Accessing data from a database
Systems, methods, devices and non-transitory, computer-readable storage mediums are disclosed for a wearable multimedia device and cloud computing platform with an application ecosystem for processing multimedia data captured by the wearable multimedia device. In an embodiment, a method for using a machine learning model to provide data corresponding to a query comprises receiving a query for data stored in one or more databases. A prompt file is determined for the query. The prompt file and the query are provided as input for a machine learning model configured to generate a prediction for the query. Training the machine learning model can include updating network parameters in the machine learning model based on the prompt file. The prediction generated for the query is received. The prediction comprises a predicted database query corresponding to the query, and a level of accuracy is determined for the predicted database query.
1. A method for using a machine learning model to provide data corresponding to a user query, the method comprising:
receiving a user query for data stored in one or more databases, wherein the user query is represented in a first natural language;
determining a prompt file for the user query, wherein the prompt file comprises one or more pairs of example queries, wherein an example query pair comprises a first query represented in the first natural language, and a respective second query represented in a database language;
providing the prompt file and the user query as input for a machine learning model, wherein the machine learning model is configured to generate a prediction for the prompt file and the user query;
receiving the prediction generated for the prompt file and the user query, wherein the prediction comprises a predicted database query represented in the database language corresponding to the user query;
determining whether the predicted database query, based on a ground-truth database query for the user query, satisfies a level of accuracy; and
providing, using (i) the machine learning model and (ii) a virtual interface corresponding to a wearable multimedia device, data corresponding to one or more user queries.
2. The method of claim 1 , wherein determining the prompt file for the user query comprises:
determining a plurality of domains for the one or more databases; and
selecting, as a query domain, a domain from the plurality of domains for generating the prompt file for the user query.
3. The method of claim 2 , wherein the method further comprises:
in response to determining that the predicted database query does not satisfy the level of accuracy, updating the prompt file by generating an additional example query pair for the query domain, wherein the additional example query pair comprises a new first query represented in the first natural language, and a new second query represented in the database language, wherein at least one of the new first query or the new second query is different from the respective first queries and the respective second queries in the one or more pairs of example queries; and
providing the updated prompt file and the user query as input for the machine learning model.
4. The method of claim 1 , wherein the method further comprises:
in response to determining that the predicted database query satisfies the level of accuracy, receiving a second user query for data stored in the one or more databases, wherein the second user query is represented in the first natural language;
providing the second user query to the machine learning model for generating a prediction for the second user query;
receiving the prediction generated for the second user query, wherein the prediction comprises a second predicted database query represented in the database language corresponding to the second user query; and
fetching data from the one or more databases that corresponds to the second predicted database query,
wherein providing the data corresponding to the one or more user queries comprises generating an output for the second user query based on the fetched data.
5. The method of claim 2 , wherein providing the prompt file and the user query to the machine learning model comprises:
generating one or more additional prompt files for the query domain, each of the one or more additional prompt files comprising one or more pairs of example queries corresponding to the query domain; and
providing the one or more additional prompt files and the user query to the machine learning model.
6. The method of claim 1 , wherein providing the data corresponding to the one or more user queries using the virtual interface corresponding to the wearable multimedia device comprises projecting light on a surface proximate to the wearable multimedia device.
7. The method of claim 1 , comprising:
providing the prompt file and the user query as input for training the machine learning model, wherein training the machine learning model comprises updating network parameters in the machine learning model based on the one or more pairs of example queries in the prompt file.
8. A method for providing data corresponding to a user query, the method comprising:
receiving, using a wearable multimedia device, a user query for data stored in one or more databases, wherein the user query is represented in a first natural language;
providing the user query as input for a machine learning model that has been trained by processing at least one prompt file determined for a particular domain of the one or more databases, wherein the prompt file comprises one or more pairs of example queries, each example query comprising a respective first query represented in the first natural language, and a respective second query represented in a database language;
receiving a prediction for the user query as an output from the machine learning model, wherein the prediction comprises a predicted database query represented in the database language corresponding to the user query;
fetching data from the one or more databases that corresponds to the predicted database query; and
generating, using a virtual interface corresponding to the wearable multimedia device, an output based on the fetched data for the received user query.
9. The method of claim 8 , wherein after receiving the user query, the method further comprises:
receiving a second user query for data stored in the one or more databases, wherein the second user query is represented in a second natural language;
determining the second natural language of the second user query is different from the first natural language; and
in response to the determination, converting the second user query to be represented in the first natural language.
10. The method of claim 8 , wherein determining the prompt file for the user query comprises:
determining a plurality of domains for the one or more databases; and
selecting, as a query domain, a domain from the plurality of domains for generating the prompt file for the user query, wherein the particular domain of the one or more databases for training the machine learning model is the query domain.
11. The method of claim 10 , wherein selecting the domain as the query domain comprises:
for each of the plurality of domains, generating a respective domain feature vector for the domain;
generating a query feature vector for the user query; and
selecting a domain from the plurality of domains based on the respective domain feature vectors and the query feature vector.
12. The method of claim 8 , wherein generating the output based on the fetched data for the received user query comprises:
converting, using a generative language model, the fetched data to be represented in the first natural language.
13. The method of claim 9 , wherein generating the output based on the fetched data for the received second user query comprises:
converting the fetched data to be represented in the second natural language.
14. The method of claim 8 , wherein receiving the user query comprises receiving audio input using a microphone of the wearable multimedia device.
15. The method of claim 8 , wherein generating the output based on the fetched data using the virtual interface comprises projecting light on a surface proximate to the wearable multimedia device.
16. A system comprising one or more computers and one or more storage devices storing instructions that when executed by one or more computers cause the one or more computers to perform operations for providing data corresponding to a user query, wherein the operations comprise:
receiving a user query for data stored in one or more databases, wherein the user query is represented in a first natural language;
determining a prompt file for the user query, wherein the prompt file comprises one or more pairs of example queries, wherein an example query pair comprises a first query represented in the first natural language, and a respective second query represented in a database language;
providing the prompt file and the user query as input for a machine learning model, wherein the machine learning model is configured to generate a prediction for the prompt file and the user query;
receiving the prediction generated for the prompt file and the user query, wherein the prediction comprises a predicted database query represented in the database language corresponding to the user query;
determining whether the predicted database query, based on a ground-truth database query for the user query, satisfies a level of accuracy; and
providing, using (i) the machine learning model and (ii) a virtual interface corresponding to a wearable multimedia device, data corresponding to one or more user queries.
17. The system of claim 16 , wherein determining the prompt file for the user query comprising:
determining a plurality of domains for the one or more databases; and
selecting, as a query domain, a domain from the plurality of domains for generating the prompt file for the user query.
18. The system of claim 17 , wherein the operations further comprise:
in response to determining that the predicted database query does not satisfy the level of accuracy, updating the prompt file by generating an additional example query pair for the query domain, wherein the additional example query pair comprises a new first query represented in the first natural language, and a new second query represented in the database language, wherein at least one of the new first query or the new second query is different from the respective first queries and the respective second queries in the one or more pairs of example queries; and
providing the updated prompt file and the user query as input for the machine learning model.
19. The system of claim 16 , wherein the operations further comprise:
in response to determining that the predicted database query satisfies the level of accuracy, receiving a second user query for data stored in the one or more databases, wherein the second user query is represented in the first natural language;
providing the second user query to the machine learning model for generating a prediction for the second user query;
receiving the prediction generated for the second user query, wherein the prediction comprises a second predicted database query represented in the database language corresponding to the second user query; and
fetching data from the one or more databases that corresponds to the second predicted database query,
wherein providing the data corresponding to the one or more user queries comprises, generating an output for the second user query based on the fetched data.
20. The system of claim 17 , wherein providing the prompt file and the user query to the machine learning model comprises:
generating one or more additional prompt files for the query domain, each of the one or more additional prompt files comprising one or more pairs of example queries corresponding to the query domain; and
providing the one or more additional prompt files and the user query to the machine learning model.