experiments than postdocs are: Wikipedia points out, on the page on high-throughput screening, that

Automation is an important element in HTS's usefulness. Typically, an integrated robot system consisting of from one or more robots transports assay microplates from station to station for sample and reagent addition, mixing, incubation, and finally readout or detection. An HTS system can usually prepare, incubate, and analyze many plates simultaneously, further speeding the data-collection process. HTS robots currently exist which can test up to 100,000 compounds per day (Hann 2004). The term uHTS or ultra high throughput screening refers (circa 2008) to screening in excess of 100,000 compounds per day.

No scientist can review this many results by hand, but these methods are medically and technologically very important:

High-throughput screening methods have become essential for sifting through large chemical libraries in search of drug candidates, and several sensitive and reliable analytical techniques have been specifically adapted to high-throughput measurements of biocatalytic activity. High-throughput biocatalytic assay platforms thus enable rapid screening.

So to take advantage of the advances in experimental technique, we need advances in informatics to let us deal with this torrent of data.

Future of chemistry

Source

http://nihroadmap.nih.gov/
http://en.wikipedia.org/wiki/High-throughput_screening
http://dx.doi.org/10.1016/j.cbpa.2006.02.033

Automatic recognition of chemical terms in free text

Andrew Walkingshaw — Fri, 02 May 2008 12:07:50 -0700

Description

Recently, ClearForest (a division of Reuters) launched OpenCalais - a web service which reads the news for you:

The Calais web service automatically attaches rich semantic metadata to the content you submit – in well under a second. Using natural language processing, machine learning and other methods, Calais categorizes and links your document with entities (people, places, organizations, etc.), facts (person ‘x’ works for company ‘y’), and events (person ‘z’ was appointed chairman of company ‘y’ on date ‘x’).

In other words, it annotates text and marks up four of the journalistic 5 Ws - who, what, where and when, hopefully making it easier for journalists to join the dots and supply the why and how.

This seems only tangentially relevant to chemistry, at first glance, but chemistry's in a sense just a special case - we want to pull the whats (chemicals) and the hows (experimental methods) out of free text - papers, theses, and journal articles. That means chemical named entity recognition, and conveniently:

Oscar3 is a system for chemical natural language processing, focussing on chemical named entity recognition.

The Royal Society of Chemistry use OSCAR3 to annotate journals as part of Project Prospect; it lets them build, automatically, a searchable index of molecular structures (and substructures) published in their journals.

So, in that, there's some overlap with what the service the Chemical Abstracts Service provides; but the CAS model is based on a small army of editors indexing papers by hand, and as such can only scale up so far. On the other hand, there's no reason OSCAR-like robot annotators, even if less accurate than the humans, can't be turned loose on much more - patent applications, university thesis repositories, scientific blogs, mainstream scientific journalism, or anywhere there's chemical text to be read. That has the potential to open up a lot of 'hidden', informal or otherwise unpublished research to chemically-meaningful indexing and search.

Future of chemistry

Source

http://opencalais.com/
http://en.wikipedia.org/wiki/5_Ws
http://oscar3-chem.sourceforge.net/
http://www.rsc.org/Publishing/Journals/ProjectProspect/
http://www.cas.org/
http://cb.openmolecules.net/

Machine-to-Machine Intelligence (m2mi) Corporation

Matt Daniels — Mon, 28 Apr 2008 16:47:03 -0700

Description

Machine-to-Machine Intelligence (m2mi) Corporation: http://www.m2mi.com/

M2MI Corp markets products that "enable machines to assess their own behavior and intelligently adapt it... m2mi provides embedded enterprise software that enables management applications to converse with all machines via their preferred native language... via open standard Semantic Web technology, the same technology present in your internet browser. m2mi allows your machines to communicate with others using technology similar to RDF, RSS and Blogs."

Machine-to-Machine Intelligence(m2mi) Corp, a partner-centric company enables companies to automate operations of vast, global networks of computers and networked equipment. Machines in an m2mi environment interoperate seamlessly, because each is augmented with knowledge of its own behavior and can communicate with all others. Information, communication and intelligence enables global system awareness and adaptive control.

Our solutions enable machines to assess their own behavior and intelligently adapt it. We fully automate global system awareness and intelligent adaptive control. Equipped with m2mi's situation and behavior models, m2mi-enabled systems provide human managers simple, intelligible, decision-oriented situation analysis. Our entire solution stack rests on a secure trust model. Our technology answers the question, "How can we control our systems when their scale and speed exceed our human capacities to monitor, understand, and alter their behavior in predictable ways?" m2mi enables executives and managers to assure that their global systems behave adaptively and effectively.

m2mi provides embedded enterprise software that enables management applications to converse with all machines via their preferred native language. We enable such mediation via open standard Semantic Web technology, the same technology present in your internet browser. m2mi allows your machines to communicate with others using technology similar to RDF, RSS and Blogs

To process huge data volumes over large distances, transform data into information, and derive actionable intelligence, m2mi leverages a
meta-data-driven architecture (MDDA). Drone trawlers are used to proactively scan both the semantic web and traditional protocols to interact with
machines and applications. When new intelligence is to be deployed (new devices, change of applications or behavior), the drone trawler spawns and manages the life cycle of multiple task agents.

Computer & Information Science

Source

http://www.m2mi.com/

Post-semantic web enhances society and the meaning of data

David Gutelius — Sat, 24 Nov 2007 16:57:31 -0800

Description

In 1999, Tim Berners-Lee first described the semantic web in this way: " I have a dream for the Web [in which computers] become capable of analyzing all the data on the Web – the content, links, and transactions between people and computers. A ‘Semantic Web’, which should make this possible, has yet to emerge, but when it does, the day-to-day mechanisms of trade, bureaucracy and our daily lives will be handled by machines talking to machines. The ‘intelligent agents’ people have touted for ages will finally materialize."

Since that time, there has been significant progress towards making such an idea reality (note Radar Networks' Twine, or Metaweb's Freebase). It has also become more tightly constrained and defined (e.g. Wikipedia's current definition: "The Semantic Web is an evolving extension of the World Wide Web in which web content can be expressed not only in natural language, but also in a format that can be read and used by software agents, thus permitting them to find, share and integrate information more easily.").

Going beyond RDF-related technologies OWL and other ontology frameworks, however, we may be approaching a post-semantic web phase of development of the Internet. It's not that the "semantic web" as Tim B-L dreamed it or Wikipedia defines has really fully appeared. In fact, I have a suspicion that in either case, it may never appear and function the way its proponents envision. For one, there is still deep disagreement over standards - for all its Sematicness, the community can't even agree on the semantics!

By post-semantic web, I do not mean that it has become irrelevant - but it is beginning to show signs of turning out far differently than anyone could have imagined.

We are now seeing advanced machine learning combined with natural language processing, social graph analysis, and data mining techniques that half a decade ago few could have imagined. These technologies are being put to use by incredibly powerful compute resources (particularly those in mesh or p2p networks) to pick up and analyze a tremendous array of "signals". By signals, I mean not just those most in vogue in "web 2.0" like tags or networks of friends, although these are new and valuable sources for machines to learn to serve people more effectively. I also mean "digital gestures" - small signals that convey meaning to others but differently than "natural language" typically conveys; examples might include symbology or avatars. We are becoming more expressive digitally, and we are now just beginning to be able to also harvest these expressions and have machines learn from them in order to adapt to us.

The artificial intelligence field has for many years been fascinated with the idea of autonomous agents - semi-stupid digital servants that can act on our behalf under certain circumstances. The recent push into probabilistic reasoning and advances in a particular subfield of AI called machine learning (a characteristically poor name for a field of inquiry, but oh well) has begun to produce something better than semi-stupid in terms of serving us users.

The promise of a post-semantic web goes beyond just a language and representation framework (the techno-wonk vision of the semantic web) or a series of agents that do things for people. It's really a combination of 1) the power of distributed computing, 2) the growing expressivity of digital life and the signals such a life leaves behind, and 3) a way for software to learn and adapt itself to serve users and the human communities that they belong to, better. The implications for such powerful applications are not that they necessarily do things for us (although that would be a useful side effect), but rather give us new cognitive, and perhaps social, capabilities that let us do what we humans already do - just more and better.

AAAI Symposium on Social Information Processing - http://www.aaai.org/Symposia/Spring/sss08symposia.php#ss06

iLink KDD - http://www.ai.sri.com/pub_list/1523

Radar Networks' Twine: http://www.radarnetworks.com/

Metaweb - http://www.metaweb.com/

Signals

Radar Networks: Twine

Machine-to-Machine Intelligence (m2mi) Corporation

Social Information Processing