Method of training a neural network

• US 6,247,001 B1
• Filed: 09/03/1998
• Issued: 06/12/2001
• Est. Priority Date: 03/06/1996
• Status: Expired due to Term

• Current Assignee: Hanger Solutions, LLC (IP Investments Group LLC)
• Original Assignee: Siemens AG
• Inventors: Neuneier, Ralph, Tresp, Volker
• Primary Examiner(s): Powell, Mark R.
• Assistant Examiner(s): Holmes, Michael B.
• Application Number:
  US09/142,275
• Time in Patent Office: 1,013 Days
• Field of Search: 706/16, 706/21, 706/25, 706/15, 706/13, 705/35-37
• US Class Current: 706/15
• CPC Class Codes: G06N 3/08 Learning methods

Abstract

A state vector (SV_t) is determined with elements that characterize a financial market (101). Taking into account predetermined evaluation variables, an evaluation (V_t) is determined (102) for the state vector (SV_t). In addition, a chronologically following state vector (SV_t+1) is determined (103) and evaluated (V_t+1). On the basis of the two evaluations (V_t, V_t+1), weights (w_i) of the neural network (NN) are adapted (104) using a reinforcement learning method (Δw_i).

Claims

1. A method for training a neural network, comprising the following steps that are iteratively executed:

determining a state vector that has elements that characterize a financial market;

determining for the state vector, an evaluation relating to predetermined evaluation variables; and

adapting, using a reinforcement learning method, weights of the neural network, at least based on the evaluation of this state vector and based on a determined evaluation of at least one following state vector.

2. The method according to claim 1, wherein the state vector respectively has at least one of the following variables:

at least one rate of at least one stock index,

at least one indication of at least one bond market interest rate,

at least one exchange rate indication for at least one first currency into at least one second currency,

a gold price,

variables that describe a price-earnings ratio of at least one business enterprise.

3. The method according to claim 1, wherein the state vector respectively has at least one indication concerning a state of capital of an investor.

4. The method according to claim 1, wherein the predetermined evaluation variables describe at least one of an investment risk and at least one determined return of at least one type of investment.

5. The method according to claim 1, wherein the predetermined evaluation variables describe at least transaction costs of at least one investment type.

6. The method according to claim 1, wherein, given several possible following state vectors a probability of occurrence is determined for all following state vectors, and

wherein the evaluation is determined by summing the evaluations of all possible following state vectors, multiplied by a respective probability of occurrence of the state vector.

7. The method according to claim 1, wherein a reduction factor is provided, by which, as the number of iterative steps executed increases, the evaluations of the at least one following state vector are reduced.

8. A method for determining an investment decision, comprising the steps of:

training a neural network by iteratively executing the following steps;

a) determining a state vector that has elements that characterize a financial market,

b) determining for the state vector, an evaluation relating to predetermined evaluation variables,

c) adapting, using a reinforcement learning method, weights of the neural network, at least based on the evaluation of this state vector and based on a determined evaluation of at least one following state vector, the following state vector that results given the respective investment decision being determined for at least two of all possible investment decisions; and

recommending that investment decision to a user that leads to a higher evaluation of the following state vector with respect to the evaluation variables.

9. A method for determining an investment decision, comprising the steps of:

training a neural network by iteratively executing the following steps;

a) determining a state vector that has elements that characterize a financial market,

b) determining for the state vector, an evaluation relating to predetermined evaluation variables,

c) adapting, using a reinforcement learning method, weights of the neural network, at least based on the evaluation of this state vector and based on a determined evaluation of at least one following state vector, the following state vector that results given the respective investment decision being determined for at least two of all possible investment decisions; and

selecting that investment decision that leads to a higher evaluation of the following state vector with respect to the evaluation variables.

10. A method for determining an investment strategy, comprising the steps of:

training a neural network by iteratively executing the following steps;

a) determining a state vector that has elements that characterize a financial market,

b) determining for the state vector, an evaluation relating to predetermined evaluation variables,

c) adapting, using a reinforcement learning method, weights of the neural network, at least based on the evaluation of this state vector and based on a determined evaluation of at least one following state vector, several sequences of investment decisions, with various values of the evaluation variables, being determined;

determining a sequence of investment decisions that is optimal with respect to a predeterminable goal from the sequences of investment decisions; and

determining an investment strategy from the evaluation variables that were used in the optimal sequence of investment decisions.