Task-specific solutions overview

This document describes the artificial intelligence (AI) features thatBigQuery ML supports. These features let you develop task-specificsolutions in BigQuery ML by using Cloud AI APIs. Supported tasksinclude the following:

• Natural language processing
• Machine translation
• Audio transcription
• Document processing
• Computer vision

You access a Cloud AI API to perform one of these functions by creating aremote modelin BigQuery ML that represents the API endpoint. Once you havecreated a remote model over the AI resource that you want to use, you accessthat resource's capabilities by running a BigQuery ML functionagainst the remote model.

This approach lets you use the capabilities of the underlying APIwithout having to know Python or develop familiarity with API.

Workflow

You can useremote models over Vertex AI modelsandremote models over Cloud AI servicestogether with BigQuery ML functions in order to accomplishcomplex data analysis and generative AI tasks.

The following diagram shows some typical workflows where you might use thesecapabilities together:

Natural language processing

You can use natural language processing to perform tasks such as classificationand sentiment analysis on your data. For example, you could analyze productfeedback to estimate whether customers like a particular product.

To perform natural language tasks, you can create a reference to theCloud Natural Language API by creating a remote model and specifyingCLOUD_AI_NATURAL_LANGUAGE_V1 for theREMOTE_SERVICE_TYPE value. You can then use theML.UNDERSTAND_TEXT functionto interact with that service.ML.UNDERSTAND_TEXT works with data instandard tables. All inferenceoccurs in Vertex AI. The results are stored inBigQuery.

To learn more, tryunderstanding text with theML.UNDERSTAND_TEXT function.

Machine translation

You can use machine translation to translate text data into other languages.For example, translating customer feedback from an unfamiliar language intoa familiar one.

To perform machine translation tasks, you can create a reference to theCloud Translation API by creating a remote model and specifyingCLOUD_AI_TRANSLATE_V3 for theREMOTE_SERVICE_TYPE value. You can then use theML.TRANSLATE functionto interact with that service.ML.TRANSLATE works with data instandard tables. All inferenceoccurs in Vertex AI. The results are stored inBigQuery.

To learn more, trytranslating text with theML.TRANSLATE function.

Audio transcription

You can use audio transcription to transcribe audio files into written text.For example, transcribing a voicemail recording into a text message.

To perform audio transcription tasks, you can create a reference to theSpeech-to-Text API by creating a remote model and specifyingCLOUD_AI_SPEECH_TO_TEXT_V2 for theREMOTE_SERVICE_TYPE value. You canoptionallyspecify a recognizer to use to process the audiocontent. You can then use theML.TRANSCRIBE functionto transcribe audio files.ML.TRANSCRIBE works with audio files inobject tables. All inferenceoccurs in Vertex AI. The results are stored inBigQuery.

To learn more, trytranscribing audio files with theML.TRANSCRIBE function.

Document processing

You can use document processing to extract insights from unstructured documents.For example, extracting relevant information from invoice files so it canbe input into accounting software.

To perform document processing tasks, you can create a reference to theDocument AI API by creating a remote model,specifyingCLOUD_AI_DOCUMENT_V1 for theREMOTE_SERVICE_TYPE value, andspecifying a processor to use to process thedocument content. You can then use theML.PROCESS_DOCUMENT functionto process documents.ML.PROCESS_DOCUMENT works on documents inobject tables. All inferenceoccurs in Vertex AI. The results are stored inBigQuery.

To learn more, tryprocessing documents with theML.PROCESS_DOCUMENT function.

Computer vision

You can use computer vision to perform image analysis tasks. For example, youcould analyze images to detect whether they contain faces, or to generatelabels describing the objects in the image.

To perform computer vision tasks, you can create a reference to theCloud Vision API by creating a remote model andspecifyingCLOUD_AI_VISION_V1 for theREMOTE_SERVICE_TYPE value. You can then use theML.ANNOTATE_IMAGE functionto annotate images by using that service.ML.ANNOTATE_IMAGE works with data inobject tables. All inferenceoccurs in Vertex AI. The results are stored inBigQuery.

To learn more, tryannotating object table images with theML.ANNOTATE_IMAGE function.

What's next

• For more information about performing inference over machine learningmodels, seeModel inference overview.
• For more information about supported SQL statements and functions forgenerative AI models, seeEnd-to-end user journeys for generative AI models.