Deep Learning Requires Large Data Sets and Significant Computational Resources

PMBOK v8 Definition

Deep learning (DL) is a more advanced machine learning type that relies on multilayered neural networks to extract features and make decisions. Deep learning often requires large data sets and significant computational resources. DL is a subfield of machine learning, which itself is a subfield of artificial intelligence, and generative AI (GenAI) is a subset of deep learning that applies large language models (LLMs) to create systems capable of generating new data such as text, speech, audio, pictures, and videos.

Why It Matters for the Exam

This concept appears in PMI exam questions about emerging technologies, resource allocation, and project constraints. You will encounter it in questions that test your understanding of how AI adoption strategies—specifically automation, classification by complexity, and human supervision needs—affect project resource planning. The exam tests whether you recognize that DL projects demand substantially more data and computing power than simpler ML implementations.

Key Points to Remember (for the exam)

• Core Hierarchy: AI → Machine Learning → Deep Learning → Generative AI (GenAI). This sequence is tested repeatedly; know which is the broadest and which is the most specialized.

• Resource Implication: DL projects require both large data sets AND significant computational resources. The exam tests this as a constraint for project budgeting, scheduling, and procurement.

• Task Complexity Classification: AI adoption strategies classify tasks into three categories: Automation (low complexity, little human intervention), tasks requiring more human supervision, and tasks needing full human oversight. DL typically falls into higher-complexity categories.

• Bias Risk: DL systems can be subject to bias if trained with biased data or if algorithms introduce bias. Mitigation strategies tested on the exam include: diversification of data sets, periodic bias-focused tests, and involvement of different teams in development.

• Privacy Requirements: DL systems use large data sets that can be sensitive and regulated by privacy policies and laws. This increases the need to properly secure data and ensure collection with a privacy policy in place.

• Human Supervision: The more complex the task, the more human intervention is required to result in high-quality outcomes. DL projects generally require significant human oversight.

• Common Confusion: Do not confuse DL with general ML. DL is a more advanced type of ML that specifically uses multilayered neural networks—this distinction is frequently tested.

Typical PMI Exam Example

A project manager is planning a new initiative to implement a system that can generate realistic product images from text descriptions. The team estimates the project will require 50 terabytes of training data and access to cloud-based GPU clusters for 6 months. According to PMBOK v8, which type of AI technology is most likely being implemented?

Answer: Generative AI (GenAI), which is a subset of deep learning, because it applies LLMs to generate new data such as pictures.

PMI Exam Traps

• Trap: Confusing "deep learning" with "machine learning" as interchangeable terms.

  • Reality: DL is a more advanced ML type that relies on multilayered neural networks. ML is the broader subfield of AI.

• Trap: Assuming all AI projects require similar data and computational resources.

  • Reality: DL specifically requires large data sets and significant computational resources. Simpler ML or automation tasks may not.

• Trap: Thinking bias mitigation is optional or only technical.

  • Reality: PMBOK v8 mandates three specific mitigation actions: data set diversification, periodic bias tests, and involvement of different teams.

• Trap: Believing privacy compliance is only about data security.

  • Reality: Privacy also requires proper data collection policies and ensuring data is not used in ways that violate ethical standards.

Important PMI Connections

Quick Review Questions

1. What is the correct hierarchy from broadest to most specific: AI, ML, DL, GenAI?

2. Name three specific actions to mitigate bias risk in AI systems as described in PMBOK v8.

3. Why does deep learning require more human intervention than simpler automation tasks?

4. What privacy-related requirements must be in place when using large data sets for DL systems?

5. Which type of AI applies large language models to generate new data such as text, speech, audio, pictures, and videos?

PMBOK v8 Reference

Section X3 - Artificial Intelligence (Appendix X3)
Figure X3-1: The Relationship Among the Technologies Needed to Make Generative Pretrained Transformers (GPTs) Work
Section X3.1.1 - Strategies for AI Adoption