Incorporating unstructured data into machine learning-based phenotyping
Implementations are described herein for incorporating unstructured data into machine learning-based phenotyping. In various implementations, natural language textual snippet(s) may be obtained. Each natural language textual snippet may describe environmental or managerial features of an agricultural plot that exist during a crop cycle. A sequence-to-sequence machine learning model may be used to encode the natural language snippet(s) into embedding(s) in embedding space. The embedding(s) may semantically represent the environmental or managerial features of the agricultural plot. Using one or more phenotypic machine learning models, phenotypic prediction(s) may be generated about the agricultural plot based on the one or more semantic embeddings and additional structured data about the agricultural plot. Output may be provided at one or more computing devices that is based on one or more of the phenotypic predictions.
1. A method implemented using one or more processors, the method comprising:
obtaining one or more natural language textual snippets, each natural language textual snippet describing one or more environmental or managerial features of an agricultural field that exist during a crop cycle;
encoding, using a sequence encoder machine learning model, the one or more natural language snippets into one or more semantic embeddings in embedding space, wherein the one or more semantic embeddings semantically represent the one or more environmental or managerial features of the agricultural field;
quantifying an amount of ambiguity of the encoding of the one or more natural language snippets into the one or more embeddings;
generating, using one or more phenotypic machine learning models one or more phenotypic predictions about the agricultural field based on the one or more semantic embeddings, the amount of ambiguity, and additional structured data about the agricultural field, the one or more phenotypic machine learning models to assign a greater weight to a first semantic embedding generated from an unambiguous natural language snippet than to a second semantic embedding generated from an ambiguous natural language snippet, wherein the structured data includes sensor data gathered by one or more sensors carried through the agricultural field by one or more agricultural vehicles, and the one or more of the phenotypic predictions are generated using the sensor data; and
causing output to be provided at one or more computing devices, wherein the output is based on one or more of the phenotypic predictions.
2. The method of claim 1 , wherein the sequence encoder machine learning model includes at least part of a transformer network.
3. The method of claim 1 , wherein one or more of the natural language snippets is obtained from speech recognition output generated using a spoken utterance captured at a microphone.
4. The method of claim 1 , wherein one or more of the natural language snippets is obtained from electronic correspondence exchanged between two or more individuals associated with the agricultural field.
5. The method of claim 1 , wherein one or more of the phenotypic predictions comprises crop yield.
6. The method of claim 1 , wherein the one or more phenotypic machine learning models include comprise a mixture of experts ensemble that includes:
a first phenotypic expert model to encode the one or more natural language textual snippets into the one or more embeddings, and
a second phenotypic expert model to process the structured data about the agricultural field.
7. The method of claim 6 , wherein the mixture of experts further includes a third phenotypic expert model that is used to process outputs of the first and second phenotypic expert models.
8. The method of claim 7 , wherein the third phenotypic expert model includes a gating network.
9. The method of claim 1 , wherein the sensor data includes image data gathered by one or more vision sensors carried through the agricultural field by the one or more agricultural vehicles.
10. The method of claim 9 , wherein the processing of the image data is performed using one or more convolutional neural networks.
11. A system comprising one or more processors and memory storing instructions that, in response to execution of the instructions, cause the one or more processors to:
obtain one or more natural language textual snippets, each natural language textual snippet describing one or more environmental or managerial features of an agricultural field that exist during a crop cycle;
encode, using a sequence encoder machine learning model, the one or more natural language snippets into one or more semantic embeddings in embedding space, wherein the one or more semantic embeddings semantically represent the one or more environmental or managerial features of the agricultural field;
quantify an amount of ambiguity of the encoding of the one or more natural language snippets into the one or more embeddings;
generate, using one or more phenotypic machine learning models, one or more phenotypic predictions about the agricultural field based on the one or more semantic embeddings, the amount of ambiguity, and additional structured data about the agricultural field, the one or more phenotypic machine learning models to assign a greater weight to a first semantic embedding generated from an unambiguous natural language snippet than to a second semantic embedding generated from an ambiguous natural language snippet, wherein the structured data includes sensor data gathered by one or more sensors carried through the agricultural field by one or more agricultural vehicles, and the one or more of the phenotypic predictions are generated using the sensor data; and
cause output to be provided at one or more computing devices, wherein the output is based on one or more of the phenotypic predictions.
12. The system of claim 11 , wherein the sequence encoder machine learning model includes at least part of a transformer network.
13. The system of claim 11 , wherein one or more of the natural language snippets is obtained from speech recognition output generated using a spoken utterance captured at a microphone.
14. The system of claim 11 , wherein one or more of the natural language snippets is obtained from electronic correspondence exchanged between two or more individuals associated with the agricultural field.
15. The system of claim 11 , wherein one or more of the phenotypic predictions comprises crop yield.
16. The system of claim 11 , wherein the one or more phenotypic machine learning models includes a mixture of experts ensemble that includes:
a first phenotypic expert model to encode the one or more natural language textual snippets into the one or more embeddings, and
a second phenotypic expert model to process the structured data about the agricultural field.
17. The system of claim 16 , wherein the mixture of experts further includes a third phenotypic expert model that is used to process outputs of the first and second phenotypic expert models.
18. The system of claim 17 , wherein the third phenotypic expert model includes a gating network.
19. A non-transitory computer-readable medium comprising instructions that, in response to execution of the instructions by a processor, cause the processor to:
obtain one or more natural language textual snippets, each natural language textual snippet describing one or more environmental or managerial features of an agricultural field that exist during a crop cycle;
encode, using a sequence encoder machine learning model, the one or more natural language snippets into one or more semantic embeddings in embedding space, wherein the one or more semantic embeddings semantically represent the one or more environmental or managerial features of the agricultural field;
quantify anamount of ambiguity of the encoding of the one or more natural language snippets into the one or more embeddings;
generate, using one or more phenotypic machine learning models, one or more phenotypic predictions about the agricultural field based on the one or more semantic embeddings, the amount of ambiguity, and additional structured data about the agricultural field, the one or more phenotypic machine learning models to assign a greater weight to a first semantic embedding generated from an unambiguous natural language snippet than to a second semantic embedding generated from an ambiguous natural language snippet, wherein the structured data includes sensor data gathered by one or more sensors carried through the agricultural field by one or more agricultural vehicles, and the one or more of the phenotypic predictions are generated using the sensor data; and
cause output to be provided at one or more computing devices, wherein the output is based on one or more of the phenotypic predictions.
20. The method of claim 1 , wherein the quantification of the amount of ambiguity is based on at least one of a presence, absence, or scope of temporal identifiers included in the natural language snippet.