770 likes | 920 Views
How to Correct the Flaws of the Patent System with a Patent?. NeuroLogic Sweden AB Business Incentive. Roland Orre <roland.orre@neurologic.se> Chairman, Director. Flaws of the Patent System.
E N D
How to Correctthe Flaws of the Patent System with a Patent? NeuroLogic Sweden AB Business Incentive Roland Orre <roland.orre@neurologic.se> Chairman, Director
Flaws of the Patent System • Main problem: The mathematical concept of software, has become considered a patentable technology. (I don’t consider it a technology) This problem is remarkable well expressed in US, but is becoming a problem also within EU. • Instead of stimulating innovation as was the original intention with the patent system, it now limits innovation and holds development back. • It creates a costly insecurity and decreasing investments in research and development. • Software patents create a form of anarchy.
NeuroLogic Provides Data Mining Services NeuroLogic is a software companyspecializing in R&D of data mining tools using Bayesian statistics and neural networks. NeuroLogic are developing and providing data mining methods and services for the World Health Organization used within the pharmacovigilance area. Since 1998 these methods started becoming a new standard within drug safety.
Two Main Services Early Warning Signalling on adverse drug relations. Unsupervised pattern recognition to find more complex relations like syndromes.
A great democratic decision for R&D in SW Article 3a in the amendend software patent directive: • Member states shall ensure that data processing is not considered to be a field of technology in the sense of patent law, and that innovations in the field of data processing are not considered to be inventions in the sense of patent law.
But... • A similar article exists earlier, but it has unfortunately not been followed by EPO, which has anyway granted around 30000 SW patents. • One of them is, for instance, an AI patent I got granted 1992 when I was developing expert systems at ABB.
SW Patents Limit Development • Instead of stimulating innovation as was the original intention with the patent system, it limits innovation and holds development back. • By causing R&D resources to be reallocated to lawyers and similar. • Also by being incentives for ”reinventing the wheel”, which is not productive (and for sw patents most often not possible, because sw patents are too broad).
About Standards • The sw patent system, instead of shaping standards, becomes an incentive to create new standards. A few simple examples: • Patent on MP3, was an incentive to create OOG, instead of doing something more creative. • Patent on GIF, was an incentive to create PNG. Neither a really productive invention. • Standards have to be free and open, otherwise it is a contradiction to call them standards.
A Paradoxical Solution • I realized that one of my old dream inventions, a system that invents from people’s ideas and wishes, could solve the problem. • A system which would speed up manufacturing of new products. • Thus, enhance the development process.
After a few years... • I finally realized how to implement this • and then immediately filed a patent. • The method is an AI approach to do business (based upon advanced data mining methods) • Actually also collaborate innovation, similar to the way GPL works.
This patented method will • Speed up the technical evolution. • It will over time correct the flaws of the patent system of today, that trivial and broad designs, as software, can be patented. • Enhance the patent system towards real innovations (new thinking, instead of trivial design patents). • Stimulate collaboration (collaborative innovation) • Stimulate shaping of standards within knowledge representation and generic design.
Patent system should • Not allow software to be patented! • Only be allowed when a patent also has a beneficial effect on the society, e.g. in shaping of standards, creating new jobs or other positive effects, like increasing collaboration.
Thank You! Roland Orre, PhD Director NeuroLogic Sweden AB roland.orre@neurologic.se
MyBackground • 2 years low tech industry. • National economy • MSc Engineering Physics • 12 years medium/high tech industry. • Patent ABB (AI, exp. sys) • Small consultancy work • Multi media pres. • Text data bases • Teaching ANN & prog tech. • Tool for interest prediction • SGML application • PhD, Comp Sc, patt rec.
How to Change This ? ? ? ? IDEAS KNOWLEDGE INFORMATION ? ?
Into This! INFORMATION KNOWLEDGE KNOWLEDGE INFORMATION KNOWLEDGE KNOWLEDGE INFORMATION INFORMATION KNOWLEDGE KNOWLEDGE INFORMATION
The Key is • Collaboration. • Data mining. • Patents as incentives to set standards.
Two Main Services • Early Warning Signalling on adverse drug relations. • Unsupervised pattern recognition to find syndromes.
Early Warning Signalling on Adverse Drug Reactions • One day you feel ill and rush to the doctor.
Early Warning Signalling on Adverse Drug Reactions • One day you feel ill and rush to the doctor. • You are then prescribed a drug for your heart.
Early Warning Signalling on Adverse Drug Reactions • One day you feel ill and rush to the doctor. • You are then prescribed a drug for your heart. • After a short while you get some unexpected reaction.
Early Warning Signalling on Adverse Drug Reactions • One day you feel ill and rush to the doctor. • You are then prescribed a drug for your heart. • After a short while you get some unexpected reaction.
Early Warning Signalling on Adverse Drug Reactions • One day you feel ill and rush to the doctor. • You are then prescribed a drug for your heart. • After a short while you get some unexpected reaction. • The doctor investigates you and suspects this to be an adverse drug reaction.
Early Warning Signalling on Adverse Drug Reactions • The doctor now writes a report on this, which will contain a lot of data about the patient. • Age, Sex, Country, Drugs taken, Other reactions, etc. • Over 70 variables are measured and reported.
40, Male, Sweden, Digoxin, Aspirin, Angry, ... The WHO Data Base • 50000 reports per quarter • from 70 countries • 3 million reports • maintained by UMC, a WHO collaborative
Recurrent BCPNN Create a BCPNN for drugs to investigate, nodes are drugs and adverse reactions Set weights by BCPNN formula: log (Pij/(Pi Pj)) Finds stored attractors, i.e. an associative memory.
Recurrent BCPNN: stimulate • Stimulate the network with a reported subset of adverse reaction from DB
Recurrent BCPNN: learn • Stimulate the network with a reported subset of adverse reaction from DB • The synaptic weight connections involved will grow.
Recurrent BCPNN: recall • Now stimulate the network with a pattern close to a stored attractor.
Recurrent BCPNN: recall • Now stimulate the network with a pattern close to a stored attractor. • Iterate the network to find the closest attractor.
Recurrent BCPNN: recall • Now stimulate the network with a pattern close to a stored attractor. • Iterate the network to find the closest attractor. A pattern which may never have been seen by the network is recalled!
Example: A recent syndrome node ci "DRUG X" 6276 "SOMNOLENCE" 1217 "AGITATION" 1162 "INSOMNIA" 953 "SUICIDE ATTEMPT" 952 "CONFUSION" 943 "ANXIETY" 762 "HALLUCINATION" 685 "NERVOUSNESS" 659 "AGGRESSIVE REACTION" 386 "DEPRESSION" 329 "MANIC REACTION" 299 "ANOREXIA" 280 "DEPERSONALIZATION" 256 "AMNESIA" 232 "THINKING ABNORMAL" 193 "DEPRESSION AGGRAVATED" 188 "EMOTIONAL LABILITY" 130 "PARANOID REACTION" 125 "PERSONALITY DISORDER" 121 "CONCENTRATION IMPAIRED" 100 "EUPHORIA" 53 "NEUROSIS" 53 "APATHY" 41
What is a syndrome? • An Invention! • Neither the doctor, nor the BCPNN have actually seen the whole pattern.
What is a syndrome? An Invention! Neither the doctor, nor the BCPNN have actually seen the whole pattern. The energy function of BCPNN maximizes the likelihood for these symptoms to occur together.
Comparision to Clustering • AutoClass is a Bayesian clustering method developed by NASA, originally for analysis of satellite images. • For haloperidol, an antipsychotic drug, BCPNN found 16 patterns, all clinically relevant, for haloperidol of which three conformed to well known (of totally five known) syndromes. This took 10.1 s (3.6s training and 6.5 s recall on PIII 1.4 GHz). • With Autoclass two patterns were found, both less clinically relevant. Autoclass was run for 20 hours on a PIII 1.4 GHz machine for this.
To What Other Business Areas can This Apply? • Marketing
To What Other Business Areas can This Apply? • Marketing of Research • Manufacturing on Demand
Research Problem 1 Funders in need of research Researchers in need of funders ? ? ? ? ? ? ? ? ? ?
Research Problem 2 Researchers in need of reviewers Potential reviewers ? ? ? ? ? ? ? ? ? ?
Research Problem 3 Researchers in need of information Information ? ? ? ? ? ? ? ? ? ? ?
Research Problem 3 Researchers in need of information Information ? ? ? ? ? ? ? ? IDEAS KNOWLEDGE INFORMATION ? ? ? ? ? ? ?
I’ll suggest a solution • Think about research as a product, produced by a demand. (also valid for some of the fundamental research if you have the right funders)
Assume Funder 1 Specifies <!DOCTYPE RESEARCH-PROJECT> <resproj> <title>Allocation of research resources</title> <abstract> I want a research project whichcan solve the problem how funders efficiently find their researchers.The problem is how to allocate these resources and make themfind each other. </abstract> <keywords> <keyword>funders</keyword> <keyword>efficient</keyword> <keyword>resources</keyword> <keyword>allocate</keyword> <keyword>finding</keyword> <keyword>researchers</keyword> </keywords> <investment>100000 Euro</investment> <deadline>Year 2005</deadline> </resproj>
Assume Researcher 1 specifies <!DOCTYPE RESEARCH-IDEA> <resproj> <title>Marketing of research</title> <abstract> I have an idea about how to solve the marketingproblem of research. How to market research. That is, how researchers could find their funders.The solution is by using a recurrent bayesian neural network. </abstract> <keywords> <keyword>funders</keyword> <keyword>marketing</keyword> <keyword>finding</keyword> <keyword>researchers</keyword> <keyword>bayesian</keyword> <keyword>neural</keyword> <keyword>network</keyword> </keywords> <costestimate>200000 Euro</costestimate> <timeestimate>1 year</timeestimate> <resources>4 people</resources> </resproj>
By applying... • A similar BCPNN technique which is described in the thesis, which has been succesfully applied to finding syndromes in the WHO database. • A similar text processing which we used in a feasability study for the British government (NPSA), when we got highest ranking in results and usage of advanced statistical methods between ten different companies, among others SAS and SPSS.
We have found a new pattern, a research project with funders. • By searching for patterns among the funder specifications, we get projects. • By searching for patterns among the researchers desires/ideas we get research groups with the same goals. • The research group can be given as a cue to find projects. • The project can be given as a cue to find research groups.