The Relation Between Acausality and Interference in Quantum-Like Bayesian Networks
1 like789 views
The document summarizes a research presentation on building quantum probabilistic models for decision making under uncertainty. It discusses: 1) Current Bayesian network models require manual parameter tuning and do not scale well for complex scenarios. 2) The presentation proposes a quantum-like Bayesian network that uses quantum interference effects and converts classical probabilities to quantum amplitudes. 3) A key challenge is that the number of quantum parameters grows exponentially large, making predictions sensitive and uncertain.
![Introduction to bayesian_networks[1]](https://cdn.slidesharecdn.com/ss_thumbnails/introductiontobayesiannetworks1-150525024327-lva1-app6891-thumbnail.jpg?width=560&fit=bounds)
More Related Content
Similar to The Relation Between Acausality and Interference in Quantum-Like Bayesian Networks (20)
Recently uploaded (20)
![LABORATORY-AQUIRED INFECTION [Autosaved].pptx](https://cdn.slidesharecdn.com/ss_thumbnails/laboratory-aquiredinfectionautosaved-250315002051-ab8c0ae4-thumbnail.jpg?width=560&fit=bounds)
The Relation Between Acausality and Interference in Quantum-Like Bayesian Networks
- 1. 9th International Conference on Quantum Interaction, 2015 by Catarina Moreira and Andreas Wichert (Instituto Superior T辿cnico, University of Lisbon, Portugal)
- 2. Motivation Quantum probability and interference effects play an important role in explaining several inconsistencies in decision-making. Moreira & Wichert (2014), Interference Effects in Quantum Belief Networks, Applied Soft Computing, 25, 64-85
- 3. Motivation Current models of the literature have the following problems: 1. Require a manual parameter tuning to perform predictions; 2. Hard to scale for more complex decision scenarios;
- 4. Research Question Can we build a general quantum probabilistic model to make predictions in scenarios with high levels of uncertainty?
- 5. Bayesian Networks Directed acyclic graph structure in which each node represents a random variable and each edge represents a direct influence from source node to the target node. 油 B Pr( 油E 油= 油T 油) 油= 油0.002 油 Pr( 油E 油= 油F 油) 油= 油0.998 油 Pr( 油B 油= 油T 油) 油= 油0.001 油 Pr( 油B 油= 油F 油) 油= 油0.999 油 B 油E 油 油 油 油 油 油 油 油 油Pr(A=T|B,E) 油 油 油 油 油 油 油 油 油 油 油 油Pr(A=F|B,E) 油 T 油T 油 油 油 油 油 油 油0.95 油 油 油 油 油 油0.05 油 T 油F 油 油 油 油 油 油0.94 油 油 油 油 油 油0.06 油 F 油T 油 油 油 油 油 油 油0.29 油 油 油 油 油 油0.71 油 F 油F 油 油 油 油 油 油 油 油0.01 油 油 油 油 油 油0.99 油 E A
- 6. Bayesian Networks Inference is performed in two steps: 1. Computation of the Full Joint Probability Distribution 2. Computation of the Marginal Probability 油 Full Joint Probability Distribution: 油 Marginal Probability: 油
- 7. Bayesian Networks Inference is performed in two steps: 1. Computation of the Full Joint Probability Distribution 2. Computation of the Marginal Probability 油 Full Joint Probability Distribution: 油 Marginal Probability: 油 Bayes Assumption
- 8. Inference in Bayesian Networks 1. Compute the Full Joint Probability Distribution B E A Pr( A, B, E ) T T T 0.001 x 0.002 x 0.95 = 0.00000190 T T F 0.001 x 0.002 x 0.05 = 0.00000010 T F T 0.001 x 0.998 x 0.94 = 0.00093812 T F F 0.001 x 0.998 x 0.06 = 0.00005988 F T T 0.999 x 0.002 x 0.29 = 0.00057942 F T F 0.999 x 0.002 x 0.71 = 0.00141858 F F T 0.999 x 0.998 x 0.01 = 0.00997002 F F F 0.999 x 0.998 x 0.99 = 0.98703198 B Pr( 油E 油= 油T 油) 油= 油0.002 油 Pr( 油E 油= 油F 油) 油= 油0.998 油 Pr( 油B 油= 油T 油) 油= 油0.001 油 Pr( 油B 油= 油F 油) 油= 油0.999 油 B 油E 油 油 油 油 油 油 油 油 油Pr( 油A=T|B,E 油) 油 油 油 油 油 油 油 油Pr( 油A=F|B,E 油) 油 T 油T 油 油 油 油 油 油 油0.95 油 油 油 油 油 油0.05 油 T 油F 油 油 油 油 油 油0.94 油 油 油 油 油 油0.06 油 F 油T 油 油 油 油 油 油 油0.29 油 油 油 油 油 油0.71 油 F 油F 油 油 油 油 油 油 油 油0.01 油 油 油 油 油 油0.99 油 E A
- 9. Inference in Bayesian Networks 2. Compute Marginal Probability B E A Pr( A, B, E ) T T T 0.001 x 0.002 x 0.95 = 0.00000190 T T F 0.001 x 0.002 x 0.05 = 0.00000010 T F T 0.001 x 0.998 x 0.94 = 0.00093812 T F F 0.001 x 0.998 x 0.06 = 0.00005988 F T T 0.999 x 0.002 x 0.29 = 0.00057942 F T F 0.999 x 0.002 x 0.71 = 0.00141858 F F T 0.999 x 0.998 x 0.01 = 0.00997002 F F F 0.999 x 0.998 x 0.99 = 0.98703198 + 油 + 油
- 10. Research Question 油 How can we move from a classical Bayesian Network to a Quantum- Like paradigm?
- 11. Quantum-Like Bayesian Networks General idea: - Under unobserved events, the Quantum-Like Bayesian Network can use interference effects; - Under known events, no interference is used, since there is no uncertainty. Moreira & Wichert (2014), Interference Effects in Quantum Belief Networks, Applied Soft Computing, 25, 64-85
- 12. Interference Effects Convert classical probabilities are converted into quantum amplitudes through Borns rule: squared magnitude quantum amplitudes. For two dichotomous random variables: - Classical Law of Total Probability: - Quantum Law of Total Probability:
- 13. Interference Effects Quantum Law of Total Probability: If we expand this term we obtain:
- 14. Interference Effects Quantum Law of Total Probability for 2 random variables: If we expand this term we obtain: Classical Probability Quantum Interference
- 15. Quantum-Like Bayesian Networks Convert classical probabilities are converted into quantum amplitudes through Borns rule: squared magnitude quantum amplitudes. - Classical Full Joint Probability Distribution: - Quantum Full Joint Probability Distribution:
- 16. Quantum-Like Bayesian Networks Convert classical probabilities are converted into quantum amplitudes through Borns rule: squared magnitude quantum amplitudes. - Classical Marginal Probability Distribution: - Quantum Marginal Probability Distribution:
- 17. Quantum-Like Bayesian Networks - Quantum marginal probability; - Extension of the Quantum-Like Approach (Khrennikov, 2009) for N random variables; Moreira & Wichert (2014), Interference Effects in Quantum Belief Networks, Applied Soft Computing, 25, 64-85
- 18. Case Study 油 We studied the implications of the proposed Quantum-Like Bayesian Network in the literature
- 19. Quantum-Like Bayesian Networks J. Pearl. (1988). Probabilistic Reasoning in Intelligent Systems. Morgan Kaufmann Publishers Inc. Pr( 油E 油= 油T 油) 油= 油0.002 油 Pr( 油E 油= 油F 油) 油= 油0.998 油 Pr( 油B 油= 油T 油) 油= 油0.001 油 Pr( 油B 油= 油F 油) 油= 油0.999 油 B 油E 油 油 油 油 油 油 油 油 油Pr(A=T|B,E) 油 油 油 油 油 油 油 油 油 油 油 油Pr(A=F|B,E) 油 T 油T 油 油 油 油 油 油 油0.90 油 油 油 油 油 油0.10 油 T 油F 油 油 油 油 油 油0.30 油 油 油 油 油 油0.70 油 F 油T 油 油 油 油 油 油 油0.20 油 油 油 油 油 油0.80 油 F 油F 油 油 油 油 油 油 油 油0.01 油 油 油 油 油 油0.99 油 E A J B M 油 油 油 油A 油 油 油 油 油 油Pr( 油M=T 油|A 油) 油 油 油 油 油 油Pr( 油M=F 油| 油A 油) 油 油 油 油 油T 油 油 油 油 油 油 油 油 油0.70 油 油 油 油 油 油0.30 油 油 油 油 油 油 油 油 油 油F 油 油 油 油 油 油 油 油 油0.01 油 油 油 油 油 油0.99 油 油 油 油 油A 油 油 油 油 油 油Pr( 油J=T 油|A 油) 油 油 油 油 油 油Pr( 油J=F 油| 油A 油) 油 油 油 油 油T 油 油 油 油 油 油 油 油 油0.90 油 油 油 油 油 油0.10 油 油 油 油 油 油 油 油 油 油F 油 油 油 油 油 油 油 油 油0.05 油 油 油 油 油 油0.95 油
- 20. Quantum-Like Bayesian Networks What happens if we try to compute the probability of A = t, given that we observed J = t? Classical Probability: Quantum Probability: Moreira & Wichert (2014), Interference Effects in Quantum Belief Networks, Applied Soft Computing, 25, 64-85
- 21. Quantum-Like Bayesian Networks What happens if we try to compute the probability of A = t, given that we observed J = t? Classical Probability: Quantum Probability: Moreira & Wichert (2014), Interference Effects in Quantum Belief Networks, Applied Soft Computing, 25, 64-85 Will generate 16 parameters
- 22. Problem! The number of parameters grows exponentially LARGE! The final probabilities can be ANYTHING in some range! Moreira & Wichert (2014), Interference Effects in Quantum Belief Networks, Applied Soft Computing, 25, 64-85
- 23. Problem! Quantum parameters are very sensitive. Small changes can lead to completely different probability values or can stabilize in a certain value!
- 24. Research Question 油 How can we deal automatically with an exponential number of quantum parameters?
- 25. The Synchronicity Principle Synchronicity is an acausal principle and can be defined by a meaningful coincidence which appears between a mental state and an event occurring in the external world. (Carl G. Jung, 1951)
- 26. The Synchronicity Principle Natural laws are statistical truths. They are only valid when dealing with macrophysical quantities. In the realm of very small quantities prediction becomes uncertain. The connection of events may be other than causal, and requires an acausal principle of explanation. Moreira & Wichert (2015), The Synchronicity Principle Under Quantum Probabilistic Inferences, NeuroQuantology, 13, 111-133
- 27. Research Question 油 How can we use the Synchronicity Principle in the Quantum-Like Bayesian Network and estimate quantum parameters?
- 28. Semantic Networks Moreira & Wichert (2015), The Synchronicity Principle Under Quantum Probabilistic Inferences, NeuroQuantology, 13, 111-133 Synchronicity Principle: defined by a meaningful coincidence between events. Semantic Networks can help finding events that share a semantic meaning.
- 29. Quantum-Like Bayesian Network + Semantic Network Synchronicity Pr( 油E 油= 油T 油) 油= 油0.002 油 Pr( 油E 油= 油F 油) 油= 油0.998 油 Pr( 油B 油= 油T 油) 油= 油0.001 油 Pr( 油B 油= 油F 油) 油= 油0.999 油 B 油E 油 油 油 油 油 油 油 油 油Pr(A=T|B,E) 油 油 油 油 油 油 油 油 油 油 油 油Pr(A=F|B,E) 油 T 油T 油 油 油 油 油 油 油0.90 油 油 油 油 油 油0.10 油 T 油F 油 油 油 油 油 油0.30 油 油 油 油 油 油0.70 油 F 油T 油 油 油 油 油 油 油0.20 油 油 油 油 油 油0.80 油 F 油F 油 油 油 油 油 油 油 油0.01 油 油 油 油 油 油0.99 油 E A J B M 油 油 油 油A 油 油 油 油 油 油Pr( 油M=T 油|A 油) 油 油 油 油 油 油Pr( 油M=F 油| 油A 油) 油 油 油 油 油T 油 油 油 油 油 油 油 油 油0.70 油 油 油 油 油 油0.30 油 油 油 油 油 油 油 油 油 油F 油 油 油 油 油 油 油 油 油0.01 油 油 油 油 油 油0.99 油 油 油 油 油A 油 油 油 油 油 油Pr( 油J=T 油|A 油) 油 油 油 油 油 油Pr( 油J=F 油| 油A 油) 油 油 油 油 油T 油 油 油 油 油 油 油 油 油0.90 油 油 油 油 油 油0.10 油 油 油 油 油 油 油 油 油 油F 油 油 油 油 油 油 油 油 油0.05 油 油 油 油 油 油0.95 油 Synchronicity Moreira & Wichert (2015), The Synchronicity Principle Under Quantum Probabilistic Inferences, NeuroQuantology, 13, 111-133
- 30. The Synchronicity Heuristic The interference term is given as a sum of pairs of random variables. Heuristic: parameters are calculated by computing different vector representations for each pair of random variables. Moreira & Wichert (2015), The Synchronicity Principle Under Quantum Probabilistic Inferences, NeuroQuantology, 13, 111-133
- 31. The Synchronicity Heuristic Since, in quantum cognition, the quantum parameters are seen as inner products, we represent each pair of random variables in 2- dimenional vectors. We need to represent both assignments of the binary random variables Moreira & Wichert (2015), The Synchronicity Principle Under Quantum Probabilistic Inferences, NeuroQuantology, 13, 111-133
- 32. The Synchronicity Heuristic Moreira & Wichert (2015), The Synchronicity Principle Under Quantum Probabilistic Inferences, NeuroQuantology, 13, 111-133 Using the semantic network, variables that did not share any dependence could be connected through their semantic meaning. Variables that occur during the inference process should be more correlated than variables that do not occur. We use a quantum step phase angle of /4 (Yukalov & Sornette, 2010).
- 33. The Synchronicity Heuristic 慮 油 慮 油 Moreira & Wichert (2015), The Synchronicity Principle Under Quantum Probabilistic Inferences, NeuroQuantology, 13, 111-133 Variables that occur during the inference process should be more correlated than variables that do not occur.
- 34. Research Question 油 How can an acausal connectionist theory affect quantum probabilistic inferences?
- 35. Classical vs Acausal Quantum Inferences High levels of uncertainty during the inference process, lead to complete different results from classical theory.
- 36. Classical vs Acausal Quantum Inferences More evidence leads to lower uncertainty, which leads to an approximation to the classical inference.
- 37. Conclusions 1. Applied the mathematical formalisms of quantum theory to develop a Quantum-Like Bayesian Network; 2. Used a Semantic Network to find acausal relationships; 3. An heuristic was created to estimate quantum parameters; 4. Quantum probability is stronger with high levels of uncertainty; 5. With less uncertainty, the Quantum-Like network collapses to its classical counterpart;
- 38. Some Concluding Reflections Can we validate this model for more complex decision problems? Can we propose an experiment?