Your team has been tasked with creating an ML solution in Google Cloud to classify support requests for one of your platforms. You analyzed the…

This approach correctly balances all project requirements. Using an established, pre-trained text model (e.g., BERT, Universal Sentence Encoder from TensorFlow Hub) for transfer learning significantly reduces development time and leverages a powerful foundation. Fine-tuning this model on the specific support request data allows for high accuracy on the custom classification task. Crucially, this method grants the team the "full control of the model's code, serving, and deployment" using TensorFlow, which can then be orchestrated within their specified Kubeflow pipelines. This aligns with using managed services (AI Platform/Vertex AI for training/serving) while retaining custom control.

A. Use the Natural Language API to classify support requests This is a pre-built API that does not allow for custom model code or fine-tuning, violating the requirement for "full control of the model's code."
B. Use AutoML Natural Language to build the support requests classifier AutoML automates model creation and abstracts away the underlying code, which contradicts the explicit requirement to use TensorFlow for "full control."
D. Use an established text classification model on Al Platform as-is to classify support requests A generic, pre-trained model used "as-is" will not be trained on the specific support request categories and will therefore perform poorly without transfer learning (fine-tuning). ---

1. Google Cloud AI Documentation: The tutorial "Classify text with BERT on Vertex AI" demonstrates the exact process described in the correct answer. It states, "In this tutorial, you learn how to train a text classifier by using a pre-trained BERT model from TensorFlow Hub and deploying it on Vertex AI." This directly supports using an established model (BERT) and adapting it (transfer learning) on a managed Google Cloud service.

Source: Google Cloud. "Get started with text classification with TensorFlow Hub and Vertex AI". Vertex AI Documentation.

2. TensorFlow Official Documentation: The core concept of transfer learning for text is a standard practice. The official TensorFlow tutorial, "Text classification with TensorFlow Hub," explains: "This tutorial demonstrates... how to use a pre-trained model from TensorFlow Hub for text classification... This is one of the most common and useful ways to apply transfer learning." This validates the methodology.

Source: TensorFlow. "Text classification with TensorFlow Hub: Movie reviews". TensorFlow Core Tutorials.

3. University Courseware: Stanford University's course CS231n explains the benefits of transfer learning: "In practice, it is very rare for anyone to train a whole Convolutional Network from scratch... instead, it is common to pretrain a ConvNet on a very large dataset... and then use the ConvNet either as an initialization or a fixed feature extractor for the task of interest." While the example is for images, the principle is identical for text models like BERT.

Source: Karpathy, A. & Fei-Fei, L. (2022). "Transfer Learning". CS231n: Convolutional Neural Networks for Visual Recognition, Stanford University, Module 2, Lecture 6.

4. Peer-Reviewed Academic Publication: The seminal paper on BERT, a foundational model for this type of task, explicitly details its design for fine-tuning on downstream tasks like classification. The authors state, "For each task, we simply plug in the task-specific inputs and outputs into BERT and fine-tune all the parameters end-to-end."

Source: Devlin, J., Chang, M. W., Lee, K., & Toutanova, K. (2019). "BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding". Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pp. 4171–4186. Section 4: Fine-tuning BERT. DOI: https://doi.org/10.18653/v1/N19-1423