Компании
<<  Обеспечение финансовой устойчивости торговой компании в условиях кризиса Издательство Дрофа Представляет  >>
Programming Neural Networks and Fuzzy Systems in FOREX Trading
Programming Neural Networks and Fuzzy Systems in FOREX Trading
Content of the Presentation
Content of the Presentation
Basic course info: Purpose of the Course
Basic course info: Purpose of the Course
Basic course info: Course Agenda
Basic course info: Course Agenda
Basic course info: Accessibility of course materials
Basic course info: Accessibility of course materials
Basic course info: FOREX Trading Bot Building Contest
Basic course info: FOREX Trading Bot Building Contest
Basic course info: Requirements, Grading and Consultation
Basic course info: Requirements, Grading and Consultation
Content of the Presentation
Content of the Presentation
Basic terms of Stock Market/FOREX 1
Basic terms of Stock Market/FOREX 1
Basic terms of Stock Market/FOREX 2
Basic terms of Stock Market/FOREX 2
Fractal Theory of Stock/Currency Pair Prices
Fractal Theory of Stock/Currency Pair Prices
Basic terms of Distribution Free Estimatiors (DFE)
Basic terms of Distribution Free Estimatiors (DFE)
Basic terms of Rule-Based Systems (RBS) 1
Basic terms of Rule-Based Systems (RBS) 1
Basic terms of Rule-Based Systems (RBS) 2
Basic terms of Rule-Based Systems (RBS) 2
Basic terms of Learning Algorithms (LA)
Basic terms of Learning Algorithms (LA)
Content of the Presentation
Content of the Presentation
Neurons in Biology
Neurons in Biology
Comparison of Neuron with Silicon-based Hardware
Comparison of Neuron with Silicon-based Hardware
Biologic Neuron and its Mathematical Model
Biologic Neuron and its Mathematical Model
Content of the Presentation
Content of the Presentation
References 1
References 1
References 2
References 2
References 3
References 3
References 4
References 4

Презентация на тему: «Programming Neural Networks and Fuzzy Systems in FOREX Trading». Автор: Balazs Kovacs, Dr. Gabor Pauler. Файл: «Programming Neural Networks and Fuzzy Systems in FOREX Trading.ppt». Размер zip-архива: 854 КБ.

Programming Neural Networks and Fuzzy Systems in FOREX Trading

содержание презентации «Programming Neural Networks and Fuzzy Systems in FOREX Trading.ppt»
СлайдТекст
1 Programming Neural Networks and Fuzzy Systems in FOREX Trading

Programming Neural Networks and Fuzzy Systems in FOREX Trading

Presentation 0 Bal?zs Kov?cs (Terminator 2), PhD Student Faculty of Economics, University of P?cs E-mail: kovacs.balazs.ktk@gmail.com Dr. Gabor Pauler, Associate Professor Department of Information Technology Faculty of Science, University of P?cs E-mail: pauler@t-online.hu

2 Content of the Presentation

Content of the Presentation

Basic course info Purpose of the Course Course Agenda Accessibility of course materials FOREX Trading Bot Building Contest Requirements, Grading and Consultation Introduction Basic terms of Stock Market/FOREX Fractal Theory of Stock/Currency Pair Prices Basic terms of Distribution Free Estimators (DFE) Basic terms of Rule-Based Systems (RBS) Basic terms of Learning Algorithms (LA) Basic terms of Artificial Neural Networks (ANN) Biologic analogy: Neurons in Human brain Comparison with silicon-based hardware Biologic Neuron and its Mathematical Model Neural references

3 Basic course info: Purpose of the Course

Basic course info: Purpose of the Course

Nowadays there are wide range of fancy FOREX „course providers” which promise that you can be a millionaire within 4 weeks, without any serious economic and mathematic training, just completing 1 week rapid course where you learn drawing curves to charts visually. By contrast, we do not teach how you can be rapidly millionaire with FOREX. Instead of it we teach how to avoid loosing everything you have very rapidly with FOREX: Participants will be trained besides basics of FOREX and how to use Meta Trader 4 (MT4) FOREX Platform: Recognizing their psychologic limits and assembling customized trading strategies accordingly, The MQL programming language of MT4 to create your own indicators, Theoretical basics of Artificial Neural Networks and Fuzzy Systems, Using the Joone open-source Neural Shell under GNU license, programming it in Java and set up its data link with MT4. To motivate gifted students, we organize a FOREX Trading Bot Contest paralel with our course, where team 3-4 students can tune their software to reach maximum amount of return from limited amount of investment, within limited time frame making limited number of trades As a unique option Trading Bots with extremely high computational requirement can be run at the new University of Pecs Supercomputer in C++ environment The course has English language course material, even if it presented in Hungarian: To close out simple-minded Gamblers (Szerencsej?t?kos) Because – even if you have trade platforms and courses in Hungarian - almost every additional resource in FOREX you really need (eg. indicator source codes, economic analysises, user guides of trading bots) are most freshly available only in English! So the main outcome of this course is not being a millionaire in 4 weeks (which is unrealistic at FOREX anyway) but to develope proficiency using Artificial Neural Networks and Fuzzy Systems in a difficult simulated battleground called FOREX. And that knowledge can result getting better paid positions in many areas of engineeering or business

4 Basic course info: Course Agenda

Basic course info: Course Agenda

Legenda:

Presented by Dr. Gabor Pauler

Presented by Dr. Gabor Pauler

Presented by Dr. Gabor Pauler

Presented by Bal?zs Kov?cs

Presented by Bal?zs Kov?cs

Presented by Bal?zs Kov?cs

Week

Presentation

Quiz

Grade%

Practice

Home assignment

Grade%

0

Introduction, grading, Neural basics1

-

-

Forex basics 1

Install MT4, create account

3%

1

Neural basics 2, Learning methods: Hebb, Delta, Backpropagation1

Neural basics 1

3%

MT4 GUI

Basic trade in MT4

3%

2

Backpropagation2

Learning methods

3%

Basic indicators

Use of indicators

3%

3

Joone GUI

Backpropagation

3%

Indicator programming in MQL

MQL programming

3%

4

Time series forecasting networks and their representation in Joone

Joone GUI

3%

Compound indicators in MT4

Compound indicators

3%

5

Trading strategies in MT4

Time series nets

3%

Connecting MT4 with Joone

MT4-Joone connection

3%

6

Rule based systems: Crisp inference CRT

Trading strategy

3%

CRT in SPSS

CRT in SPSS

3%

7

Rule based systems: Fuzzy basics

CRT

3%

Fuzzy inference, multi valued results

Stock Futures

3%

8

FuzzyTech1

Fuzzy basics

3%

FuzzyTech2

Breasts

3%

9

Special topologies: RBF, ART, Kohonen

FuzzyTech

3%

Topology diagrams

Character design

3%

10

Neurofuzzy, FAM

Spec topology

3%

FAM in FuzzyTech

FAM

3%

11

Text mining basics

FAM

3%

Text mining in SPSS

Text mining

3%

12

Text mining topologies

Text mining

3%

Text mining in Joone

-

-

5 Basic course info: Accessibility of course materials

Basic course info: Accessibility of course materials

We can’t use TAB here!

All course materials are available at PTE-TTK Szent?gothai Szakkoll?gium website: ftp://szentagothai.ttk.pte.hu/pub/pauler/Forex/ in form of PowerPoint presentations and practices These are NOT conventional „three sentences/slide” projectable presenta-tions, but almost full-text materials with: Linked-in case study materials Step-by step animated software usage usable at computer lab However it is highly recommended for stu- dents to print them out in handout format and taking notes to slides, as questions in quiz may be represented from oral comments of tutor also All course materials are in English to cap- ture Business English But presentations are in Hungarian, and we have Hungarian Notes MT4 GUI can be both

6 Basic course info: FOREX Trading Bot Building Contest

Basic course info: FOREX Trading Bot Building Contest

To motivate gifted students, we organize a FOREX Trading Bot Contest paralel with our course, where team 3-4 students can tune their software to reach. Rules of the contest are: Server: FxPro MT4 Base currency: USD Maximum Leverage: 1:50 Demo account capital: 5000 USD http://usd.kurs24.com/huf/?q=5000 Platform: FxPro MT4 Client Terminal http://www.fxpro.com/hu/downloads/platforms/client-terminal Operating system of trading bot: Windows 2000, XP, Vista, Windows 7 Time range: 2011.11.10. 8:00:01 - 2011.12.08. 7:59:59 Trading hours: whenever markets are open Currency pairs: all possible pairs of EUR, USD, GBP, CHF, JPY, CAD, AUD, NZD can be traded in demo account Who can participate: registered students of current course Performance benchmarks: Passively managed static currency portfolio, Tutors demo account Maximal number of modifications on a trading bot: 5 Minimal number of trades completed by bot: 10 (without closing) Using any foreign code in bots without referencing it will result in immedaiate exclusion from contest Opened positions will be closed at the end of time range by tutors Winner team will be the one with the highest balance at the closing Identical balances among more teams will result in deuce Relative result% = Team balance/Tutor benchmark balance of teams compared to benchmarks can be published in university media/certificates

7 Basic course info: Requirements, Grading and Consultation

Basic course info: Requirements, Grading and Consultation

Mid-semester requirements: Max. 10 ? 3points = 30 points from simple 5-question quizes written at the beginning of presentations where students are evaluated individually Quizes are from the last presentation and practice Missed quizes can be substituted by one extra 6 point quiz ad the end of semester Max. 10 ? 3points = 30 points from home assignments evaluated at project team-level. Teams are free to reallocate their home assignment points internally to proportionate it to contribution of their members! Home assignments are due to the beginning of next practice Missed home assignments cannot be replaced after deadline as they are group assignments Max 40 team points from trading bot contest = 40 ? Relative result% Grading of individual students: 0-29points:Reject signing course(0), 30-49points: Fail(1), 50-59points: Sufficit(2), 60-69points: Medium(3), 70-79points: Good(4), 80-points: Excellent(5) In case of Fail(1), there are 2 possibilities for correction at oral exam from course material of presentations to get credit Consultations: Tutors will provide consultation at Department of Informatics, PTE-TTK, at times prearranged at pauler@t-online.hu or kovacs.balazs.ktk@gmail.com Results: Students can track their mid-semester results at ftp://szentagothai.ttk.pte.hu/pub/pauler/Forex/ExamForex/

8 Content of the Presentation

Content of the Presentation

Basic course info Purpose of the Course Course Agenda Accessibility of course materials FOREX Trading Bot Building Contest Requirements, Grading and Consultation Introduction Basic terms of Stock Market/FOREX Fractal Theory of Stock/Currency Pair Prices Basic terms of Distribution Free Estimators (DFE) Basic terms of Rule-Based Systems (RBS) Basic terms of Learning Algorithms (LA) Basic terms of Artificial Neural Networks (ANN) Biologic analogy: Neurons in Human brain Comparison with silicon-based hardware Biologic Neuron and its Mathematical Model Neural references

9 Basic terms of Stock Market/FOREX 1

Basic terms of Stock Market/FOREX 1

The Stock Exchange (r?szv?nyt?zsde) is a Non-profit Company (non-profit t?rsas?g ), what is Exclusive (Kiz?r?lagos) trading place of stocks of Publicly Quoted (t?zsd?re bevezetett, ny?lv?nosan ?rfolyam-jegyzett) Companies (r?szv?nyt?rsas?gok). These are larger, stabile firms complying strict Accounting (Sz?mviteli) rules. Stocks of smaller companies not quoted publicly are traded at Over The Counter (OTC) market. Macroeconomic (Makrogazdas?gi) function of stock exchange is Effective Allocation of Investment Resources (hat?konyan ossza el a v?llalkoz?sok k?zt a beruh?z?si er?forr?sokat) allocating more money to more profitable companies with larger growth in a public, open competition. Microoeconomic (Mikrogazdas?gi, v?llalati szint?) functions of Stocks/Equity (R?sz-v?nyt?ke): It helps Raise Funds (T?k?t gy?jt) necessary for operating a company and: Represents proportional Ownership/share (tulajdonr?sze) in a company, giving the right to Vote (Szavaz) in Board of Directors (igazgat?tan?cs) governing it, except: non-voting stocks Profitable companies pay Dividend (osztal?k) of profit for that, but it is not guaranteed, except: if it is Preferred Stock (els?bbs?gi/aranyr?szv?ny): always pays dividend, but cannot be sold and usually does not have vote Ordinary stocks can be Sold (eladhat?) on the stock exchange any time at Spot Stock Price (aktu?lis ?rfolyam), or at Futures (hat?rid?s ?rfolyam) except: if the company has Pre-emptive Option (el?v?teli jog), to block Hostile Takeover (t?mad? c?l? r?szv?nyfelv?s?rl?s) by competitor firms If we Bought (vett?k) or Underwrite (Lejegyezt?k) a stock in the past, and there was Hausse, Bull (?rfolyamemelked?s) we can earn Yield (?rfolyam-nyeres?g). If there was Baisse, Bear (cs?kken?s) then we Loose (veszt).

10 Basic terms of Stock Market/FOREX 2

Basic terms of Stock Market/FOREX 2

Equity is the most profitable but most risky tool of Investment Portfolio Management (T?ke befektet?si portfoli? menedzsment): You can buy and hold (Long) stocks of profitable and less risky companies (Blue Chips) to make profit from dividend or price increase, and liquidate stocks (Short) of bad companies to avoid loss. It can use 3 basic techniques: Hedge (Fedezeti ?gylet): to short/long a stock whose price tendencyously moves against a price of another stock or Currency (Valuta) longed/shorted to eliminate risk of loss from adverse price movement. Less risky, less profitable. Arbitrage (Arbitr?zs): short/long a stock very rapidly (1day-some hours) to make profit from minor price fluctuations. Medium risk, medium profitable. Speculation (Spekul?ci?): open a short/long Position (Poz?ci?) for longer time frame against the price Expectations (V?rakoz?s) of the whole market, and try to influence them with tricks to rapidly change their expectations. Very risky, very profitable. Actors of stock exchange: Broker (br?ker): does not own stock just trades it by Comission (Megb?z?s) of the owner for a Fee (D?j), Dealer (d?ler): can own stocks Underwriter (undervr?jter): can buy all stocks of a new company for re-sale. From broker to underwriter they have more rights to perform difficult and risky trades, but they have to comply more and more strict accounting and stock exchange rules FOREX, FOReign EXchange (Devizat?zsde) differs from ordinary stock exchange 2 ways: Instead of trading stocks against one Currency (Valuta) eg. (Sell IBM?for USD), several Foreign Exchanges (Deviza, valut?ra sz?l? sz?mlak?vetel?s) are traded against each other in Currency Pairs (Valutap?rok): eg. USD?EUR, GBP?CHF, JPY?EUR, etc. There are only brokers called FOREX companies/providers trading with someone else’s money, who want to hedge, arbitrage or speculate

11 Fractal Theory of Stock/Currency Pair Prices

Fractal Theory of Stock/Currency Pair Prices

Both at Stock Exchange/FOREX there is strong Information Asimmetry (Inform?ci? aszimmetria): most investors do not have any direct information about: Changing technology level and marketing efficiency of a company (denoted with green) Plans of Governments (Korm?ny) and Central Banks (K?zponti Bank) of 2 countries determining at most price of a given stock/currency pair long term They have to decide allocation of their money among stocks/currencies from partial information and their expectations, so they tend to fall in selling/buying panic at sudden big changes. Therefore, both Stock Exchange/FOREX are strictly controlled markets with many safety rules. But this will result in a Stepped (L?pcs?s) price (denoted with red) update behavior: Without strong external impulse brokers tend to build „dream worlds” setting up prices by their expectations ignoring slow and small changes of reality (eg. In „.com boom” of 2000s, small internet-based companies were worth more than General Electric and other industrial giants) But when the difference between them gets to big, they update price in smaller-bigger sudden steps, instead of continous change As prices are influenced by many different lenght cycles (eg. 1 year:seasons .. 1day:daily close), sudden steps are Embedded (Be?gyaz) into each other at several levels, it creates Fractal (Frakt?l)-type structures: price steps in time have self-similar details embedded into each other It makes Price Forecasting (?rel?rejelz?s) necessary for trading extremely difficult Function Estimation (F?ggv?ny becsl?si) problem: Prices of stocks/currency pairs are influenced by numerous parameters creating complex multivariate (Sokv?l- toz?s) functions Price data is very Noisy (Zajos) dis- torted by random disturbances, so Sto- chastic (Sztochasztikus) function esti- mation is necessary from a Sample (Minta) of prices Sometimes it is hard to assemble any function from future price expecta- tions collected from different informati- on sources: spot price( ) can adapt to reality in more alternative fractal path

12 Basic terms of Distribution Free Estimatiors (DFE)

Basic terms of Distribution Free Estimatiors (DFE)

Distribution Free Estimators (Eloszl?sf?ggetlen becsl?si rendszer) can estimate output of a complex, multivariate function from inputs. Functional transformation is estimated from previously observed (Megfigyelt) Sample (Minta) of noisy input-output values, and it does not make any assumptions on Probability Distribution (Val?sz?n?s?gi Eloszl?s) of sample. It means that the function can be reasonably complex. There are Analytic (F?ggv?nytani) methods of Approxi-mating (K?zel?t) complex functions: Taylor Series (Taylor-sor): it approximates a non-linear function (Eg. Sin(x)) with a suitably paramete-red higher order polynom in a given range Fourier Transform (Fourier-transzform?ci?): a complex nonlinar function (eg. Stock price, sound wave, etc.) is assembled as weighted sum of Sin(x) type functions with different wavelenght and phase. Evaluation of analytic methods: ? They have relatively low Computational requirement (Sz?mol?sig?ny) ? They require high level analytic mathematical knowledge ? They are not Modular (Modul?ris): modeling any additional local „bumps” or „steps” will result exponentially more complex global formulation Therefore, we will not deal with analytic approximation methods in this course. Instead of them, we will use Rule Based Systems, RBS (Szab?lyalap? rendszerek)

13 Basic terms of Rule-Based Systems (RBS) 1

Basic terms of Rule-Based Systems (RBS) 1

Rule-Based Distribution Free Estimators (szab?ly-alap? eloszl?sf?ggetlen becsl?si rendszerek) approximate Control Function (Vez?rl?si f?ggv?ny) ( ? ) among Input-Output Variables (I/O v?ltoz?k) of Decision Space (D?nt?si t?r) with the help of Rule Basis (Szab?lyb?zis) containing k=1..l finite set of rk Rules(Szab?ly): They Associate (Egym?shoz rendel) vi, vj values, or [vil, viu] intervals of xi i=1..n input and yo o=1..O output variables They have Linguistic(Nyelvi) representation: IF InputVar1 = Intreval AND InputVar2 = Interval AND.. THEN OutputVar = Interval They have Graphic (Grafikus) representation: multi-dimensional Hyperbars (Hipert?glatest) in decision space (we denote them Yellow ?) Rules of a rule basis can be Mutually Exclusive (k?lcs?n?sen egym?st kiz?r?ak): they have no Intersection (Metszet) = Common subset (K?z?s r?szhalmaz) in decision space. Alternatively, they can be Overlapping (?tlapol?ak) All rules of the basis has mx(rk) Validity (?rv?nyes-s?gi) value, which shows whether the rule is Valid/ Fires (T?zel) (Red ?) at a given x vector (Green O) of input variables: x ? rk If there is only one rule in the base to fire at any x input vector then rule basis is Non-Contradictive (Ellentmond?smentes), else Contradictive (?nel-lentmond?)

14 Basic terms of Rule-Based Systems (RBS) 2

Basic terms of Rule-Based Systems (RBS) 2

Effective approximation of continous functions would require large number of rules in the base to get rea-sonable Resolution (Felbont?s), eating up resources To avoid this, rules can have wk?[0,1] importance weights. Estimated otput yx* is computed as weighted sum of output values of firing rules. This is called Interpolation (Interpol?ci?) among rules: yx* = Sk wk ? mx(rk) ? yk (0.1) Interpolation enables to model continous control func-tions with less rules more effectively. It has 2 methods: Bayesian Probability (Bayes-i val?sz?n?s?g) rules: It uses mutually exclusive, Crisp (?les) rule base Where multiple rules can fire binary mx(rk)?{0,1} for a given x input vector But simutaneous firing rules Occour (Bek?vetkez) only with a pk?[0,1] Probability weight (Val?sz?n?s?gi s?ly), where sum of their probabilities is 1 creating Probability distribution (Val?sz?n?s?geloszl?s): Sk pk ? mx(rk) = 1 (0.2) ? It is supported by Probability theory (Elm?let) ? It requires data about probabilities of relatively large number of mutually exclusive rules, which is unrealistic to get in the practice Fuzzy Rule Inference (Fuzzy szab?ly k?vetkeztet?s): Rule basis has overlapping rules: Boundary (Hat?r) of Support (Tart?) of one rule are in the middle of support of neighboured rules mx(rk)?[0,1] validity of a rule can change continously: It is 1 in the middle of support and 0 at boundary (we denote it with yellow shading), forming not crisp/fuzzy rules: they occour certainly but their validity is uncertain/changing gradually wk weights do not form probability distribution ? Theoretically it is less sound method ? But can model complex nonlinear continous functions using much less rules/weights to tune

15 Basic terms of Learning Algorithms (LA)

Basic terms of Learning Algorithms (LA)

Manual definition of several thousand rules and their weights with the help of experts is expensive, slow Thats why Expert System Shells, ESS (Szak?rt?i rendszer shell) - using manual Bayesian probabilistic rule bases - failed to become the mainstream of Artificial Intelligence, AI (Mesters?ges Intelligencia) So we need Learning Algorithms(Tanul? Algoritmus) which can set up rules and their weights automati-cally form an X,Y Sample database (Minta adat-b?zis) of pre-viously Observed (Megfigyelt) j=1..m xj ,yj vectors of xi i=1..n input/yo o=1..O output vars. They have 2 groups: Classification and Regression Trees, CRT (Klasszifik?ci?s ?s regresszi?s f?k) algorithms: They can estimate only discrete valued (Diszkr?t ?rt?k?) output variables from continous/discrete inputs (Eg. Estimate Bankrupcy/Survival of a company from its financial rates) Building Decision tree (D?nt?si fa) of connected crisp Bayesian probability rules Trying to set up rule boundary values at each input variable, which separate best output values ? Low computational reqirement ? Can use only crisp hyperbar rules, which are ineffective modelling complex nonlinear Transversal (?tl?s) control functions Artificial Neural Networks, ANN (Mesters?ges Neur?lis H?l?zatok): they can estimate continous/discrete outputs from continous/discrete inputs Building kind of „implicte fuzzy rules”, without liguistic represntation and direct acces by user From random initial boundaries and rule weights They can model complex nonlinear, transversal control functions (Eg. Recognizing a letter „N” from dots of ink scanned in a picture) effectively At a price of difficult parametering and brutally high computational requirement

16 Content of the Presentation

Content of the Presentation

Basic course info Purpose of the Course Course Agenda Accessibility of course materials FOREX Trading Bot Building Contest Requirements, Grading and Consultation Introduction Basic terms of Stock Market/FOREX Fractal Theory of Stock/Currency Pair Prices Basic terms of Distribution Free Estimators (DFE) Basic terms of Rule-Based Systems (RBS) Basic terms of Learning Algorithms (LA) Basic terms of Artificial Neural Networks (ANN) Biologic analogy: Neurons in Human brain Comparison with silicon-based hardware Biologic Neuron and its Mathematical Model Neural references

17 Neurons in Biology

Neurons in Biology

Human brain contains 1011 Neurons (Idegsejt) connected with 1016 Synapses (Szinapszis) organized in Hemi-spheres (F?lteke) > Cortexes (K?reg) > Layers (Mez?) > Blocks Unisolated short Dend-rits(?g) transmit inco-ming electric signals at 2.3m/s to Cell membra-ne(Sejtfal)of neuron col-lecting electric charge At certain mV potential Treshold (Hat?r?rt?k), neuron emits electric signal by its Signal function (Jelz?si f?ggv?ny), which is tranmitted at 90m/s on long synapses covered with isolator Myelin (Mielin) jumping over the Ranvier-gaps (R?s) Excited Sytaptic termi-nals (V?gbunk?) emit Neurotransmitter(Inger?-let?tviv?) molecules (Eg. Acetilcolin, Opiats) Opening ion channels on other neurons mem-brane making them ac-cumulat electric charge

18 Comparison of Neuron with Silicon-based Hardware

Comparison of Neuron with Silicon-based Hardware

In the electron microscope image above we can see a neuron laid on leads of a modern microchip. Neuron is 10-12 times bigger than condensers and transistors of basic logic gates, however it can perform such a non-linear computing function, which requires hundreeds of basic logic gates in a math cooprocessor. Moreover, neurons require much less energy and produce much less heat than silicon-based chips. Currently a 100TByte blade-supercomputer compa-rable in storage capacity with human brain - but still inferior in speed, as brain can share work among 1011 simple processors instead of 103 more difficult ones - consumes 2-3 m3 space, 380V industrial current and cooling capacity of a supermarket Human brain consumes 1500cm3 volume even storing oxygene and glucose for 15-20 secs of work, and requires 5-10 Watts of power input and cooling

19 Biologic Neuron and its Mathematical Model

Biologic Neuron and its Mathematical Model

Fuctions of a neuron in ANN Mathematical Model: Non-volatile Memory (Permanens mem?ria): ji synapses connecting j=1..m neurons with i=1..n neurons in the network during t=0..T time periods transmit sjt?R signals of jth neuron in tth period with changing wjit Intensity/ Weight(S?ly). Teaching/ Training(Tan?t?s) of net means changing the initially random wji0?R weights. All information learnt is stored as synaptic weights Volatile Memory (R?vid t?v? mem?ria): a neuron aggregates wjit?sjt weighted signals of incoming synapses into a xit Membrane value (Membr?n ?rt?k) in the Activation Process (aktiv?ci?s folya-mat), additionally they Passively decay (Passz?v lecseng?s) membran value by (1-di) Decay Rate (Lecseng?si r?ta) to keep membrane value within [li, ui] Lower/Upper bounds (Als?/Fels? Korl?t) and smooth (Sim?t) its changes in time. There are 2 methods of membrane value aggregation: Additive (Addit?v): xit=di(Sj(wjit?sjt)/Sj(wjit))+ +(1-di)?xit-1 i=1..n, j=1..m, t=1..T (0.3) Multiplicative (Multiplikat?v): xit=diPj(sjtwjit)(1/ Sj(wjit))+ +(1-di)?xit-1, i=1..n, j=1..m, t=1..T (0.4) Aggregated membran value emits signal by mono-tonic increasing (Monoton n?vekv?) signal function with ai inflexion point as signal treshold and bi slope: sit = 1/ (1+e-bi?(xit-ai)), i=1..n, t=1..T (0.5)

20 Content of the Presentation

Content of the Presentation

Basic course info Purpose of the Course Course Agenda Accessibility of course materials FOREX Trading Bot Building Contest Requirements, Grading and Consultation Introduction Basic terms of Stock Market/FOREX Fractal Theory of Stock/Currency Pair Prices Basic terms of Distribution Free Estimators (DFE) Basic terms of Rule-Based Systems (RBS) Basic terms of Learning Algorithms (LA) Basic terms of Artificial Neural Networks (ANN) Biologic analogy: Neurons in Human brain Comparison with silicon-based hardware Biologic Neuron and its Mathematical Model Neural references

21 References 1

References 1

Hungarian language course notes: Notes Neural networks biologic analogy: http://health.howstuffworks.com/brain.htm Neural networks chatroom: http://www.geocities.com/siliconvalley/lakes/6007/Neural.htm GNU-licensed neural software: Source code libraries in C++, without install utility: SNNS: http://www-ra.informatik.uni-tuebingen.de/SNNS/ (+install and user guide) http://www.generation5.org/xornet.shtml http://www.netwood.net/~edwin/Matrix/ http://www.netwood.net/~edwin/svmt/ http://www.geocities.com/Athens/Agora/7256/c-plus-p.html http://www.cs.cmu.edu/afs/cs.cmu.edu/user/mitchell/ftp/faces.html http://www.cog.brown.edu/~rodrigo/neural_nets_library.html http://www.agt.net/public/bmarshal/aiparts/aiparts.htm http://www.geocities.com/CapeCanaveral/1624/ http://www.neuroquest.com/ http://www.grobe.org/LANE http://www.neuro-fuzzy.de/ http://www.cs.cmu.edu/afs/cs/project/ai-repository/ai/areas/neural/systems/cascor/ http://www.cs.cmu.edu/afs/cs/project/ai-repository/ai/areas/neural/systems/qprop/ http://www.cs.cmu.edu/afs/cs/project/ai-repository/ai/areas/neural/systems/rcc/

22 References 2

References 2

GNU-licensed neural software: Source code libraries in Java: Java Neural Networks by Jochen Fr?lich: http://fbim.fh-regensburg.de/~saj39122/jfroehl/diplom/e-index.html (Java Class, Internet applet about Kohonen-nets, free, no GUI, Tutorial in HTML) http://www.philbrierley.com/code http://rfhs8012.fh-regensburg.de/~saj39122/jfroehl/diplom/e-index.html http://neuron.eng.wayne.edu/software.html http://www.aist.go.jp/NIBH/~b0616/Lab/Links.html http://www.aist.go.jp/NIBH/~b0616/Lab/BSOM1/ http://www.neuroinformatik.ruhr-uni-bochum.de/ini/PEOPLE/loos http://www.neuroinformatik.ruhr-uni-bochum.de/ini/VDM/research/gsn/DemoGNG/GNG.html http://www.isbiel.ch/I/Projects/janet/index.html http://www.born-again.demon.nl/software.html http://www.patol.com/java/NN/index.html http://www-isis.ecs.soton.ac.uk/computing/neural/laboratory/laboratory.html http://www.neuro-fuzzy.de/ http://openai.sourceforge.net/ http://www.geocities.com/aydingurel/neural/ http://www-eco.enst-bretagne.fr/~phan/emergence/complexe/neuron/mlp.html Biologic modelling software: Neuron: http://www.neuron.yale.edu/neuron/ (free, GUI, Win XP install, Tutorial in HTML) Genesis: http://www.genesis-sim.org/GENESIS/ (free, GUI, Win XP install, Tutorial in HTML) PDP++: http://www.cnbc.cmu.edu/Resources/PDP++//PDP++.html (C++ source code library, GUI, Win XP install, Tutorial in HTML)

23 References 3

References 3

Decision support software: JNNS: http://www-ra.informatik.uni-tuebingen.de/software/JavaNNS (Simplified SNNS in Java, GUI, Win XP install, Tutorial in PDF) JOONE: http://www.joone.org (Java, GUI, Win XP install, Tutorial in PDF) Commercial neural decision support software: NeuroSolutions: http://www.neurosolutions.com/download.html (60 days shareware, no save, GUI, Win XP install, Excel Add-in, Excel Wizard, MATLAB modul, Tutorial in PDF Medical, automotive appliacations) NeurOK: http://soft.neurok.com/dm/download.shtml (Excel Add-in, C forr?sk?d, XML-es fel?let, Win XP install, financial applications) EasyNN: http://www.easynn.com/dlennp.htm (30 days shareware, GUI, Win XP install, Tutorial in HTML, financial forecasting applications) ALNFit Pro: http://www.dendronic.com/downloadalnfit_pro.shtml (30 days shareware, GUI, Win XP install, Tutorial in PDF, p?nz?gyi el?rejelz?si applications) Trajan: http://www.trajan-software.demon.co.uk/Downloads.htm (30 days shareware, GUI, Win XP install, Tutorial in HTML, no real application) AINet: http://www.ainet-sp.si/NN/En/nn.htm (1 days shareware, GUI, Win 95 install, Tutorial in PDF, nincs m?g val?s alkalmaz?sa) NeNet: http://koti.mbnet.fi/~phodju/nenet/Nenet/Download.html (performance limited shareware, GUI, Win 95 install, Tutorial in HTML, SOM networks oriented) Add-Ons for Statistical Packages: Statistica Neural Networks: https://www.statsoft.com/downloads/maintenance/download.html (no shareware, GUI, Win XP install, Tutorial in MPEG)

24 References 4

References 4

Add-Ons for MATLAB: Matlab Neural Toolbox: http://www.mathworks.com/products/neuralnet/ (No shareware) SOM ToolBox: http://www.cis.hut.fi/projects/somtoolbox/download/ (Matlab 5, free, GUI, Tutorial in PDF) FastICA: http://www.cis.hut.fi/projects/ica/fastica/code/dlcode.shtml (Matlab 7, free, GUI, Tutorial in PDF) NetLab: http://www.ncrg.aston.ac.uk/netlab/down.php (Matlab 5, free, GUI, Tutorial in PDF) NNSysID: http://www.iau.dtu.dk/research/control/nnsysid.html (Matlab 7, free, GUI, Tutorial in PDF) Excel Add-Ins in Financial Forecasting: NeuroShell: http://www.neuroshell.com/ (no shareware) NeuroXL: http://www.neuroxl.com/ (no shareware) Comparison of 50 commercial licensed neural software: http://wwwcs.uni-paderborn.de/~IFS/Tools/neural_network_tools.htm

«Programming Neural Networks and Fuzzy Systems in FOREX Trading»
http://900igr.net/prezentacija/ekonomika/programming-neural-networks-and-fuzzy-systems-in-forex-trading-197212.html
cсылка на страницу
Урок

Экономика

125 тем
Слайды
900igr.net > Презентации по экономике > Компании > Programming Neural Networks and Fuzzy Systems in FOREX Trading