# US 2018/0025282 A1 · Systems and methods of determining sufficent causes from multiple outcomes

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2018/0025282 A1
- Document ID: US20180025282A1
- Status: Published application
- Kind: A1
- Application number: 15/215,513
- Filed: 2016-07-20
- Published: 2018-01-25
- Earliest claimed: 2016-07-20
- Assignees: Baidu USA LLC
- Applicants: Baidu USA, LLC
- Inventors: Yusheng Xie; Nan Du; Jing Zhai; Weicheng Zhu; Dawen Zhou; Wei Fan
- Web page: https://patentgrep.com/patent/US20180025282A1

## CPC

- G06N 7/01: Probabilistic graphical models, e.g. probabilistic networks
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Systems and methods are disclosed for question generation to infer the most probable cause from the observable outcome and known Noisy-OR causal relations. In embodiments, the outcomes are sorted by indices according to an order including but not limited to the outcomes' natural frequency order, expert-labeled order, machine-learning derived order, etc. According to their assigned indices, in embodiments, observed outcomes with lower indices are assigned for exact inference while observed outcomes with higher indices are assigned for variational inference. In embodiments, results of exact inference and variational inference are combined to predict the most probable cause. The unique combination of exact inference and variational inference according to outcome indices makes the probable cause inferring process faster.

## Claims

### Claim 1

A computer-implemented method for determining one or more potential causes from outcomes, the method comprising:
obtaining a set of observed outcomes;
ordering at least some of the observed outcomes according to an assigned indexing;
dividing the ordered outcomes into a first set of outcomes and a second set of outcomes based on the ordering;
determining an inferred probability for each cause of a set of causes using the first set of outcomes in a variational inference process and the second set of outcomes process in an exact inference process; and
outputting a set of probable causes as likely causes based upon the causes with probabilities above a threshold probability value.

### Claim 2 (depends on 1)

The method of claim 1 wherein the ordering step comprises removing one or more of the set of observed outcomes that are associated with low importance.

### Claim 3 (depends on 1)

The method of claim 1 wherein an outcome's assigned indexing is based upon its natural frequency of occurrence.

### Claim 4 (depends on 1)

The method of claim 1 wherein the assigned indexing is based upon an expert-labeled ordering.

### Claim 5 (depends on 1)

The method of claim 1 wherein the assigned indexing is ascertained from an outcome's machine learning derived order.

### Claim 6 (depends on 1)

The method of claim 1 further comprises picking a cause with highest probability from the set of causes as the most probable cause.

### Claim 7 (depends on 1)

The method of claim 1 further comprising updating the ordering using ground truth and machine learning.

### Claim 8 (depends on 1)

The method of claim 1 wherein the variational inference are implemented using convex conjugate minimization or loopy propagation.

### Claim 9 (depends on 1)

The method of claim 1 wherein the exact inference are implemented using quick score, singly-coupled pair elimination or quartet test.

### Claim 10

A method for determining potential causes from multiple sufficient outcomes, the method comprising:
obtaining a set of observables comprising positive observables and negative observables, the negative observables being the confirmations of not observing, the set of observations being related with a set of possible causes;
determining a single negative observable probability for each negative observable;
applying each single negative observable probability to determine a joint probability of the negative observables;
selecting exact inference or variational inference for each of the positive observables according to a present indexing assigned to each positive observable;
determining a probability of each positive observables selected for variational inference using variational transformation;
determining a joint probability of all positive observables selected for variational inference;
determining joint probability of the set of observables using the determined joint probability of all positive observables and determined joint probability of all negative observables; and
determining an inferred probability for each cause of a set of causes using the determined joint probability of the set of observables.

### Claim 11 (depends on 10)

The method of claim 10 further comprises filtering out some of the positive observables.

### Claim 12 (depends on 10)

The method of claim 10 further comprises optimizing an upper bound for the variational transformation before implementing the variational transformation.

### Claim 13 (depends on 12)

The method of claim 12 wherein optimizing the upper bound comprises:
transforming a probability function of each positive observable selected for variational transformation as a concave function; and
using a conjugate function of the concave function as the upper bound.

### Claim 14 (depends on 10)

The method of claim 10 further comprises picking a cause with highest probability from the set of causes as the most probable cause.

### Claim 15 (depends on 10)

The method of claim 10 wherein the variational inference are implemented using convex conjugate minimization or loopy propagation.

### Claim 16 (depends on 10)

The method of claim 10 wherein the probabilities of negative observables are transformed into a factorized form using an exact transformation.

### Claim 17

A non-transitory computer-readable medium or media comprising one or more sequences of instructions which, when executed by one or more processors, causes the steps to be performed comprising:
receiving a set of observables, the observables being related to a set of possible causes;
dividing the set of observables into a first set of observables and a second set of observables based on an index pre-assigned to each observable;
calculating a probability for each of a set of possible causes using variational inference for each observable of the first set an and exact inference for each observable of the second set; and
selecting a cause with highest calculated probability as the most probable cause.

### Claim 18 (depends on 17)

The non-transitory computer-readable medium or media of claim 17 wherein the set of observables comprises positive observables and negative observables, the negative observables being the confirmations of not observing.

### Claim 19 (depends on 18)

The non-transitory computer-readable medium or media of claim 18 wherein the negative observables are used for exact inference.

### Claim 20 (depends on 18)

The non-transitory computer-readable medium or media of claim 18 wherein the positive observables are omitted from calculation, used for exact inference, or used for variational inference depending on the pre-assigned indices assigned to the positive observables.